(Български - по-долу)
Hey, I realized that I must finally start doing real work in AI. I should complete a Thinking Machine 0.001, 0.0001 or 0.00001 - but something with an architecture which is complete and does something.
An agent which sees, feels something, finds some patterns, try to simulate and predict its inputs. A basis of conversational agent, research accelerator, computational creator; an environment which embodies many applications...
I decided... I should start real development of *something* thinking and creative, not just researching and fantasizing in zillions directions or doing much ordinary research in fields like Comprehension Assistants or Speech Synthesis...
It must learn... It must evolve. Offf... Must stop saying big words, start working on real stuff...
I decided! :))
Well, hope I'll have time soon... And a team... :( :)
...
Реших... Най-накрая трябва да започна да работя по нещо истинско... По някаква архитектура на мислеща машина... И да завърша нещо. Версия 0.001, 0.000001 - но да е нещо пълно. С възприятия, цели, симулация на възприятията и предсказване, обучение и пр... Основа на разговарящ агент, на машина творец... на машина програмист, която се учи да програмира... на Ускорителя на изследователската дейност.... Трябва да свърша нещо реално и да мисля по начини за решаване на реални инженерни проблеми за решаване. И вярвам, че нещата ще тръгнат....
Не само да изследвам паралелно в безброй насоки и да работя по по-обикновени проекти като интелигентни речници и синтезатори на реч.
Крайно време е да насоча въображението си към целта и да локализирам конкретни *инженерни* проблеми, цели и модели и по-важното - да започна да ги решавам.
И машината 0.0001 трябва да се учи, трябва да се развива от входните данни, без да се препрограмира код. И трябва да спра с големите приказки, и да започна да ги подплътявам с реални неща....
Реших! Ще се стегна и започвам!
Надявам се скоро да имам време... И екип. :( :)
7 коментара:
Fine words Todor! Do you realize how many smart people over the last 50 years made the same resolution, & achieved next to nothing? You can't start building a system till you formalize its purpose: functionally define intelligence. This is not a programming problem, it's introspective generalization problem. For my approach, check out the "Intelligence" knol: http://knol.google.com/k/boris-kazachenko/intelligence/
Boris.
Hi Boris, thanks for the comment!
Interesting article! Good to see somebody in the club of Thinking machine dreamers. :)
Of course I'm aware of those people, so? Check my Computational Creativity If you don't do nothing, you will certainly not achieve anything, you should try.
I may also mention that I believe my combination of creativity inclinations is very unique, so I might have more chances. ;)
I'm in this "club" of dreamers since the previous century or so, speculating and writing theories. My main theory about the mind is actually very similar to Jeff Hawkins's one, but it's online in Bulgarian only, since 2002-2004.I've been speculating also about the creativity, since many years, and I do also like introspective analysis, because sometimes you're able to track the thoughts and intentions, and to find good reasons why did you invent this or that...
In very brief my theory of mind is that mind is an universal simulator of virtual universes, which it captures in the data of its senses, and that those virtual universes in mind are at different levels of complexity; the simpler universes are building blocks of the more complex, higher ones.
Theorizing, speculation... Real research and work on a real framework helps finding and formalizing specific and real issues, and since my approach is definitely learning-one, you need to test it with real word input, and must see how to develop real senses and how to predict future senses, so you need to touch real machines.
"Introspectively" I've solved the problem since 4 or 5 years, I sort of feel I know what thinking and creativity is, I've defined it in my theories, but this is not a thinking machine and right, you need to define it in more detailed level and in more functional level.
This level is on a machine that really does something, starting from something simple etc.
From what I can tell, your introspection hasn't produced a quantitative criterion of intelligence yet :). You're right, we have to start from the simplest, but theoretically rather than by trial & error.
How about defining intelligence on the level pixels & patterns, instead of universes & creative writing? You do realize that language-level data is not simple, it's not a start but a heavily processed product of human thought?
As for your creativity inclinations, I have theory about a mindset required for AI discovery: http://knol.google.com/k/boris-kazachenko/cognitive-focus-speed-vs-scope-tradeoff/, see if you agree.
Hi again, Boris!
Well, I guess you haven't read my stuff, while the essence many times bigger than what's in the blog is in Bulgarian and you surely hadn't read it...
In this blog I've mentioned about my speculations about measuring intelligence with regular cognitive requirements which are used by development psychologists, nursery teachers etc. They are very detailed, indeed. I'd like to develop creativity measure, but it will need time...
http://artificial-mind.blogspot.com/2007/11/faults-in-turing-test-and-lovelace-test.html
There are standards of what a baby or a child at age so-and-so should be able to do. And besides that, studying development of children tells a lot about what intelligence is.
...
Of course, language is very "digested". Generally again, "universe" in my theories is something which has casuality, dynamics, physics and of course has patterns. If the environment didn't have regularities, patterns and redundancy, the future couldn't be predictable as it is now, and intelligence wouldn't be possible. Intelligence is about making predictions that came to be true, isn't it?
Language and anything could be modeled as a by-effect of "physics" inside, or between virtual universes.
You're right - "the first" intelligence is visual/spatial (basic light/shape/pattern recognition), tactil (patterns from other sensory inputs), auditory etc. Language develops after there is already a cognitive basis with patterns supposed to be called and structured somehow.
Creativity is imagination. Imagination is simulation of perceptions, simulation of the world (virtual universe), as it's felt by us, it's richer than language we write...
...
If I understand correctly your view about the AI creators, I sort of agree with you.
I agree about the programmers whose big love is coding. I'm not that type, either. I'm master of programming, but it's not my favourite activity, I like to research and to imagine things, coding is a work that just needs to be done. I'm also a writer, a photo story maker, a filmmaker, I draw, I play...
BTW, I believe "Artificial Programming Intelligence" could arrive soon, if somebody sit and work. It's something sometimes I speculate about. I think "API" could be easily developed, if we could teach the system to know what it wants to achieve. I imagine how astonishingly fast it would be, even just because of the speed of typing... ;)
Programming is formalized per se, but the abstract global intentions of the programmers are not yet formalized.
The fact that I'm writing that I "finally have to start doing something real..." displays that I probably had preferred to theorize and "feel" intelligence and stuff, than sit and code... My general theories about what is mind were expressed in metaphysics philosophy works, when I was still a teenager, but as you mention, too general things are not too practical either. For real systems, we need to focus in details. So, it's both: generalization and specialization...
My teenage philosophy of mind and Universe (Bulgarian):
http://eim.toshuniverse.com/razum
...
I believe people who combine scientific talents (maths, physics, engineering) and humanity and artistic talents (philosophy, creative writing in all forms, photography and movie making) would reach understanding and implementation of intelligence.
Their left and right hemispheres make use of each other and they are able to analyze better their creativity decisions. These individuals have wider range of sensory inputs and wider possibilities to make generalizations and connections between diverse concepts.
http://artificial-mind.blogspot.com/2008/04/twenkid.html (see "Younak")
Hi Todor!
I speak Russian, & can make out most of your Bulgarian, but it does take an effort. I did notice your idea about "educational tests", but it's quantitative only in superficial sense, - it requires rating by an external "black box" intelligence.
If you agree that intelligence is an ability to predict, then it must quantify predictive value of individual variables, then patterns, then algorithms, on its own. That's what I mean by a quantitative criterion, in your words: teaching the system to know what it wants to achieve. Having a low level selection criterion is the only way an intelligence can evolve.
If you agree that language will develop as a natural consequence of high-level pattern recognition (which will translate images into words), then you shouldn't be doing NLP directly, or relying on verbal tests. Creativity doesn't need a different definition either, it's simply a deductive bias of intelligence, - you do more projection than recognition.
Art, games... sorry, I don't care much for it.
We seem to disagree on what a broad intellectual focus means: you thinks it's doing many different things well, & I think it's only being good at discovering common patterns among these things. That's not the same mindset, your version I'd call "a serial specialist" rather than a generalist.
I agree that AI can be programmed very quickly, just as soon as we know what we want it to do :).
If you want to continue, I think e-mail is more convinient: boris(dot)k(at)verizon(dot)net.
Boris.
You're right that developing intelligent system must be able to find out whether it progresses or not. Educational test is more like to prove to humans machine intelligence.
My theories were dealing with entities called "Virtual Control Units", they are the causality entities in the Virtual Universes. "Virtual universes" are modeled using such units, generally they're similar to multi-agent systems. Indeed, I have to refresh and rewrite my theories in English...
About the measure of achieving the goal - I've been thinking about a something like "pleasure" for the control units. The unit/entity is searching for states where it feels "more pleasure", pleasure is indication of reaching the goal, it's something in the design of the unit. The unit may do not understand the goal, but can feel that it's reached.
The main "pleasure" for CU is to control - to cause the future to be what it wants/expects it to be.
I've speculating a lot, but I can't explain it shortly.
http://artificial-mind.blogspot.com/2008/02/2004.html
...
Yes, NLP now is like doing "tricks", but sometimes the tricks are useful and there are applications. For example - fast computer-aided processing of text and dictionary-quering, information retrieval, data mining... It's useful to search quickly etc. One of the applications I want to create using NLP is a system for very fast writing.
I'm not convinced about the "serial specialist", have to reflect about that. Because if you do many things well, you know the patterns in them, and you inevitably find analogies and common patterns. Also, because being able to do all these usually is indication of something unusual in your brain, i.e. "intelligence nucleus" which has deeper pattern-matching. For example in creativity. I've written one short article about that 5 years ago: http://eim.toshuniverse.com/eim22n/eim23/emil04052003.htm
There I use a bit different terminology - "algorithm recognition" for pattern recognition, and about the "mind nucleus" - an entity which is able to grasp algorithms of how to do things.
I do want to continue this interesting conversation, but I'll have a little break. Maybe I'll read your articles; and refresh mine too...
OK, via email.
Tosh
Note: that term "algorithm" in the end of my comment could be called also "scenario".
Post a Comment