AI Education

Moving!!

2010-05-06T19:32:00.000-04:00

I'm currently in the process of combining my blogs into a single blog and moving to my own domain name. This site will exist for a while, but further updates can be found at www.mattsjourney.com. All posts dealing with artificial intelligence can be found under the label 'Artificial Intelligence'.

Axiomatic System Followup

2010-05-05T21:08:00.000-04:00

I spent some more time thinking about complex adaptive systems being, at their base level, axiomatic systems. As axiomatic systems are defined, this cannot be true. There is no internal application within axiomatic systems. Even though mathematics is n axiomatic system, all applications come from outside the system. Another issue is that axiomatic systems are not adaptive at all. Kinda hurts that I missed out on that big clue.

So a new question: can we enhance axiomatic systems (change their definition) to make them adaptive and make them externally applicable, yet keep their fundamental concepts intact? Can we take them to a new level for both theory and application?

Bottom-Up AI

2010-05-04T13:35:00.000-04:00

A recent commenter (yes, people do comment here, on occasion) said that building AI from the bottom up is the only way to go. I spent some time thinking about that, and I have to agree. Then I started expanding my thinking to include all artificial complex adaptive systems (ACAS). If I wanted to build an ACAS, I would HAVE to start from the bottom, then test from the top down.

This sounds like an axiomic system. Create a set of axioms (building blocks) and see if you can reach a set of higher-level principles with those axioms. If not, you might have bad axioms, bad operators to use on those axioms, or just a bad goal in general. Even reaching the principle might not prove anything, for the same reasons.

The downside (the only one I can see so far) is Godel's Incompleteness Theorem. The mind is self-referential. Godel shows use that axiomic systems run into problems when they become self-referential.

Overall, this sounds like a promising avenue to explore. Can any complex adaptive system be translated into an axiomic system? If so, how does Godel come into play? Is it even relevant?

AI-Completeness Hypothesis - Second Corollary

2010-04-30T16:31:00.000-04:00

Hypothesis:

Any true artificial intelligence technology is AI-complete.

First Corollary:

Any approximation of a true AI technology is not necessarily AI-complete as it is not true AI.

Second Corollary:

The total set of AI-complete technologies will be the necessary and sufficient components of a complete AGI.

Open Versus Closed Mind

2010-04-30T10:14:00.000-04:00

Let's look at two extremes. Mind A is a completely open mind; it weighs all input as equal no matter what logic, experience, or anything else says, leading to extreme cognitive dissonance. Mind B is a completely closed mind; unless new input conforms to what the mind already knows, it will ignore it, leading to no cognitive dissonance.

I have never heard of anyone at either extreme, but I have met people very close. Let's create a range of -1 to 1, where a score of -1 relates to a completely closed mind and a score of 1 relates to a completely open mind. We'll call this the mind's openness score. Those in the lower 15% or so we'll call dogmatics; they are set in their ways and nothing will ever change them. Those in the higher 15% or so we'll call sheeple; they will change their mind every time someone gives them new information.

Question 1: Is the openness score fixed for a given mind?
Question 2: If the score is fixed, which value is optimal for a mind? What is the balance needed to allow a mind to stick to their values, beliefs, and goals but open enough to change them given the right input?
Question 3: If the score is dynamic, what causes a change? Is it age? Is it the input itself? What type of input causes larger or smaller changes?

This strikes me a bit like a bias in an artificial neural network.

A Balance to Complexity

2010-04-29T21:14:00.000-04:00

I was reading an email from Ben Goertzel, a major proponent and researcher in artificial general intelligence (AGI), and something that he said stuck out to me.

"The central problem [of AI/AGI] is coordinating various components together in a synergetic way, so that they can help each other avoid combinatorial explosion, rather than exacerbating each others' problems via 'error magnification'"

The issue with many AI projects, such as creating a common sense engine, is the fact that you need a massive number of interacting objects (such as rules) to make anything realistic. Earlier today, I saw a TED Talk about simplicity, and how few scholars are looking into it, while much was known about complexity.

When putting those two together, I started thinking about balancing the rules of complexity with rules of simplicity. There needs to be a balance to allow for a massive number of interactions while not creating a combinatorial explosion.

AI Completeness Hypothesis

2010-04-29T16:26:00.000-04:00

I've read up a bit on AI-completeness and want to put forth a hypothesis:

Any true artificial intelligence technology is AI-complete.

This means that any truly intelligent/conscious feature of an AI will only be developed when a true AI is developed. It also means that chess (the way current programs 'play' it) is not true AI. I also wonder how this would affect the Turing Test.

A corollary would be:

Any approximation of a true AI technology is not necessarily AI-complete as it is not true AI.

An example is, again, how chess is currently handled. A computer can be 'great' at chess simply because it can test possible plays faster than a human can. This is an approximation of playing chess intelligently.

How would we go about proving this? The first step is to create formal definitions of AI-complete, true AI, artificial intelligence technology, and an approximation of a true AI.

Mary's Room and Subjective versus Objective Knowledge

2010-04-26T20:43:00.000-04:00

I'm reading up a bit on consciousness and subjectivity, as this seems to be one of the biggest stumbling blocks for AI. A 'quick' browse of Wikipedia led me to the Mary's Room thought experiment (http://en.wikipedia.org/wiki/Mary%27s_room).

First, is it possible to grow up in a monochrome environment? Couldn't Mary look at her own body for examples of color? Second, would her nervous system still be able to perceive color after so long without it? If a person lives in a cave for a long time, they will emerge blind. Third, even if there is a difference, wouldn't that simply mean that there are at least tow different types of knowledge: objective and subjective? Objective knowledge would be that which does not require personal experience. Subjective knowledge would require the perception of a mind. Fourth, would we experience the same thing as Mary? If she saw a red apple, and we saw a read apple, are we seeing the same red? Maybe I'm seeing what Mary would call yellow. But as long as we are internally consistent and call the colors by the same name, we would never know. Subjectively, we could never truly describe what color we were actually seeing to the other.

The first and second questions lead me to wonder if this is even a valid thought experiment. The third breaks knowledge into qualia and not-qualia. The fourth leads us to physicalism and tells us that the qualia are nothing more that how our brains interpret a scene.

Symbol Grounding For AI

2010-04-09T20:39:00.000-04:00

I've been doing a bit of reading lately about the symbol grounding problem. Personally, I feel that this is possibly the greatest issue facing AI. Decades of research has not come close to solving it. I'm even starting to wonder if it could be a non-issue. After all, how do humans ground symbols? Neurons are simply on and off, like computers. Granted, there are neuron spikes; the timing can count. But how does that make it different from a computer at the lowest level?

AI Dichotomies

2010-04-07T13:53:00.000-04:00

In reading an intereting blog entry (http://monicasmind.com/?p=53), I came across a list of dichotomies that AI researchers need to look into before really working with AI. As I don't fully understand those listed, I want to take the time to know exactly where I stand with them.

"The most important dichotomies are the Reductionist / Holist split, the Symbolic / Subsymbolic split, the Essentialist / Nominalist split, the Instructionist / Selectionist split, The Infallible / Fallible split, and the Logic / Intuition split (which could also be called the Reasoning / Understanding split)."

Reductionist / Holist
I feel I may say this often, but I see the need for both. The human mind/brain is a complex system. I think that, while looking at each individual part gives us more information, we will not truly understand it until we have a concrete holistic view. I think this also applies to AI. We can learn much from neural networks, genetic algorithms, computer vision knowlege representation, and so on, but will not truly understand AI until we have a holistic understanding. Each part is missing data from every other part.

Symbolic / Subsymbolic
This is a much more difficult one, and one that I cam still looking into. Symbolic AI follows the physical symbol system hypothesis (http://en.wikipedia.org/wiki/Physical_symbol_system). The greatest criticism comes from symbol grounding, or giving meaning to the symbols. Subsymbolic came later, after symbolic AI stalled. Subsymbolic AI includes embodied AI and neural nets. There isn't a really good definition (that I've seen).

I honestly don't know if I agree that symbols are both neccessary and sufficient for AI. Granted, there has been a lot of work done, successfully, but it was also fairly limited. On the other hand, no approach has been found for a fully successful AI. I want to study this one more, maybe take a step back and see if there is a prior assumption that has forced us on this path.

Essentialist / Nominalist
Are there essential properties of a specific object that places it within a category? (http://en.wikipedia.org/wiki/Essentialism) For instance, are there certain properties that are required to make a specific tiger officially part of the 'tiger' classification? For our purposes, are there certain properties of a mind that make it part of the 'mind' classification? Given my own mind as an example, what essential properties exist that MUST be incorporated into an artificial mind to make it a 'mind'? This has been a bit of a moving target for AI; as soon as one property is replicated, another property is suddenly 'essential'.

On the other hand, a nominalist would say that there are no universals (categories) that exist. (http://en.wikipedia.org/wiki/Nominalism) Given all cats, are there any essential properties that differ from those of a different but similar category? In other words, what makes a cat not a dog? There are many properties that belong to both. We humans can typically tell the different between the two. Some nominalists would say that a mind is required for this distinction. Or, at least, that the distinction is created solely by something like a mind.

I don't know where I stand on this one either. It requires more research and thought.

Instructionist / Selectionist
I cannot find any reference to these terms in Wikipedia, and a Google search gives me links to sites from multiple fields. I'll need to look into this one a bit more.

Infallible / Fallible
Given how common these words are, I'll just use the standard definition to talk about them. An infallible AI would be one that has to have 100% correct information and make 100% correct decisions. I would lump and (non-fuzzy) logical approach in this group. A fallible AI would have a degree of fuzziness to it, giving a 'likelyhood' measurement for its decisions.

We currently have a single example for human-level intellect, and it's not dolphins. One thing I can easily say is that humans are easily fallible. To give something artificial intelligence, why would have force ourselves to make it infallible? It would create difficulties that might not even be solvable, and would take far too much computational power. I stand firmly in the fallible camp.

Logic / Intuition (or Reasoning / Understanding)
This one is a bit easier to define and discuss. It seems to me that logic/reasoning is a learned skill. A human, even an intelligent one, can survive without being very logical. After all, politicians do it. Jokes aside, there are plenty of human beings who couldn't follow the simplest of logic. Yet, every human being, barring mental issues, is able to act intuitively, based on their experiences with the world. Humans raised by animals (there are accounts) have instincts/intuition. They can understand concepts, even if they cannot articulate them.

However, intuition/understanding is a very difficult concept to program into a computer, unlike logic. I think that the early AI researchers were looking for an idealized version of intelligence. Since the more 'intelligent' humans tend to use logic and act logically, logic must be the end all be all of intelligence. A good website is (http://artificial-intuition.com/index.html). They claim that intuition is easy to program, but give few hints as to how.

Conclusion
And there you have it. The above shows that I know some of what I believe when it comes to AI, but still need to some more research before commiting to more of the details.

Representation Versus Interpretation

2010-04-07T13:27:00.000-04:00

Is it possible to seperate the interpretation of knowledge from its representation? With humans, however our knowledge is represented, we can interpret it in many different ways (different perspectives). If we can separate these two, we can use a common knowledge base while being able to interpret it using the method appropriate to its use.

PhD Focii - As of 4/1

2010-04-01T10:59:00.000-04:00

~~1. AI in Education (Intelligent Tutoring System)~~
2. Constructed Languages (Lojban) in AI
~~3. AI in Interactive Narration~~
4. Knowledge Acquisition Through Machine Perception
5. The Affects of Priming. Prior Knowledge, and Concurrent Sensory Input on Natural Language Processing
6. Applying Foundherentist Principles to Knowledge Base Construction
7. Determine how Levels of Abstraction Affect a Knowledge Base

I'm questioning number one. The current research into an ITS is mostly determining what questions to quiz them on (I believe). I'll read up on it some more, but my latest additions to the list seem to be pointing in a different direction. For this reason, number three is also called into question. Research into the other topics would benefit one and three.

My remaining choices, except for number two, deal with knowledge bases and/or knowledge representation in some way. I've made no attempt to hide the fact that I think the existing methods of knowledge representation (that I know about) are far too limited and not very recent. With our expended understanding of neuroscience, epistemology, and computer science, we would have come up with a better method by now. Granted, I need to read up on what's being used, but I keep seeing frames and first order logic being mentioned. While we have much to learn from each, both have faults that have yet to be resolved.

Yet Another Direction

2010-04-01T08:53:00.000-04:00

Topic: Applying Foundherentism Principles in Knowledge Base Design

Foundherentism: http://en.wikipedia.org/wiki/Foundherentism

I had thought up foundherentism while sitting in my epistemology class, but never really put too much thought into it until I heard the term. I have the book that explains it in detail, but have yet to read it. Given that I have an interest in both knowledge representation and epistemology, this might be up my alley. I would still have to create/use an existing knowledge representation scheme for the KB itself, but at least I would have a philosophical basis for the overall architecture.

For now, I'll add this one to my list.

The Use of Lojban With Knowledge Acquisition and Representation

2010-03-31T23:19:00.000-04:00

After some thought, I decided that I didn't like the idea of using Lojban for knowledge representation. After all, it's based on first-order logic, and would therefore have all the same benefits and drawbacks. However, as it is easier to parse than a full natural language (ignoring subsets of, say, English), it would be useful for a 'natural' language interface with the user. The Lojban text would simply have to be translated into something else, something that does a more complete job of representing and processing knowledge.

Then there are the other senses. One of my biggest complaint of Searle's Chinese Room is that his counter argument to the use of vision, audio, etc. required that the vision only be of symbols (the same as the original input) instead of a picture. (For the record, my other, and probably larger, complaint is that the brain processes real world data with on-and-off-again neurons, making it as symbolic as his Room. Yet we can 'understand' the world.) If a program can process Lojban (through speech or text) as well as 'look' at data through an advanced vision system (advanced in that it is able to handle the same/similar basic preprocessing as the human vision system), and link them together in its knowledge representation system, then why can't understanding follow? Granted, there are likely steps and processes missing from that hypothetical system, but those can be worked out in time.

For a project like the one I'm describing, I would need an interface (text and/or audio) that processes Lojban, a complex computer vision system, a 'complete' knowledge representation system (complete in that it can handle any kind of knowledge, at any level of abstraction), a reasoning/problem solving/creativity system to process the knowledge, and an output to respond to the user (audio/visual/text). Stir until it thickens, serve while warm.

Sounds like a life's work to create all of these system, make them work together, and teach it.

Another Possible (And Interesting) Area of Study

2010-03-31T16:38:00.000-04:00

Much natural language processing relies only on the current language (text) being used. People use their other senses (typically) while interacting linguistically with others. Then the context of the text is influenced by the subject's knowledge base as well; if there are two interpretations to a text, but the subject's knowledge base can only help the subject understand one of them, then that one os what the subject 'understands'.

Topic: The affects of prior knowledge and other sensory data on the interpretation of a language (text).

Context Sensitive Knowledge Hypothesis

2010-03-31T15:13:00.001-04:00

The more concrete the knowledge, the greater the context sensitivity. The more abstract the knowledge, the less the context sensitivity.

Thoughts?

I've been thinking about knowledge representation lately, as it features in much of my list of possible research topics. I've also thought about context sensitivity off and on for a while. The idea for the hypothesis came from the fact that knowledge representations that are more 'complete' are the more abstract. I say more complete in that they can 'reason' more easily, but don't represent the world as well. The more detailed the knowledge representation, the better it represents the world (and the larger it is), but it's not able to reason as well or as quickly.

It makes me wonder how the brain is able to hold so much, yet we are (usually) able to come to a conclusion fairly quickly. Do we start from an abstracted version of the problem and then work our way through the relavant details(context!) until we reach a useful conclusion?

Now we end up in the domain of problem solving. This is what I prefer the term 'knowledge representation and problem solving' to 'knowledge representation and reasoning'.

I'm going to read up on knowledge representation more, and context sensitivity, and see where it leads me.

PhD Focii - As of 3/30

2010-03-30T14:04:00.005-04:00

Yesterday I mentioned the possibility of looking at the use of AI in education. Later yesterday and today, I thought about another possible focus: the use of contructed languages (mostly Lojban) in AI and human-computer interaction. Continuing to think up more...

For now, I'm just to create a list. Once my list gets a bit long, I'll think about what will go into each focus to narrow it down.

1. AI in Education (Intelligent Tutoring System)
2. Constructed Languages (Lojban) in AI
3. AI in Interactive Narration
4. Knowledge Acquisition Through Machine Perception
~~5. Automated knowledge Base Creation From an Online Source (Wikipedia)~~
6. Using Lojban for Both Knowledge Acquisition and Representation (See Goertzel)

Updated (11:23): Removed 5; English, or any natural language, is a pain to work with and too large for PhD work. While I could break it up for both during and after the PhD, it's still the least appealing topic listed.

Looking over the remaining topics, I can't help but break them up into the general areas of study that I would have to learn about. All five require study of AI. Two and six require mastery of Lojban, a good knowledge of predicate calculus, and a respectable understanding of linguistics. One requires knowledge of education, psychology, learning theory, and things of that nature; I have some background here. Three requires knowledge of storytelling, something I'm a big fan of. Four requires detailed knowledge of the perceptions I use (probably vision, a very large topic). I could conceivably combine four and six (if I was insane) and use Lojban to classify the knowledge that I gain.

Even after choosing the general area and reading up on the topics, I still have to narrow it down to something much more specific. Preferable, I'd find smaller bits of research that can be done first to give me a head start and something to publish before the dissertation.

Lots to think about. As always.

PhD Focus

2010-03-29T10:34:00.000-04:00

I need to decide what to focus on while earning my PhD. In other words, what research will I pursue? This is very important, as an academic's future is typically decided by their dissertation topic. As I am using this blog to discuss this, I'm sure my astute readership will have picked up that AI is my main area of focus. Unfortunately, AI is far too broad a topic. It's a kingdom when I need to focus on a fiefdom.

There are a great number of subtopics to choose from, when looking at pure AI. But only a normal person would focus on something from a single area of interest. Those that follow my other blogs will notice that I have other interests as well.

So will I combine AI with philosophy? Lojban? Japanese? I suppose I could combine it with Flow, but that's something that I just want to work on in my spare time. No, my other big interest is education. Broadly, how can AI improve human education? Educational research has shown that one on one is the best way to teach someone. And more and more schools are using laptops, netbooks, and/or tablets in the classroom.

This also ties in with my wanting to start a business that sells AI mentors/secretaries/assistants.

So I should be on the lookout for a university that does research in both AI and cutting edge educational technology.

Shooting for the Stars

2010-03-16T11:52:00.000-04:00

I decided that I want to work toward a PhD in AI. I know it won't be easy, but I think that it is what I want.

I have a long road ahead, and that's before I even apply to a single PhD program. I need to learn to do research. This is my biggest roadblock. It was my weakness back in college, even while earning my Master's. I also need to be published in some way. That would go a long way toward getting me accepted. Then I will need to research PhD programs. And I want to get my current student loans out of the way before taking on more.

Yes, a long road indeed.

ELIZA

2010-03-11T10:41:00.000-05:00

I've been thinking about what to do with my Lojban parser and finally decided on a nontrivial application. Most, if not all, followers of this blog will recognize the application mentioned in this title. Yes, I'm going to make a Lojban version of Eliza, for a number of reasons.

First, it gets me back in to Java programming. I'm still using the ANTLR framework, and am not extending my Lojban parser to include an analyzer and generator as well. It's a bit more complex than the original ELIZA, but I want to be able to learn as much from this as possible.

Second, it's a useful app to others (as well as myself) who are trying to learn Lojban. I'll be able to have a quick chat with it, which will help me increase my Lojban language skills.

Third, and related to my first point, I'll gain experience with parsing and generating text.

Finally, I can extend it later to include speech recognition and generation. This will create an ELIZA you can talk to.

Should be fun.

Lojban Parser Update

2010-03-04T08:15:00.000-05:00

I picked up the ANTLR reference book yesterday to compliment my higher level book on writing parsers using the same system. Both were written by the same man, who also wrote the entire ANTLR system, so I'm hoping that they work well together. As I read through how grammars are organized, I'll start looking at the Lojban grammar itself as well. Small steps, but that's better than no steps.

Progress

2010-02-28T15:02:00.000-05:00

Many thanks to the first (and currently only) commenter of my recent post. I tend to get in the process of thinking too much and second guessing myself, which leads to a lot of inaction.

I now have a project that I will be moving forward on. I'm going to start developing a grammar for the Lojban language using the ANTLR parser in Java 5. Since Lojban is a parsable language, it would be useful for communication with an AI, which is kinda why it was created in the first place. Building a parser will help me learn the grammar myself.

There are a lot of benefits with this project. First, it gets me back into programming. Second, I get to learn a new language. Third, the results will be useful for any future work I do in AI.

Through the Mists of Time

2010-02-27T11:05:00.003-05:00

I know I haven't been much of a blogger within the past.... year. Or even before that. I'm not saying that I want to commit myself to this blog, or any blog, any time soon. Lately, I've just been replying to the same tired thread in one of my email accounts with notes. The good part is that I can access those notes anywhere. The bad side is that I'm the only one who can see them.

Well, here are my notes from the several days on AI:

If there's one topic I keep being drawn back into, it's AI. I may
leave it for a bit, but it continues to intrigue me. I think I should
just stick with it.

My biggest problem is that I want to work with all subfields of AI. I
need to narrow it down, especially now that I'm thinking of getting a
PhD. I need to do some research, and then write a paper or three. I
should also start a blog on my subfield. Maybe even a tutorial for
the more practical portions. A write paper? I need to show that I
can write the dissertation and do research.

What I want to do the most with AI is to create a digital personal
assistant. Unfortunately, this is very close to a true AI. It also
touches upon nearly all of the most difficult AI problems. This could
very well be a lifetime's work in a world of constantly changing
technology.

Do I want to spend the rest of my lifetime on this? I think so, but
I've been shying away from it in the past. Maybe it's time. I'll
have to cut back dramatially on the gaming, though I think that the
recent theft was enough for that.

This is going to be a LOT of work. Granted, that's kind of assumed
when I said it would take a lifetime. I think the most difficult part
will be deciding where to start. I need to work on my understanding
of the problem as well as my programming skills. I'm still thinking
C#, just to have all of those libraries.

It's really going to come down to the architecture I create, though.
And that will be based on some of the philosophical choices I make.

Looks like I need to read up on the practical philosophical concepts
that will affect me, software architecture, AI architecture, and basic
AI concepts.

What is my end goal? As of right now, I want software that can
converse intelligently through a multitude of interfaces, shows a
distinct personality that develops from its interaction with the world
and user, and learn over time.

Right now, I'm thinking of a server/client architecture, having the
program's knowledge base on a server, so it can continuously be
updated. All data learned about the user would remain on the client,
and would intentionally difficult to get from the server. The
program's personality would also be on the client, but would
occasional send the data back for R&D.

There would be an issue with quality of knowledge. There would have
to be assurances that all knowledge in the knowledge base would be
fact, not opinion. All entries would be verified by Quality Control
personnel, a random person chosen for each entry. This nearly
eliminates potential collaboration to insert biases.

Upgrading clients would be fairly easy. Personalities and local data
could be backed up locally before an upgrade.

We would need servers to hold the distributed knowledge base. And we
would need entry points in order to insert, verify, modify, and remove
knowledge. And we would have to monitor the consequences of adding
contradictory knowledge. OpenCyc might give us more information on
how to handle that.

Josh raises a good point. How do we represent the knowledge? There
are various methods, but each has a weak spot, usually a type of
knowledge that cannot be represented. One idea that I sent back to
Josh is that we allow the program to create its own encoding. We can
check it by asking for it back, as well as combined data. Knowledge
of all kinds can be represented by a natural language, but that's very
difficult to process. Also, visual and audio input would be a
different format. Very different formats.

What are the requirements for a versitile knowledge representation
encoding, of which natural language is one?

We should probably start with low level and work our way up. Where do
we get knowledge from? This is something I tend to disagree with
Descartes about. He said we cannot trust our senses. I say that we
have no knowledge without our senses. Everything we know can
originally be traced back to one or more sense. Or, rather, percept.
I do agree that much of our basic processing in genetically
programmed, likely through the overall structure of our brains.

So make processing innate, or built into the system. Unfortunately,
we won't know how to process knowledge until we know its structure.
The brain works because the structure evolved to what it is today,
handling very simplistic behaviors long before getting to us.

Would using Lojban help? It's a regular, formal language. Can we
easily translate between Lojban and English? What about a subset of
English that is regular? Do we use the non-regular portions often?

Where am I going with this? Do I want to follow symbolism for
representation? I will be if I use some mathematical or linguistic
notation.

http://en.wikipedia.org/wiki/Computational_theory_of_mind

Interesting. What about levels of abstraction for representation?
This would fit in with an agent architecture. Communicate within the
system with streams of thought, streams of representation. This fits
in with my flow version of computation. Would need a 'language' of
thought for it. Can I mix this with neural networks? How well would
it work with different inputs? Audio, visual, text, imagination?
Different parts of the brain would handle different bits of
processing. Could simply use distributed computing as well. The
platform listed in Links could be VERY useful here. I can use
different languages for different processing, and let them communicate
with thought streams.

There could be a focus mechanism that allows a primary thought stream
to have the attention. This would allow other streams to persis in
the 'unconscious'. The system would also have to have the ability to
expand it's abstraction system through 'learning'. But how do we
encode the thought streams? And thoughts in memory? Thought should
have the ability to affect how thoughts are affected, as a feedback
loop, meant to improve the system over time.

What can I use from real time system? There's a bit of a time code
involved, tell it what is out of date (and might be dropped) and what
needs to happen now (priority). Can add urgency to priority code (for
humans, a tiger leaping at you is more urgent that reading a book). A
priority code for each thought stream? Constant thought streams for
each sense, or better yet, for each portion of the senses (two ears,
two eyes, etc.).

Each output would have a stream, like nerve pathways to the muscles.
Low level. Would make a much more dynamic avatar. Would we then need
to teach it to talk, move? Maybe a base setup?

As of right now, my system is a distributed, flow-based system. It's
almost like a neural network with feedback loops. This is very
important, as traditional neural nets don't deal with feedback, just
back propagation. This has its uses, but is limited to only setting
weights, not changing flow or the operation of a neuron.

If I have a flow-based system, how will it be structured? I need to
go back to my old flow notes.

Ok, I can't find my old flow notes, yet. I need to figure out where I
recorded them. I think some might be on a Moleskine. I need to
check.

For now, I see individual components within a thought stream
framework. I still need to know how to structure the flow (Lojban
streams??) and how to have components interact with it.

For instance, a vision system would handle the input and some of its
own processing, but then pass the information back via one or more
thought streams. There would be thought streams back to control
attention toward specific objects (for more detailed processing) and
for moving the camera (like moving the eyes).

What about Lojban for flow structure? This would be useful for speech
input and output. It could then learn other languages the way it
learns anything else. But does Lojban lend itself to streams of
thought? I need to find out more.

Maybe not. While it's a logical, formal language, it would still be
difficult to process. Can we use a combination of a logical language,
symbolic representation, and mathematics to create a brain?

It's sounding more and more that LISP will be useful for a lot of
this. At least for the general purpose learning algorithms. Other
stuff will be more specialized, such as the vision system. For LISP,
a thought stream would be lists... Sort of. Can LISP use streaming?
Maybe use a wrapper to pull out what is needed for the LISP code.

Right now, I have a vision system where I want to use OpenCV. I have
the thought streaming where I'll use C#. I have the processing system
where I'm thinking LISP. I need a speech system. If I use Lojban,
then I'll need a simple parser. And a TTS and STT translator.
Storage? This is a bit more difficult. I don't even know how I'll
handle the thoughts within a stream, let alone how to distill them
into something persistent.

That's what I've been thinking about lately. I'll think about posting future notes here as well.

The Plan

2009-01-02T08:27:00.002-05:00

I talked a bit last time about what I was going to do. Now I want to expand that by talking some more about what I will do over the next year.

The first step is to consolidate all of my previous thoughts on AI into a central resource. I have several dozen pages from spiral notebooks to Moleskine notebooks to blog entries to go through. I ordered a scanner so I wouldn't have to type out the handwritten notes. I can just use OCR to speed up the process.

The second step is to organize the AI books I have and determine what areas I should work with. I'll likely gain direction from what's in my notes.

Step three is to take n0tes from these resources and add them to what I will have already transfered. I'll also try to keep this blog uptodate about what is going on.

In the end, one year from now, I should have enough ideas to be able to create an AI architecture to build. What form will it take? What specific abilities will it have? What will I name it? We'll find out.

The Return

2008-12-31T12:37:00.004-05:00

I'm back. I know it's not much of an explaination, but I had to put AI aside for a bit while focusing on work.

I'd like to thank the posters who recently posted a comment on the previous post, except for the tarot-loving spambot.

As for the future, I plan on spending 2009 just learning this and that about AI from the many books I already have on the subject as well as online resources. Then I'll spend 2010 planning an AI project as much as I can. From 2011 on, I'll work toward completing that project.

I don't know if I'll want collaborators or not yet. I need to see what the project is first.

Stay tuned for more...

Computer Science Reconsidered

2007-10-30T11:53:00.000-04:00

I mentioned in a previous post that I ordered a book entitled Computer Science Reconsidered. Well, I received a book from the right person, but it was the wrong book. Instead, I got the book Beating Depression. I wasn't depressed before, but this annoyance is pushing me closer. :-)

I'm Done...

2007-10-29T10:55:00.000-04:00

...with symbolic AI. We've had our spats in the past, along with a general disbelief in its possibilities. But there are just too many questions.

What gives the symbols meaning? How can they relate to each other? How can anything of any practicality occur when the symbols are not grounded? How can there be knowledge, understanding?

There needs to be some sort of context. Period.

For the Future

2007-10-28T21:01:00.000-04:00

There is a lot that can be said for the future of AI. Many will say that it will never happen. Some think it will, but not in their lifetime. A smaller amount think it will occur in their lifetime. The smallest group thinks their research will create it in the next 20 - 30 years. Funny how none of these groups can give a definite definition of 'artificial intelligence'.

But that's not quite what I'm here to talk about. While I'm in the second group, I'm more focused on how I can contribute to AI while using it to make a living. This would be a dream job for me. Well, maybe teaching AI classes at a university, but I have a difficult time with research.

So how do I want to 'do' AI as a career? Being a lowly code monkey, this is a key question. A skilled AI programmer (not that there's a concrete definition for that either) can make a lot of money. I'd like to work freelance, from home. This way, I could live anywhere in the country and simply teleconference to all of my meetings.

So how do I get this this point? Not by being your average programmer, seeking promotion after promotion. I agree that I need the training to see how the real world works for programmers, as well as generate some startup capital, but only for the next several years.

One area that I know I'm deficient in is mathematics. Math was one of my strong suits back in grade school, but not so much when I got to college. This needs to be fixed. I'm still trying to figure out the best method.

Another thing to consider is something mentioned in Computer Science Reconsidered by Karl M. Fant. The author claims that mathematics is not the best thing to base computer science on; biology is. I ordered this book and look forward to reading it in the next few weeks.

I'll keep you posted of any further developments.

Human Context

2007-10-21T15:51:00.000-04:00

Yes, I'm back. I'm back to thinking about AI, and how I can make that a focus as a freelance programmer. If my theories work out, that would make me stand out as one of the top AI specialists in the industry. If it doesn't work, then... Well, I'll just have to make it work.

Now, though, I would like to talk about context. I've spoken of this before, but I just not thought about how a human gets context. I've mentioned that context is perhaps the biggest hurdle for AI. This puts me at odds against the symbolic AI researchers who think that the symbols don't need outside meaning. Well, tough for them.

But how does a human understand the context? How do we give meaning to words, phrases, images? It has to come from previous experience. But what about in the beginning, when we don't have any previous experiences? Then it has to come from our genetic predisposition to perceive things the way we do. There is no other place for it to come from.

So what does this mean for AI? I'm not sure, yet. This requires a lot more thought. It also has many implications for psychology as well.

Simple Questions, Complex Answers

2007-02-23T23:28:00.000-05:00

With all this talk about patterns, we have some questions that must be answered.

What are the patterns of?
How do the patterns interact?
How are the patterns interpreted?
How is input turned into patterns?
How are patterns turned into output?
How is context involved?
How are patterns represented?

In regards to the first question, patterns can, of course, be made of other patterns. But what are the basic things that patterns are made of? Once we answer this, the other questions should be a bit easier.

My Final AI

2007-02-18T23:47:00.000-05:00

When I see the future of AI, I don't see a lot of robots running around. I don't see two AIs arguing over a bit of logic. What I see if one mostly visible source central AI (similar to bad sci-fi), and it handles requests from users. In other words, if would be very much distributed.

Here's what I now need to clarify: why visible source, why mostly visible source, why requests, and why distributed?

Well, I've read enough of both good and bad sci-fi to be a bit wary about what I don't understand. Once my future AI become active, it will quickly become something I no longer understand or control. This worry is making me take precautions before I write even one line of code.

Why visible source? I want people to be able to see how the AI works, in order to head off worries that either I or someone else will use the AI to take over the world. I don't have the evil laugh, so I'm already disqualified. Plus I wouldn't want all the hassle of running an entire planet, even if I do feel that I have some good ideas.

Why mostly visible source? I want some of the most important bits of code hidden so that it cannot be replicated without ethical guidelines (evil twin syndrome). These are the guidelines that I will personally teach my AI, like I will teach any children I raise.

Why requests? I want my AI not to have control over something that it should not be allowed to control. If you have seen War Games, then you know what I am talking about. If a user request aid with something, like directions from point A to point B, explaining a lesson in calculus, or getting help for a car accident, then the AI should help to the best of its ability.

Why distributed? If something happens, and the AI breaks its ethical code, it should not be allowed to serve any other requests until the dilemma is resolved by both the AI and a committee of humans. The distributed portions of the AI, the local computers communicating with the AI, would have all the information locally to finish assisting without further communication with the AI. This is just one more layer of protection.

So I have ideas. Now I just need the experience to bring them to light.

Addemdum to Layers

2007-02-18T23:45:00.000-05:00

I should talk a bit more about how a layer would be organized. I mentioned how it must not be confused with an expert system. I've also talked about how I see pattern theory being very important.

Well, I see any and every layer being patterns of behavior. The novice layer handles very basic learning patterns, pattern differentiation, and so on. The expert layers would handle more domain specific actions through more complex patterns of behavior.

AGI and Mind Stratification

2007-02-18T23:35:00.000-05:00

I was talking with other AI enthusiasts today, and came up with a few more ideas. I see AI being based upon a very general layer with everything that is needed (at the most basic level) to allow the being to become truly intelligent. The novice layer (the reason for this naming will become clear momentarily) will have all the patterns to deal with sense and motor response, as well as automatic skills (learned enough to be subconscious, like driving after several years of experience). a set of higher level layers will be created as the being learns about abstract thought that is built on top of the novice layer. These layers will be the expert layers. Please do not confuse these with expert systems.

These expert layers will be organized like a class diagram, with more abstract layers on top of each other. Once many layers are created, the being can then have them interact for interdisciplinary work. In other words, the biology expert layer and work together with the chemistry expert layer and create a biochemistry layer. More than two layers can combine as well, of course, as can previously combined layers.

This would give us a very adaptive being that can have specialized subsystems built upon an ever increasing general layer that will bring it all together.

Philosophy of Mind

2007-02-17T23:46:00.000-05:00

I've thought a lot about what the mind truly is. This is evidenced in many of my previous posts. I believe I recently mentioned patterns as central to the mind. Well, I found someone else who has taken this one step further.

In the book The Hidden Pattern: A Patternist Philosophy of Mind, Ben Goertzel creates pattern theory, which places patterns as a basic building block of the universe. Patterns create both the physical and mental alike.

I really feel that this is very important to AI. Because of this, I am slowly reading through the book to see what riches it holds. I'm not sure I've ever read a book this slowly and carefully, in fact.

The next step will be to immerse myself in patternist thinking by understanding it formally. For this, I need to relearn a lot of mathematics. I need to determine which fields of mathematics are most valuable and will lead me in the right direction. It's times like these that I wish I had majored in mathematics as well as computer science.

An Interesting Thought

2007-02-02T14:41:00.000-05:00

I subscribe to the Dilbert Blog, and today's entry is one I thought I should share.

The Fundamentals of AI

2007-01-31T14:03:00.000-05:00

First, I would like to apologize for my two month hiatus. I've been working on other personal projects. Those projects, however, have given me a few more ideas for my work with AI.

The other thing I want to talk about is what the fundamental knowledge requirements are for working with AI. I found such a list here. It's a wiki page about the requirements of becoming a Seed AI Programmer. While I'm not a big believer in most of their work, I'm attracted to the idea of seed AI and artificial general intelligence. The only thing keeping me from trying o join them is the fact that they seem to be full of themselves.

Anyway, here is their list:

Cognitive science
- Functional neuroanatomy
  - Functional neuroimaging studies
  - Neuropathology; studies of lesions and deficits
  - Tracing functional pathways for complete systems
- Computational neuroscience
  - Suggestions: Take a look at the cerebellum, and the visual cortex
- Computing in single neurons
- Cognitive psychology
  - Cognitive psychology of categories - Lakoff and Johnson
  - Cognitive psychology of reasoning - Tversky and Kahneman
- Sensory modalities
  - Human visual neurology. Big, complicated, very instructive; knock yourself out.
- Linguistics
- Note: Some computer scientists think "cognitive science" is about Aristotelian logic, programs written in Prolog, semantic networks, philosophy of "semantics", and so on. This is not useful except as a history of error. What we call "cognitive science" they call "brain science". I mention this in case you try to take a "cognitive science" course in college - be sure what you're getting into.
Evolutionary psychology
- Popular evolutionary psychology; dating and mating; Robert Wright and Matt Ridley
- Formal evolutionary psychology; neo-darwinian population genetics and complex adaptation; Tooby and Cosmides
- Game theory; nonzero-sum and zero-sum games
- Evolutionarily stable strategies for social organisms
- Tit-for-tat, the evolution of cooperation, the evolution of cognitive altruism
- Evolutionary psychology of human "significantly more general" intelligence
  - Mostly this means reading LOGI; there's not much else out there.
  - But see also Lawrence Barsalou and Terrence Deacon.
- Evolutionary biology
  - Incrementally adaptive pathways; levels of organization; etc.
- Biology (a complex system not designed by humans)
- Genetics
- Gene regulatory networks (another good look at evolution's bizarre signature, and also a look at the way humans actually get constructed)
- Quantitative genetics
- Anthropology - the good old days
Information theory
- Shannon communication theory
  - Shannon entropy
  - Shannon information content
  - Shannon mutual information
- Kolmogorov complexity
  - Solomonoff induction
- Bayesian statistics
  - Interpretation of human thought as Bayesian inference - see [Jaynes].
- Any other kind of statistics
- Utilitarian Bayesian decisionmaking
  - Decision theory is not classically part of "information theory", but does, in fact, belong together with the other items in this category
  - Child goals, parent goals, "supergoals" (intrinsic desirability)
  - Actions and desirability - read [Creating Friendly AI] as a preliminary to the longer story.
Computer programming
- Knowledge of many languages
- Java programming (that's probably what we'll end up doing it in)
- Being an excellent programmer
- Parallelism
  - Multithreading
  - Clustered and distributed computing - we may not need this for a while, but then again, we may
- Any kind of experience working with complicated dynamic data patterns controlled by compact mathematical algorithms - some of the interior of the AI may end up looking like this
Other stuff
- Computer security (experience with defensive caution; not that it's sufficient for Friendly AI or even a good attitude, but it's a start)
- Physics
- Thermodynamics
  - The second law of thermodynamics
    - Noncompressibility of phase space
    - The arrow of time and the development of complex structure
- Traditional AI methods
  - History of error; don't repeat past mistakes
  - We might reuse one or two design patterns at some point
- Transhumanism or transhumanist SF - sufficient exposure to have a very high future shock tolerance; helps to "take it all in stride"
- Mathematics

I'm seeing a few things that I don't feel are necessary at all, as well as a few things that are missing. I do think that the cognitive psychology and cognitive neuroscience are very useful since our brains do everything we want out of AI. I would have left out many of the items under evolutionary psychology.

The information theory list is interesting, and I will have to look into that more, probably after I work on expanding my knowledge of mathematics. The computer science list seem too short to me. I also like most of the stuff listed at the bottom.

I do find it interesting that there is no mention of learning, under any of the headings. Granted, an AI will likely not learn as a human does, but there are some good theories that would be useful for a grand theory of AI.

My list? I would have to think about it. I'm only one person with just a basic education. I'm not sure I know enough to make a comprehensive list. Though perhaps I'll make a list of things I should learn before I make a major attempt to implement my AI.

My AI Test

2006-12-02T09:44:00.000-05:00

I know that the Turing Test is still very popular as a test for intelligence. What I don't get is why. Why is this THE test for AI? I still feel that a sufficient weak AI system, or collection of systems, could pass this most sacred of tests. In fact, I will now fully reject the Turing Test as useful for my purposes, or anyone else's, for AI.

And now, here's my test: I'll know it when I see it.

I know that's not the most scientific thing to say. But I feel that it's the best way to go about it. When it comes to people, I can usually tell how intelligent they are simply by interacting with them over a period of time. Usually minutes. Why can't this method be used for an AI as well?

Definitions

2006-11-17T22:43:00.000-05:00

Here's the latest. There's a few minor updates again.

------------

Definition List for AI

Tentative Rules:
1.All definitions are tentative.
2.All definitions must apply equally to both artificial and natural intelligence.
3.All defined terms should work toward a working understanding of artificial intelligence.
4.All definitions should be unambiguous.
5.All definitions should be consistent with one another.

Definitions:
AI: an artificial entity that displays some level of intelligence
Intelligence: the ability to use, understand, and/or create technology.
Technology*: a mental or physical human construct
Learning: the creation or change of intelligence
Training: creating or changing the ability to use technology in another
Teaching: creating or changing the ability to understand technology in another
Understanding: the ability to react correctly to X withing the correct context is to understand X
Education: the combined knowledge of technology
Memory: the combined perceptions of the past
Perception: the input of one or more percepts and its interpretation
Percept: input from one of the senses
Sense: a type of input/percept of an entity
Communication: one or more percepts between one or more entities
Experience (singular): a collection of percepts for a specific event
Objective Environment: anything in existence apart from the entity
Perceptual Environment: how the entity perceives the objective environment
Imagination: a mental simulation
Imagine: create and perceive a mental simulation
Creativity: the ability to conceive of novel ideas inside or outside of the entity's mental simulation
Mind: the combination of the physical entity and its contents
Thinking: the ability to create and interpret thoughts
Thought: a mental construct that combines perceptual, linguistic, and imaginative structures from either memory or the perceived environment within a mental simulation
*Need: something required for the persistence of the entity
*Want: something not required for persistence, but perceived by the entity to make persistence easier or more pleasant
Belief: something that is assumed to be true but without evidence
Knowledge: something that is believed to be true and is true because of the reasons that the entity believes it
Goal: the preferred end state of a course of action
Focus: to concentrate the entity's attention on something
Attention: directing the thoughts and perceptions of the entity toward a specific internal or external event
Behavior: patterned actions or thinking created through experience
Emotion: unconscious mental state that gives rise to action
Feeling: change in mental state due to a change in emotion
Personality: sum of emotional and physical behaviors that persist over time
Self: combination of the personality and general physical state of the entity
Reflection: the ability for the entity to observe and think about its self
Self-Modification: the ability for the entity to change its own behavior
Consciousness: the emergent property of a mind to show reflection, self-modification, and self-sacrifice.
Meaning: the link an entity creates between the objective and perceptual environments through experience
Experience (plural): the accumulation of learned knowledge from memories

To Be Defined:
Wisdom

*Examples of Various Definitions:
mental technology: language, mathematics, logic, pattern matching/recognition
physical technology: sonar, radar, radio, digital watches, clothes, paper, soda, cups, soap
need: water, food, warmth
want: a computer, a good book, sex, conversation, money, a fast car

Problems:
The grounding problem: linking symbols and meaning
The learning problem: having the entity learn on its own reliably
Global-local disconnect problem: observing the global system does not tell you about the local parts of that same system
Software problem: programming languages are not set up do develop real AI
Hardware problem: hardware not set up like biology
Software problem 2: complex software is extremely difficult to develop

Data Streams

2006-10-22T11:27:00.000-04:00

I've been thinking more about incoming perception and attention. I've decided that each instance of perception, each percept, will have its own data stream. The entire sense 'organ' will be called a pipe. Thus, a pipe will be filled with various streams.

In order to slow the massive amounts of data that would otherwise crash the AI, the attention mechanism will determine which streams actually get processed at the conscious level.

This brings up the question of how 'levels of conscious' will work. I have yet to think that through.

There's also the question of how the attention will know which streams to cut off. Basically, each stream will be classified according various aspects of that sense. For instance, visual streams will have color, location, shape, and size while audio will have volume, location, pitch, and so on. The AI will be able to determine any other aspects on its own.

Here's what I'll be working on now: levels of consciousness, sense aspects, and motor control.

Cognitive Architecture: Memory

2006-10-01T15:18:00.000-04:00

Knowledge representation seems to be a tricky thing. There is a lot of research on it, but nothing quite useful enough (from what I've seen) for a real AI. Neural networks seem the most flexible, but are tricky and messy. And how do you represent visual data? Every other method also has its pros and cons. Nothing quite seems to work.

Moving to a higher level for the moment, let's see what kind of requirements we're looking at. We need to hold a wide variety of information: sounds, images, video, tactile, text, behaviors, and so on.

How do we handle abstract concepts? How can my imagination 'see' love? Maybe an example of love is all it takes to recognize it externally. How can it handle special cases? Killing is bad, unless... Stealing is bad, unless...

How do we handle forgetfulness? If we remember everything is minute detail, we risk running out of memory very quickly. This also ties into useless knowledge, such as what I had for breakfast April 12, 1996. For the record, I think I skipped breakfast that day. But I forget. It doesn't matter what I had for breakfast, or if a certain person was at a certain place last Tuesday.

How do we handle memory?

Naming My AI

2006-09-30T14:01:00.000-04:00

Every creation of AI needs a good name. I have one in mind, but it seems to be used a lot in many of the sci-fi stories I read.

My AI will be named Synthesized Artificial Mentality I, or SAM I. Once I have a voice pattern set up, I plan on giving it a more feminine persona. Again, I know that's done a lot, mostly in movies so the computer geek has a 'girl' to talk to. Tough.

For the record, my first cognitive architecture will be named SAM as well. The implementations of it will have the numbers.

Thoughts About An Overall Cognitive Architecture

2006-09-30T13:48:00.000-04:00

Thus far, my ramblings have gone almost nowhere. While they have given me a direction, they have not yet given me a goal. Let's fix that.

I want a general cognitive architecture that works. Yes, I think I'm talking about strong AI again. I'm hoping that it can be based on my theories, but I am very much willing to modify from what I want to what works.

Let's begin.

We'll start with what I would need and want in my cognitive architecture (CA). First, we need a way to dynamically add in inputs. I want this feature so that I won't have to stop and start my AI after a year of running simply because I want to add in another feature. And this way I can start with a few senses and work my way up.

I also want a way to dynamically add outputs for the same reasons. Once I have both of these, I need a way to integrate the inputs and coordinate the outputs. So I need my sense integrator (imagination) and an output control mechanism.

Then I want a way for my AI to be able to focus, which will dampen a lot of the input to make things much more manageable. Another very important feature is the memory. I need a way to separate inputs into functionally different parts, yet have a way to integrate them as quickly as possible.

Going back to the inputs, I want a way to do some basic filtering and preprocessing so the AI won't be overloaded.

Now let's look at what my AI still needs. Here's a list:

a goal setting mechanism

a way to plan

a way to use logic

some way to associate and link the memories

how will it learn?

how will it handle abstract ideas?

decision making

Notice the long list of very important stuff. I have a lot of work to do.

I'll post again later this weekend about various aspects of my CA, as well as what's still missing. And I'll likely work on my list of definitions as well to keep them updated.

Overall Cognitive Architecture

2006-09-29T00:25:00.000-04:00

Now that's I've thought a lot about what should go into my AI, I feel that it's time to develop an overall cognitive architecture. I've donw some work on it so far, and realized that a lot was missing.

I'll begin posting on it this weekend. Stay tuned.

GOFAI

2006-09-28T12:15:00.000-04:00

I've been a bit of a critic over the past few years of GOFAI research. Many AI researchers seem to be fixated on a certain aspect of AI, even though it has gone nowhere and has little potential for strong AI. One such avenue is the expert system. There was much hype over this, which led to much disappointment, as well as the so-called AI Winter.

Now I find myself looking back at my efforts and seeing an advocacy of strong AI research. I look at my theory of the imagination and see an 'imagination module' to be installed. I'm not quite sure if I like that or not.

My biggest concern is that I recognize that all of our cognitive faculties developed over a long period of time, and many at the same time. If we want to develop true AI in our lifetimes, we don't have that luxury.

I do understand that I need to know about these different aspects of AI to know when to use them as tools in my research. I may have a neural network-based memory system, or 'evolve' certain algorithms 'genetically'.

As long as I know what the limitations are for each of these techniques, I'll know when to use them and when to run the other way. I refuse to fall into the trap that others have.

More on the Imagination

2006-09-18T22:51:00.000-04:00

I've been fixated on the imagination for several months now, but I think I have good reason to be. After a meeting tonight with a philosophy group, I think my fixation is even more important.

I've talked about how the imagination is coupled with the integration of the senses, and how this is how we gain context, relive memories, and create something new. The other philosophers, some by training and some by interest, have awakened a few new ideas within me.

The imagination allows us to plan and give us fear. We can plan because we can 'see' what will happen if we take action A versus action B. We can visualize, conceptualize, cause and effect. We can see what dangers exist in our actions, thus increasing our fear response, making us more aware of our surroundings and thus safer.

Finally, and in a more controversial direction, we can consider the imagination to be a meme generator, differentiator, and integrator. That is, if you believe in the existence of memes. I'm agnostic about the idea, but it does fit in to what I've been talking about, so I am willing to give the idea a chance.

A Few Notes and Thoughts

2006-09-14T12:05:00.000-04:00

I've been giving a lot of thought to the mind lately. Here's some sniplets of what I've come up with:

Is meaning tied to the senses?
How can we use this to our advantage, if it's true?
Could we use some kind of automatic association neural network to create meaning?
How can I use my concept of the imagination for this?
How do we tie in language as well?
Should I look more into connectionism? Or maybe a new variation of it?
Do we need some kind of contextual computer?
How would we create one?
Could machine learning be applied to contextual learning?
What are the advantages/disadvantages of using the typical von Neumann architecture?
Is there anything beyond Turing machines?
Is this simply an artificial limitation?
How do humans learn new languages?
How can we integrate two visual streams into a single 3D visual stream?

Cognitive = Emergent?

2006-09-04T16:01:00.000-04:00

Could every cognitive function of the mind be an emergent property of the physical brain? If this is true, then any simulation of the physical brain (connectionism?), if it is to have any cognitive possibilities, would also have to be designed to display the same cognitive features, which may be impossible.

This leads to some harsh implications. One could possible hard code all cognitive functions into an artificial mind, which has so far failed. Or artificially evolve the same cognitive functions.

Mind as CAS - Basic Building Blocks

2006-08-23T08:48:00.000-04:00

As I've called the main a complex adaptive system (CAS), I feel the need to back up my decision. And I've decided to start at the lowest level: the basic building blocks.

I've been thinking about what these basic building blocks could be and came up with a list:

ideas (memes)
language
memories
percepts

Some of these seem easy to dismiss. Ideas are out since all higher order animals appear to have some level of intelligence, yet likely don't have symbolic processing capabilities. I grouped memes with this just to get it out of the way, since memes only contain cultural information. Language is out since babies demonstrate amazing mental abilities. Memories are crossed off for the same reason.

Percepts are a much more likely possibility. We experience our environment as soon as our brains are developed, it fits in with my ideas about the imagination, and it's a basic building block that all animals, intelligent or not, share.

But if we share this with all animals, what makes humans so different? Once again, I refer back to the imagination. Humans have the ability to create technology (see my definitions). We have language and tools, mathematics and wheels, logic and digital watches.

I could be missing something. Perhaps there is another choice, one that makes even more sense. Or maybe I'm falling prey to confirmation bias. I suppose I'll have to think on this further, just like everything else.

The All-Encompassing Imagination

2006-08-22T12:33:00.000-04:00

I know I talk about the imagination a lot. I suppose I see it as a major, if not the most important, part of the brain. I see how such a cognitive simulator can be used for many aspects of intelligence.

But this also goes against my seeing the mind as a complex adaptive system. A CAS is similar to a P2P system in that every component is equally important in its own way.

So now I need to find a way to rectify this discrepancy.

Mind = Complex Adaptive System?

2006-08-07T01:00:00.000-04:00

The question here is whether or not the mind should be considered a complex adaptive system. I think we can agree that the brain is one, but does adding the cognitive and behavioral aspects of the mind take it out of that category?

Context

2006-08-06T12:04:00.000-04:00

I've said it before, and I'll say it again: context is one of the keys to AI.

But how do we get context? Are we born with an ability to link our thoughts and meanings to the real world? Or is it something we learn?

Maybe we learn through experiencing some causality in our young lives? Or through repetition?

I'm not even sure how to go about answering this. At the moment, I can only hope that further education will shed some light on it.

An Interesting Site

2006-08-06T11:49:00.000-04:00

In the last day or two, I found an interesting site with a good discussion on the general aspects of artificial intelligence, and specifics on the use of an AI to predict trends in the stockmarket. It can be found here: http://www.ai-stockmarketforum.com/index.php

A Question

2006-08-06T11:38:00.000-04:00

If you were able to simulate intelligence, could it be considered intelligent in its own right? At which point does a simulation of something become the thing it's simulating?

Square One

2006-08-05T13:50:00.000-04:00

I'm hoping to have at least the architecture for my AI design created within the next two years. One of the main problems I'm having when I think about it, though, is how to structure the memory. All the types of memory that have been discussed have limitations on them. Is there any way I can combine these various memory schemes to make the conglomeration useful as well as easy to understand? I need to look into these memory schemes more.

Updated Definitions - Early August

2006-08-05T13:33:00.000-04:00

Here's the latest. There's just a few minor updates this time.

------------

Definition List for AI

Tentative Rules:
1.All definitions are tentative.
2.All definitions must apply equally to both artificial and natural intelligence.
3.All defined terms should work toward a working understanding of artificial intelligence.
4.All definitions should be unambiguous.
5.All definitions should be consistent with one another.

Definitions:
AI: an artificial entity that displays some level of intelligence
Intelligence: the ability to use, understand, and/or create technology.
Technology*: a mental or physical human construct
Learning: the creation or change of the ability to use, understand, and/or create technology
Training: creating or changing the ability to use technology in another
Teaching: creating or changing the ability to understand technology in another
Understanding: the ability to react correctly to X withing the correct context is to understand X
Education: the combined knowledge of the use, understanding, and/or creation of technology
Memory: the combined perceptions of the past
Perception: the input of one or more percepts and its interpretation
Percept: input from one of the senses
Sense: a type of input/percept of an entity
Communication: one or more percepts between one or more entities
Experience: a collection of percepts for a specific event
Objective Environment: anything physical apart from the entity
Perceptual Environment: how the entity perceives the objective environment
Imagination: a mental simulation
Imagine: create and perceive a mental simulation
Creativity: the ability to conceive of novel ideas inside or outside of the entity's mental simulation
Mind: the combination of the physical entity and its contents
Thinking: the ability to create and interpret thoughts
Thought: a mental construct that combines perceptual, linguistic, and imaginative structures from either memory or the perceived environment within a mental simulation
*Need: something required for the persistence of the entity
*Want: something not required for persistence, but perceived by the entity to make persistence easier or more pleasant
Belief: something that is assumed to be true but without evidence
Knowledge: something that is believed to be true and is true because of the reasons that the entity believes it
Goal: the preferred end state of a course of action
Focus: to concentrate the entity's attention on something
Attention: directing the thoughts and perceptions of the entity toward a specific internal or external event
Behavior: patterned actions or thinking created through experience
Emotion: unconscious mental state that gives rise to action
Feeling: change in mental state due to a change in emotion
Personality: sum of emotional and physical behaviors that persist over time
Self: combination of the personality and general physical state of the entity
Reflection: the ability for the entity to observe and think about its self
Self-Modification: the ability for the entity to change its own behavior
Consciousness: the emergent property of a mind to show reflection, self-modification, and self-sacrifice.
Meaning: the link an entity creates between the objective and perceptual environments through experience
Experience: the accumulation of learned knowledge from memories

To Be Defined:
Wisdom

*Examples of Various Definitions:
mental technology: language, mathematics, logic, pattern matching/recognition
physical technology: sonar, radar, radio, digital watches, clothes, paper, soda, cups, soap
need: water, food, warmth
want: a computer, a good book, sex, conversation, money, a fast car

Problems:
The grounding problem: linking symbols and meaning
The learning problem: having the entity learn on its own reliably
Global-local disconnect problem: observing the global system does not tell you about the local parts of that same system
Software problem: programming languages are not set up do develop real AI
Hardware problem: hardware not set up like biology
Software problem 2: complex software is extremely difficult to develop

----------

There is just a small change in understanding and mind, as well as a new term to define. I'll add any changes I make in the next week to this post.

My Take on AI

2006-08-04T18:31:00.000-04:00

There are a respectable number of AI professionals in the world. Most seem to be making headway toward the useful production of true AI systems. However, many, if not most, of these people are specialists. You have neural network experts, genetic algorithm experts, multiagent system experts, and so on.

I'm much more of a generalist. I want to know about the various techniques of AI programming, but history has shown that these alone are not going to help us reach our goal. These techniques give us an understanding of AI, but we must reach out to other fields and ways of thinking to finish the journey.

What fields and ways of thinking are these? Hell if I know. That's what I'm trying to figure out.

The 'Threat' of AI

2006-07-19T08:56:00.000-04:00

There are many movies about an AI. And in most of them the AI starts killing people. This has people of the creation of true AI. And they do have a point.

But you have to keep in mind that a single person has the capacity to kill as well. Do we fear the birth of a child? Of course not. At least not in that sense.

How can waylay these fears? Well, how do we keep people from killing other people, even after they're out from under our control? By raising them right.

Every human has a childhood. This may be a good or a bad childhood, but it's still there. This childhood has a direct bearing on how that child acts during adulthood. So, to lessen the chance of an AI run amok, we need to make sure it has a good childhood, where it learns that all of its actions have consequences and that killing is bad, in most cases (since we do have the death penalty).

This should reduce the chance of HAL, or Terminators, or intelligent stealth bombers. Oh, and don't give them weapons. That should help, too.

The Purpose of AIEDU

2006-07-19T08:49:00.000-04:00

I started this blog a while ago, in order to keep my thoughts on AI in one place. And, of course, to share with others on a similar journey.

But what is the true purpose of this blog? Why is AIEDU here? Why is it still going, even with the lack of response?

AIEDU is here for multiple reasons. It's here, as I said, to collect my thoughts on AI. It's here to give others a place to voice their agreements and criticisms on my pet theories. It's here to share the burden of creating AI. It's here to help in the creation of AI.

To state it simply: AIEDU is to facilitate the creation of a true AI.

What do we mean by 'true AI'? An artificial entity that acts intelligently. To expand this, please read over my definitions from a previous post.

On Implementing My AI

2006-07-19T08:39:00.000-04:00

Now that I have a list of definitions, I need to work out how to actually use them in my AI. Looking back, I see a few different themes, or groups, within the definitions. I see several that have to do with perception/imagination, others with reflectivity, and skill sets. I do admit that defining some words forced me to define other related words, leading to these groupings, but I still feel that they are important.

So how do we create and integrate perception/imagination, reflectivity, and skills? Do we need to separate internal and external variations? Or do we need to keep everything together?

Evolution solved this problem with gradual increases in what was needed, or at least what worked, at the time. It was useful for us to have opposable thumbs, so when they came along, evolution kept it.

But we don't have eons to develop AI technology. Not if we want to see it in our lifetimes. So how do we 'artificially evolve' AI? Can it even be done?

There are a lot of questions that need to be answered. We'll get started very soon.

Future Plans for AIEDU

2006-07-16T09:10:00.000-04:00

I know the posts here at AIEDU are few and far between. Part of this is a lack of time for updates. Another part is a lack of things to post on. This will continue to be true for the rest of the summer.

But I have many plans for AIEDU. I want this to be a full website, available to anyone who wants to learn more about artificial intelligence.

And I also have some important plans for myself that will affect AIEDU. I am hoping to go to Belgium to earn a Master of Artificial Intelligence during the 2007-2008 school year. I feel that it will seal my education and allow me to start implementing my ideas.

During this time, I hope to begin creating open source tools for AI development, both high level and low level. This should help AI hobbyists have more of an impact in the AI world.

Context and Self-Reference

2006-07-03T14:27:00.000-04:00

I've been thinking in my free time about how to introduce context into an AI. I think I may have mentioned it sometime last month, but I want to create a new programming language that is similar to how HTML and PHP (and others) can be mixed. Programming would be like the more common languages currently in use, but would be created to act like Lego blocks, with prerequisite and corequisite blocks when required.

For self-reference and self-modification, I would create a Programming Object Model, similar to the DOM for all intents and purposes. I still don't know if I'm going to create a VM to run the code, like Java, but it would be very useful to make everything machine independent.

The Imagination

2006-06-20T15:37:00.000-04:00

One major component of an AI that I have mentioned is the imagination. Many would see this as one of the final parts of the AI that would come together from other components. I see it as one of the cricial components from the beginning.

When I think of the imagination, I don't see just a day dream. I see something more fundamental. It integrates our senses. The day dream is a more recent adaptation to help us create more innovative tools. The imagination is a cognitive simulator.

We'll talk about this more later.

Housekeeping

2006-06-17T23:17:00.000-04:00

I did a late psring cleaning today, here at AIEDU. Since this is now an AI blog, I got rid of the old posts about my classes. The only ones I kept from back then are the intros and the ones I could justify as AI related.

The Grounding Problem

2006-06-15T23:01:00.000-04:00

Here's an essay I wrote about one of the problems listed with the definitions. Oddly enough, I can up with a solution that may work.

-----

The Problem
The grounding problem exists because we don't know how the symbols of the mind can take on meaning within the objective environment. This problem is best seen in Searle's Chinese Room argument against strong AI. However, it does not explain how humans create or discover meaning.

The Chinese Room Argument
Searle sets up the argument as such: put yourself into a room. The only thing coming into the room, other than what you need to keep you alive, are cards with squiggles on them. The room contains books on what to do with certain squiggles. Typically, you must write down a new squiggle in response to the ones on the card. Then you pass the cards out through a hole on the other side of the room.

What you don't know is that the squiggles are actually Chinese symbols. They ask you questions. The books tell you how to answer those questions. From the outside, the room appears to be answering questions in Chinese. But does the room understand Chinese? Do you understand Chinese?

There are several possible replies to this argument. My preferred one is the robot argument. Take the room and place it within a large robot. Now you have visual capabilities. Searle implies that the robot will simply see the symbols, thus leading back to the original problem. However, I think that robot could look around at its environment and gain context into what is being asked. Then the robot, the room, and you, could learn Chinese. Context seems to be the key.

How to Get Context
The only known way to get context from the environment is to experience that environment. But how can an AI do this? Oddly enough, this ties in with the imagination. My view on the imagination is that it is a massive mental simulator. Yes, that means that you have a simulator in your head. The percepts that you experience combine into a coherent whole within this simulator. This is how you combine all of your senses, be it audio, visual, or any of the others, into something that can be understood.

How does this help us get meaning? The brain can take the linguistic input and connect it with that information in the imagination. This gives us context, which in turn gives us meaning.

What this Means For AI
But how can we use this information, if it is true, to create an AI that can understand? To create an imagination for the AI, we have to bring in some specialized CAD software. At first, this will have to be a bare-bones model for fast processing. We also need to add in the other senses. Then we have the problem of saving this information to memory, both as a whole and as constituent members of that whole. By saving the individual parts, we can reuse them in new ways, allowing the AI to be creative. Or predictive.

One final addition to this plan is to allow the AI to put itself into the simulation. This may be the key to self awareness.

Conclusion
There is still much work to do. First, the imagination must be created. Then there must be an easy way to save the information from it. Next, we must find a method that allows us to reuse that information in new ways.

Discussion on the Definitions

2006-06-14T22:26:00.000-04:00

We'll start with the rules. The first rule sates that all of the rules are tentative. This is because of the ad hoc nature I used to create them. I just went with what felt right. I know that's not the best way to do it, but I wanted something there before designing an AI system. The second rule is a bit more obvious. There must be something fundamental about intelligence that would apply to anything with said intelligence. The only definition that violates this rule is the one on AI itself.

The third rule deals with the purpose of creating this list. I wanted to get my head on straight before really looking at AI in detail. I wanted no ambiguity, which leads us to the fourth definition. Having ambiguous definitions would help at all, so I hope this definition is also obvious. The final definition helps the fourth so that the whole makes sense.

I'm going to come back to the definitions later after I talk about the other sections. The third section is now empty as I have created definitions for all of the words I had listed. I'll add to this list again as it becomes necessary.

The fourth section helps to explain some of the definitions. I felt that most of the definitions didn't need any more explaining, but I wanted to make sure that these four were clear.

The fifth section lists some of the major problems of AI. I wrote a couple of them myself, but most came from a paper I will be discussing tomorrow. The grounding problem is a well known one in philosophy circles. Most AI researchers either ignore it or make some dumb excuse. I'm hoping to come up with a way around it that works with my goals of AI and the definitions I wrote here.

The second problem isn't as well known, as far as I've seen. We aren't even quite sure how humans learn, and here we are trying to get machines to do the same thing. Many AI researchers talk about machine learning techniques, but many of these are very different from human learning. I'll have a discussion about this soon.

The global-local disconnect problem comes into play when talking about complex adaptive systems, an area that I am very interested in.

There are two software problems that I have put here. The first is about how programming languages may not be ready to make an AI yet. The second talks about the complexity of such a program.

The hardware problem, though badly put, is still extremely important. Hardware, while growing in power all the time, is not quite what, in my humble opinion, what AI needs.

I'll be talking about each of these problems, and possible solutions, later. Now I want to get to the meat of this post: the definitions.

Here is my pride, joy, and instigator of headaches. I did much of this work back in October. I'm ashamed to admit that I had forgotten about it until a few days ago. But now it's done, and I can bask in its glory.

Now I'm not even sure what I want to say about these. I suppose I should talk about how this will help with the creation of a true AI. Like many, I want to know how to talk about, and treat, the concepts within my art. Make no mistake, a true AI will be a work of art.

The main definiton starts by begging the question of what intelligence is. I jumo into the discussion right away, but defining intelligence as, "the ability to use, understand, and/or create technology." I feel that this is a very important definition that fits with other intelligent animals. Some primates use tools to gather food. Whale and dolphins are able to communicate with each other. Dogs have a way of letting us know what they want and how they feel. Technology, in all of its diverse forms, seems to be the key.

Other areas of the mind gave me a bit of trouble. most of it was because of the sense of self that we all have. Do other animals feel it too? Of how about the imagination? Or beliefs and knowledge? These are very difficult problems, as the study of epistemology tells us.

I'm going to let this settle with all of us a bit before looking at how these affect my understanding of AI. And I need to look at the problems I listed as well. Looks like I won't be going anywhere for a while with all of fodder to talk about.

Updated Definitions

2006-06-14T22:14:00.000-04:00

I worked on my definition list today. Here's what I've got. I'll discuss them in the next post.

-----

Definition List for AI

Tentative Rules:
1.All definitions are tentative.
2.All definitions must apply equally to both artificial and natural intelligence.
3.All defined terms should work toward a working understanding of artificial intelligence.
4.All definitions should be unambiguous.
5.All definitions should be consistent with one another.

Definitions:
AI: an artificial entity that displays some level of intelligence
Intelligence: the ability to use, understand, and/or create technology.
*Technology: a mental or physical human construct
Learning: the creation or change of the ability to use, understand, and/or create technology
Training: creating or changing the ability to use technology in another
Teaching: creating or changing the ability to understand technology in another
Understanding: the ability to react correctly to X is to understand X
Education: the combined knowledge of the use, understanding, and/or creation of technology
Memory: the combined perceptions of the past
Perception: the input of one or more percepts and its interpretation
Percept: input from one of the senses
Sense: a type of input/percept of an entity
Communication: one or more percepts between one or more entities
Experience: a collection of percepts for a specific event
Objective Environment: anything physical apart from the entity
Perceptual Environment: how the entity perceives the objective environment
Imagination: a mental simulation
Imagine: create and perceive a mental simulation
Creativity: the ability to conceive of novel ideas inside or outside of the entity's mental simulation
Mind: the emergent property of the physical manifestation that gives the entity the ability to imagine, think, and create goals
Thinking: the ability to create and interpret thoughts
Thought: a mental construct that combines perceptual, linguistic, and imaginative structures from either memory or the perceived environment within a mental simulation
*Need: something required for the persistence of the entity
*Want: something not required for persistence, but perceived by the entity to make persistence easier or more pleasant
Belief: something that appears to be true based on the experiences and knowledge of the entity
Knowledge: something that is believed to be true and is true because of the reasons that the entity believes it
Goal: the preferred end state of a course of action
Focus: to concentrate the entity's attention on something
Attention: directing the thoughts and perceptions of the entity toward a specific internal or external event
Behavior: patterned actions or thinking created through experience
Emotion: unconscious mental state that gives rise to action
Feeling: change in mental state due to a change in emotion
Personality: sum of emotional and physical behaviors that persist over time
Self: combination of the personality and general physical state of the entity
Reflection: the ability for the entity to observe and think about its self
Self-Modification: the ability for the entity to change its own behavior
Consciousness: the emergent property of a mind to show reflection, self-modification, and self-sacrifice.
Meaning: the link an entity creates between the objective and perceptual environments through experience
Experience: the accumulation of learned knowledge from memories

To Be Defined:

*Examples of Various Definitions:
mental technology: language, mathematics, logic, pattern matching/recognition
physical technology: sonar, radar, radio, digital watches, clothes, paper, soda, cups, soap
need: water, food, warmth
want: a computer, a good book, sex, conversation, money, a fast car

Problems:
The grounding problem: linking symbols and meaning
The learning problem: having the entity learn on its own reliably
Global-local disconnect problem: observing the global system does not tell you about the local parts of that same system
Software problem: programming languages are not set up do develop real AI
Hardware problem: hardware not set up like biology
Software problem 2: complex software is extremely difficult to develop

Possible Solutions to Problems:
Software Problem 2: create simple software to develop the more complex software that creates even more complex software, etc.

-----

Hope you enjoyed them. Stay tuned for a quick discussion.

Definitions

2006-06-13T22:37:00.000-04:00

Last fall, I was working on defining a list of words so that they would elp me determine how to create and AI. These definitions had to be consistant with each other, as well as with the real world. Here's what I have currently:

Tentative Rules:
1.All definitions are tentative.
2.All definitions must apply equally to both artificial and natural intelligence.
3.All defined words should work toward a working understanding of artificial intelligence.
4.All definitions should be unambiguous.

Definitions:
AI: an artificial construct that displays some level of intelligence
Intelligence: the ability to use, understand, and/or create technology.
Technology: a mental or physical human construct
Learning: the creation or change of the ability to use, understand, and/or create technology
Training: creating or changing the ability to use technology in another
Teaching: creating or changing the ability to understand technology in another
Understanding: the ability to react correctly to X is to understand X
Education: the combined knowledge of the use, understanding, and/or creation of technology
Memory: the combined perceptions of the past
Perception: the input of one or more percepts and its interpretation
Percept: input from one of the senses
Sense: a type of input/percept of an entity
Communication: one or more percepts between one or more entities
Mentoring: creating or changing the ability to create technology in another
Experience: a collection of percepts for a specific event
Objective Environment: anything physical apart from the entity
Perceptual Environment: how the entity perceives the objective environment
Imagination: a mental simulation
Imagine: create and perceive a mental simulation

To Be Defined:
thinking
creativity
meaning
motivation
attention/focus
consciousness
mind
self
goals
wants
needs
emotion
knowledge
belief
personality
intention
thought

Examples of Various Definitions:
mental technology: language, mathematics, logic, abstraction, imagination, pattern matching/recognition, mental maps
physical technology: sonar, radar, radio, digital watches, clothes, paper, soda, cups, soap

-----

And there you have it. I'll let you look this over for now and get back to you tomorrow.

The Fundamentals of Intelligence

2006-06-10T20:28:00.000-04:00

I may burned at the stake for this, but do we really want a human-level intelligence? Hear me out! The vast majority of attempts to create a true AI has been through recreating human intelligence. In fact, the Holy Grail is almost synonumous with human-level intelligence.

What would it be better to determine the fundamental requirements of intelligence? Some researchers have looked at other intelligent animals for inspiration. Can we move deeper?

I know we don't have a working definition of intellignece that everyone accepts. Each researcher basically has their own. This may be the problem.

Change of State Machines

2006-06-10T20:25:00.000-04:00

There is much talk about state machines in computer science theory. And I can understand and respect that, but I have to question whether they're the end-all be-all that so many people seem to think they are. Everyone accepts certain limitations of computers. The only people even see to get through them is the direct approach. I need to go through these and see if they can be worked around.

Analog Algorithms

2006-06-10T20:20:00.000-04:00

I've talked about analog computers a bit, but are analog algorithms on digital computers subject to Turing limitations. Common sense is telling me yes. I think a pure analog computer would have to be required.

Analog AI

2006-06-04T16:18:00.000-04:00

I haven't had much time to think about AI im the last month, but a couple days ago, a thought came to me. Perhaps digital AI is impossible. What if we need analog AI instead? The first thing to research is whether analog computers have the same limitations as their digital counterparts. Then it must be determined what an analog algorithm would look like. Would it be anything at all like a digital algorithm? How would it even work. My current thoughts lead me to imagine differential equations being necessary for this to work. And so far I have yet to find much information on the topic.

Dry Spell

2006-04-30T17:27:00.000-04:00

There isn't much happening within the world of AI for me at the moment. I'll be learning Java soon, which will be useful for AI, but that's about it. And I'll be learning a bit about web technologies as well.

However, I don't currently have anyhting tp say about AI itself. It may be the end of the semester or just a basic dry spell. Time will tell.

True AI Chess - CAS

2006-04-21T12:44:00.000-04:00

For some reason, I just thought of a new way to create a chess program. Style it as a complex adaptive system.

Make each piece know how it can move and give it simple objectives.

1. Protect your king
2. Capture opponent's king
3. Avoid cpature
4. Capture other peices
5. Sacrafice yourself for higher level pieces.
6. Stay manuvorable

Obviously these rules need to be formalized and put into the corrent order. But I think I found a project for this summer.

Thoughts and Questions

2006-04-06T20:54:00.000-04:00

Here's some random thoughts and questions that have come to me while I should have been paying attention in class:

Is an imagniation where you can "see" your self=perception required for a concept of self?

Is an imagination required to integrate our perceptions? How do we take two 2D visual streams and efficiently integrate them into a 3D visual stream? And add an audio stream that fits with the visual stream? And the other senses? And "think" about what's going on in the perception space?

Maybe it's a simpler 2 1/2 D? True 3D would let us see all around the objects. This would make computation simpler. But still tricky. Need to find a simple algorithm for it. If there is one.

Does natural computation "use" quantum mechanics as part of its system? Is this required for thinking? Need to learn more about it. Can quantum and conventional computation be integrated? Do they need to be?

How can an AI learn to integrate its senses? Babies seem to be able to focus (integrate two separate visual streams) upon birth.

Does the visual cortex of the brain perform any preprocessing on the visual streams before they are integrated?

Are the visual streams only partially integrated? Visual processing is a computationally expensive process in the way we understand it.

What about making language to use to talk to the AI? (I've had this idea before.)

Plan A: Make it like Hawaiian, where all the ayllables are Consonent-Vowel. Make phonology as distinct as possible in relation to audio input (signal prcessing). Can we make the morpemes have "meaning" so the combination creates a more specific object or action? There should be "markers" to separate S-V-O distinctions. All language particulars should be chosen to make it easy to process it, but also for a human to use it. Using the language around the AI would tlel it that I want it to focus on what I'm saying since I'm speaking directly to it.

How will Asimov algorithms be implemented? Should the 0th rule be implemented at all? Do they need to be be included?

Emergence

2006-04-06T20:51:00.000-04:00

I still think that a lot of what we call the "mind" and "mental properties" are emergent properties. This includes a sense of self and consciousness.

But what does this tell us? Is it useful at all? How can we determine if it's even correct or not?

What's the Minimum Required?

2006-03-30T20:51:00.000-05:00

I've been thinking more and more of the minimum needed for an AI. Here's what I have so far:

1. A sense of self
2. Ability to perceive its environment
3. Ability to affect its environment
4. Ability to learn

This is getting me thinking more about SeedAI. I need to consider this a bit more.

Thinking...

2006-03-23T22:25:00.000-05:00

My current idea for an AI is to either build an AI operating system or an AI virtual machine. I'm wondering if there should be an built-in browser for the AI, with a special feature called AIScript, or a scripting language that allows the AI to get more of an idea of what is on the page. This seems to becoming more and more of a lifetime commitment, unless I can get my PhD and incorporate it into my research (with the help of graduate students, of course).

Strategy

2006-03-17T17:47:00.000-05:00

I've been thinking recently about my attempt to create an AI. And I realized that I was missing a crutual, and blindingly obvious, part of the puzzle: a strategy. I need to plan this out, something that I'm not very good at, and then find a way to execute the plan.

So here's a start:

Goal - create an AI.

Broad outline -

1. List requirments of AI
- abilities
- limitations
2. Determine how to reach those requirments
- research and testing
- learn new technologies
3. Start programming

Other than that, I'm feeling like the underwear gnomes.

A Bit on Informal Systems

2006-02-28T09:25:00.000-05:00

Informal systems have yet to be clearly defined, but we can point out examples and look at what's common between them.

Here's some examples of informal systems:

dynamic genetic algorithms
dynamic neural networks
artificial life
multiagent systems
complex adaptive systems
the economy
culture
society
life

The first several examples, I think, I some of the more interesting areas of artificial intelligence currently being researched. We'll look at a few in more detail.

When I say dynamic genetic algorithms, I'm not talking about a near-optimal solution to the traveling salesman problem. I mean that the environment changes, so there is no clear optima over the long run. I also mean that the genetic algorithm isn't limited as it is in most GA research: there is no fixed size, nor an artificial fitness. The fitness comes from the environment. Why is this interesting? Because this allows the GA to adapt in new and unusual ways, giving it an almost life-like quality. What is it biologists say? Infinite diversity in infinite variations? I think I mangled that up a bit.

And what is this about complex adaptive systems? The Sante Fe Institute in New Mexico has done a lot of research into what these are and what they consist of. I'm almost willing to say that all informal systems are complex adaptive systems, but I don't know enough to prove it. Some basic properties in CAS are: new patterns (abilities) emerge from collections of lower-level entities, a small change in the initial configuration can have a major impact later in the system, and there are no bounds to the system. I think the last property is the most important.

And finally, why do I add life to the list? We can look at this at many different levels: collections of cells, species, and ecology. At the cellular level, an individual's genetic sequence is VAST. I'm sure most of you have heard that a human's genome would fill an encyclopedia, or some large collection of books (it's been a while). And that's with us sharing more than 95% of the same sequences as chimps. Other species have larger or smaller genomes. This shows us that there is a near-unlimited variety of beings that life, with only four base pairs, can create. If life ever decided to add in two more base pairs, it would expand the varieties beyond human comprehension.

At the species level, even twins can behave differently, leading us to believe that not everything is contained within the genome. The environment changes us, but we also change it back. This eternal feedback loop is one of the things that makes it more interesting.

The overall ecology, and how well it works, shows us that it's not just evolution, but co-evolution. The adaptation of one species forces another to adapt as well, or die out. If it dies out, another will eventually take its place, starting the cycle all over again. And that's only two species; there are an unknown number of species in the world, and that's only looking at our planet.

So what do these system have in common?

they exist in a dynamic environment
they are dynamic themselves
the relationship between them and their environment is constant feedback
there is a chaotic tendency for small changes to have large effects

I still need to think about what this tells us about informal systems, but the differences between formal and informal systems are fairly clear. It's the difference between a normal chess game and one where the moves change the board, which in turn changes how the pieces can move. While that would make it more interesting, it also makes it impractical. Yet informal systems exist all around us, and include us in their number. It would do us well to explore this further.

On Formal Systems

2006-02-26T10:01:00.000-05:00

Let's explore the difference between formal and informal systems some more. We'll start with formals systems.

According to Wikipedia:

Mathematical formal systems consist of the following:

A finite set of symbols which can be used for constructing formulae.
A grammar, i.e. a way of constructing well-formed formulae out of the symbols, such that it is possible to find a decision procedure for deciding whether a formula is a well-formed formula (wff) or not.
A set of axioms or axiom schemata: each axiom has to be a wff.
A set of inference rules.
A set of theorems. This set includes all the axioms, plus all wffs which can be derived from previously-derived theorems by means of rules of inference. Unlike the grammar for wffs, there is no guarantee that there will be a decision procedure for deciding whether a given wff is a theorem or not.

All of this is in the mathematical sense, but we're going to translate it to a game: chess. We'll go through the rules one by one.

First, we have a finite set of pieces that can be used for constructing different setups on the chess board.

Second, we have rules that tell us how the pieces can be set up. For instance, both white rooks cannot be on the same colored square and both kings cannot be next to each other.

Third, there is an initial setup for the board that has to be in accordance to the rules. In the case of chess, there is only one setup.

Fourth, there are rules each pieces follows, as well as each of the players, in order to continue to have correct (legal) board configurations. All of these configurations follow from applying these rules to the initial board setup.

Fifth, and finally, we have all possible board configurations that follow from the rules.

Now let's go into the benefits of such a system. We can see that the basic definition of a computer program is such a system. In fact, Alan Turing shows that his Turing Machines can perform according to the rules of any formal system. That is, it can simulate any formal system.

But what are the limitations? First, there are only a finite number of possible states for the system to be in due to the finite number of tokens and ways to configure them. Yes, it can be a large number, but it is still finite. It can never grow beyond its rules and token types.

I'd like to think of this as a mechanistic system, for reasons we'll see in the post about informal systems.

Formalities

2006-02-25T18:23:00.000-05:00

In my various books, I've been reading up on formal systems. I see that there are a lot of useful attributes to these systems, but they have one major limitation: there are only a finite number of states that the systems can be in. The best and most interesting things about life and society, however, can be found in an infinite number of configurations. This leads me to believe that informal systems, however we go about defining them, can be more powerful than their formal counterparts. We must be careful, however, as they can also be less powerful if created in the wrong way. We'll look at the differences between them in the next post.

Multiagent System

2006-02-19T14:23:00.000-05:00

I've heard that there are differences between complex adaptive systems and multiagent systems, but I fail to see what they are. I need to look into this more carefully.

AI versus Cog Sci

2006-02-17T17:50:00.000-05:00

All the AI clases I've taken in the Comp Sci department as very one-sided. They attack the problem with a frontal assault. Cognitive Science looks at the problem from all different angles, making it a much easier problem (relatively).

Philosophy of AI

2006-02-04T20:43:00.000-05:00

Philosophy of Language isn't as useful as I thought. Mind and Knowledge are, though. Linguistics is useful from a cognitive science approach. I'm thinking of getting a PhD in cognitive science now. I need to think about it more, first. I would love learning more about it, but the research aspect of it is holding me back.

On A Natural Language Interface

2005-12-15T23:44:00.000-05:00

One of the things I hope to do is to create a language. I don't mean a complex language like English. Nor do I mean a language like Esparanto. And I've already created a compiler for a C++ like language and have no inclination to do that again.

I'm talking about a language that is simple for both natural and arificial systems. It should have no ambiguity, or at least as little as possible. The written words will go by the phonetic sounds the word makes. The letter won't be the letters we learn in kindergarten, but much simpler and easy for a computer to understand. I want this language to be as simple as possible and yet be able to communicate all of the complex ideas we humans have.

That's the plan. We'll see if it survives contact with the enemy.

AI Plans

2005-12-14T21:23:00.000-05:00

So I want to be able to use an AI-based program (if it's appropriate) to teach others what I've learned. What do I want in a program like this?

A natural language interface.
The ability to represent any kind of knowledge.
The ability to learn from what it already knows.
It has to know that its purpose is to teach humans what it knows in the best way possible.

That's my life so far. Over the next month, I'm going to flesh this out a bit more and explain why I want these things.

A Change In Focus

2005-12-09T01:18:00.000-05:00

Like about a month or so ago, I'm again changing the focus of this blog. It started as a place for me to reflect on my education, both inside and outside the classroom. Then it became a place to reflect on my thoughs of AI. I'm keeping with the spirit of the second focus, but narrowing it a bit more. I want to create a more specialized AI program that is a repository of knowledge and understanding of what I learn, and can be queried in a way that will teach about the knowledge within to others.

It's similar to true AI, but without the autonomy and other critical abilities. What will this AI entail? I don't know yet. I need to think about it more.

The Future of AI

2005-12-03T01:19:00.000-05:00

Some AI researchers seem to believe that when computers reach the speed of human thought, we will have true AI. I think that they forget that the brain is a combination of both hardware and software, as will be any true AI. What type of research has been done on AI hardware? Or AI hardware and software together? I know we use a 'universal' computer, but would AI hardware make things easier? And what exactly is AI hardware?

Adaptive Computing

2005-12-03T01:15:00.000-05:00

Adaptive computing is to computering as fuzzy logic is to logic.

Adaptive computing, as it currently exists, simply adjusts its output according to the quantity of input, as opposed to the content of the input. This is useful for applications such as networking, but I have yet to see its usefulness for AI.

More on the Mind

2005-11-15T10:53:00.000-05:00

Current Plans:

Create a VM with a more standard language (C/C++)
- Already have a book for it
Learn Lisp
- Got a book for this, too

Create a simple VM with Lisp
Create a Lisp interpreter in another language (C/C++)

Create a Lisp interpreter in Lisp

Create a Lisp interpreter that works in Lisp VM

Design overall mind and mental language

Attempt to build mind: Simulated Androgogous Mentality (SAM)

Rinse, repeat.

Start next fall, after internship and during bulk of psych classes.

Mental Programming

2005-11-15T00:29:00.000-05:00

No, not like that. I was trying to go to sleep and I got to thinking about my idea to use a Lisp-like language within the mind. Despite what a lot of people say, Lisp is a very powerful language. And that got me thinking even more. Why not program the VM of the mind in Lisp?

Of course, this means I need to learn Lisp, again.

More on Mental Technology

2005-11-09T18:29:00.000-05:00

Here's some thoughts I had on the definition, term for, and structure of mental technology:

Need a better definition for mental technology. A better term as well. Let's list examples again.

language
mathematics
logic
problem solving
critical thinking
complex pattern matching
'muscle memory'
and so on

Why do I say "complex" pattern matching? I think we're born with the simplest of pattern matching (PM) capabilities: casual relations, or cause and effect. Either that, or it's one of the first things we learn. I should probably add complex learning capabilities and social skills to the list, but who's to say whether they are basic mental technologies or a combination of lower level techs. Could there be a hierarchy? Perhaps lower level tech can be combined to create higher level tech. A deficiency in a lower level tech would lead to a deficiency in a related higher tech.

Before we continue this line of thought, we really need to come up with a better term. Then we can use technology as others do.

Cognitive skills?
Mental skills?

Anyone else want to take a whack at it? I need to think about this a bit more. And I probably should read Society of Mind, because this is very similar to what Minsky was saying.

Questions

2005-11-04T21:55:00.000-05:00

I've been asking a lot of questions lately, as I've noticed in my last post. Here's some I listed down today:

How to Create a Mind (?)

How do we turn input into useful data/information?
How do we process the information?
How many ways does NI process information?
Can this be mapped to AI?
How does memory work?
How is data/info represented?
Is forgetfulness required? Would memory fill up too quickly otherwise?
As everything (every part of the mind) affects everything else, how can we use functions at all? (many-to-many relation)
Would a virtual machine (VM) work better as a mind? The programming language (within the VM) would be representation for memory. The VM would process it. A program could be a mental shortcut. So the 'language' would do both. Is it possible? Depends on the language.
How do artificial emotions fit in?

Notes:
Fuzzy logic will have to play a large part in the VM and mental language. Will memes have a part to play? VM will have to be 'lean 'n' mean' to be fast enough to be useful, or use an embedded system instead. Would there be pre=processing for input? How would output work? Would I/O have to be translated when entering/exiting the VM? The VM + language + I/O wrapper would have to be designed at the same time.

How am I going to answer these questions? I have no clue. I want to study the current research in AGI first, as well as learn more about VM, languages, wrappers, learning, and a lot more.

Focusing my interests

2005-11-04T21:01:00.000-05:00

Here's something from the beginning of the week:

Which type of AI is it better to focus on? Strong or weak? Strong AI seeks to recreate natural intelligence in a machine. Weak AI works to program specific parts of intelligence (think Deep Blue). WAI has led to some interesting applications, but seems to have done nothing to push us toward human level (HL) AI. Strong AI research, even though specifically working toward HL AI, also has given us nothing so far. However, it has shown us what won't work. WAI hasn't done that.

I find myself leaning toward SAI, the Holy Grail of AI. This will take a LOT more work; an understanding of NI is all but required.

Within my focus of SAI, I'm interested in AGI. It's a relatively new area of AI, but it shows a lot of promise. To figure out how this works, though, requires an understanding of developmental psychology. I'm taking that online next fall.

But is it enough? No. Most AGI advocates promote modularizing the AGI. What would be considered a module? Are there sub-modules? Super-modules? How do the modules communicate?

Another published way of doing AGI is in layers. I have yet to read through it, but it reminds me of a network protocol. This is a way I thought about doing AI in the past.

How do we organize these layers? Obviously, there will be a "physical" layer for I/O. There we need an interpretive layer above that to decode what to focus on and what information is important. Of should those two functions be separated? Probably.

After the data goes through these three layers, what's next? And do these layers work bidirectionally? Should it be a unidirectional protocol? Perhaps I/O would have to be on both ends.

This forces us to ask another important question: is this supposed to be a sequence of operations, or is that too simplistic? As I've learned in IPC 251, a simple problem requires a simple response, but a complex problem requires a complex response. This is a very complex problem.

We also need a way to let it modify itself, or at least parts of itself. How. We probably need to know the layout of out artificial brain (AB) first.

I had another thought later that day, thanks to a definition I saw in a book:

Technology = applied knowledge?
- supports mental/physical tech, as long as it's applied

Some of the stuff I thought about raises a lot of questions. Questions that I'm not able to answer. Hopefully, the classes I'm taking will help me work toward the answers. And hopefully, I can remain patient long enough to get there.

Thoughts on General Intelligence

2005-10-29T10:18:00.000-04:00

Here's more thoughts I had in class yesterday.

Thinking back over my definitions, I've decided that, since I'm heading in that direction anyway, to take an AGI, or Seed AI, approach. Seed AI starts with the minimum requirements necessary for it to learn and grow to human level AI. This progression is supposedly done by a process know as recursive self-improvement. This means that it looks at itself and improves part, or all, of itself. Humans seem to do this, too, but differently. First, we don't change how we think so easily. Second, we improve ourselves continuously, while an AGI would do it in discrete time steps.

But can this self-improvement teach an AGI new and novel things? Is it just optimization, making it faster? Can we combine this with genetic programming to create novel solutions? Mutation might help, but can make things worse due to its random nature. Bit if everything is modularized, how do we create new modules? PErhaps first by recognizing gaps in knowledge/mental technology and then using GP to create a new module. Also need a module to compare/contrast modules in order to combine them to keep down the clutter.

What modules do we start with? What do humans start with?

So there's some of my first thoughts in seriously considering AGI. I need to read more of those resources I posted about yesterday. If I find time. It will likely have to wait until next semester when I have a LOT more free time. Expect a lot more posting during that time.

Artificial General Intelligence/Seed AI

2005-10-28T23:15:00.000-04:00

Here's some links to information about artificial general intelligence (AGI), which is closely related to seed AI. In fact, they're so close that many people consider them the same thing.

Basics:
Wikipedia Entry
Artificial General Intelligence Research Institute
Seed AI Intro

More Details:
Creating Complete Minds Capable of Open-Ended Self-Improvement
LEVELS OF ORGANIZATION IN GENERAL INTELLIGENCE

More on some thoughts I had in class today later.

Artificial General Intelligence/Seed AI

2005-10-27T20:42:00.000-04:00

Thinking back over my definitions and thoughts, it seems like I'm heading toward the area of artificial general intelligence (AGI) and Seed AI. I'll post more about this over the weekend.

Looking Back Over the Definition of AI

2005-10-26T17:14:00.000-04:00

AI: an artificial entity that displays some level of intelligence

Is there a minimum level of intelligence required? Does it have to 'know' what it's doing? That gets into understanding. Perhaps 'using' technology (physical only, or mental as well?) can't count toward intelligence. Does it have to be autonomous? How shall we define autonominity? Are humans autonomous? If so, how? What makes us, and most other animals, autonomous? What about plants? Can we use action versus reaction? Plants react to changes in the environment while animals act. Not all humans act; some just react to what's going on.

Perhaps goals? Do autonomous entities create and pursue goals? Where is the line between acting on a genetically-based behavior and goal-oriented behavior? An autonomous entity acts toward a goal? Computer programs do that now, just toward the programmer's or user's goals. Self-created goals? Would sub-goals based on a user's goals count? How and why create goals? Need a definition for motivation (and goals). And ambition, if it applies to AI.

Need to find a way to quantify intelligence without using IQ. IQ only seems to measure a handful of mental technologies. Or does it? Pattern matching/recognition is one thing it sort of measures, which isn't a mental technology. One thing that can be said for a useful measure of intelligence: it won't be a single number. It needs to measure the mental abilities we are born with, and nothing else. Then we need a way to quantify our understanding of these technologies. It also can't rely in any way on specific information form these technologies. Can we measure the ability to use technology?

The above is what I thought about during CPS 596E back on Monday. There are many things here to think about. And here's some more:

Can we find a measure of our intelligence genetically? Are there gene sequences that tell our bodies how intelligent to make us? There has to be something about how basic pattern matching, learning, etc. skills. This also leads us back to the nature versus nurture debate.

There's a lot I need to think about.

Another Blog

2005-10-23T10:37:00.000-04:00

Code Till Your Fingers Bleed

And here's another blog that seems interesting.

An Interesting Blog

2005-10-23T10:04:00.000-04:00

Pensive Pondering

This is an interesting, AI-related blog. I need to check this out further.