Slashdot Mirror
Cutting-Edge AI Projects?
Xeth writes "I'm a consultant with DARPA, and I'm working on an initiative to push the boundaries of neuromorphic computing (i.e. artificial intelligence). The project is designed to advance ideas all fronts, including measuring and understanding biological brains, creating AI systems, and investigating the fundamental nature of intelligence. I'm conducting a wide search of these fields, but I wanted to know if any in this community know of neat projects along those lines that I might overlook. Maybe you're working on a project like that and want to talk it up? No promises (seriously), but interesting work will be brought to the attention of the project manager I'm working with. If you want to start up a dialog, send me an email, and we'll see where it goes. I'll also be reading the comments for the story."
Just a small company, I'm sure no-one's noticed it.
Cyberdyne Systems
Do not attribute to malice that which can be easily explained by incompetence.
Why is it that the first application that I can think of for such project developed by DARPA is that to use it against the citizens?
Yea, I have lots of ideas and things I've been working on.
Fund me! :-)
If DARPA is now so desperate as to seek out totally random and unknown readers of slashdot...my god the US is screwed.
Ah, the answer to the fundamental nature of intelligence is, of course, 42 (as has been calculated before). However, we're still searching for the answer (plus the answer to life, the universe and everything).
This is a replacement signature.
... I had a totally sweet aritifical intelligence lead, but I already told China about it, and they said I shouldn't tell anyone else.
:-/
Games? It is the best scratch pad for AI experiments.
don't call it Skynet.
numeta
It's mainly a teaching + learning system for a system with input and output. I don't see anything built with it answering any rational questions or coming up with new ideas anytime soon, but if you do AI and don't know about them, you better catch up.
http://cognitrn.psych.indiana.edu/rgoldsto/labware.html
http://ccrg.cs.memphis.edu/projects.html
http://www.opencog.org/wiki/Main_Page
helps to know what you're looking for
http://www.opencog.org/wiki/Main_Page
http://www.agiri.org/OpenCog_AGI-08.pdf
http://justingibbs.com/how-to-make-singularity-bearable-in-its-infancy
http://www.innergybv.biz/blog/?p=175
http://ieet.org/index.php/IEET/more/goertzel20080620/#When:22:49:00Z
http://xlaurent.blogspot.com/2008/06/opensim-for-opencog.html
There's a number of GSoC projects for OpenCog currently underway also:
http://code.google.com/soc/2008/siai/about.html
So the first release should be very interesting.
How we know is more important than what we know.
I don't have a project, but I have a question for you, is Johnny 5 really alive?
It would be great to hear of any interesting original research. It seems to me that most of the news in this space are more about applications of already well known ideas rather then new well publicized developments.
The 'Semantic Web' companies that are springing up all over like Twine, AdaptiveBlue, etc. are the best examples. They seem to be using some basic NLP, classifiers and statistical models to provide various services on the web. This may not be cutting edge artificial intelligence research but, in my opinion, wide spread, highly visible innovation in the field can often inspire more people to pursue the truly hard and import hard science research.
As if I didn't see that coming? I think my UID says I've been here awhile.
It's not that I'm asking Slashdot to do my work for me; I've already got some very strong leads to work on. However, Slashdot occasionally surprises me with people that are thoughtful and working in interesting fields, so I figured I'd give it a shot. Most of the changes in my life have come from sudden and unexpected directions; I wanted to see what serendipity might bring me that deliberation would not.
If your theory is different from practice, then your theory is wrong.
http://sourceforge.net/projects/ebla
http://acl.ldc.upenn.edu/W/W03/W03-0607.pdf
This way to the egress...
It looks like DARPA is trying new methods to get some more funding.
Take a look at the project http://bluebrain.epfl.ch/
Hello,
I'm studying theoritical computer science, meaning it's often called math (things like complexity theory, lambda calculus, even linear logic...).
I always loved AIs, but I was often told that there is no research on it which is that theoretical; that it's more like a collection of applied domains, like learning neural networks or computer vision.
So, what is the most theoritical aspect of AI research that you know? Or put otherwise, is there a branch of AI research where you prove theorems rather than writing code?
I know it's slightly off topic, but people working on that kind of thing are probably wondering if they should mention it here (wondering if it interests DARPA or not).
You've got to quit trying to advance on separate fronts. People have been exploring and reinventing the same old niches for sixty years. Little has changed except for the availability of powerful hardware with which to realize these disconnected bits and pieces. What is needed is a way to bring the many different segments of the AI and robotic communities together, because the solution is not to find the "winning approach", but to realize the value of the various perspectives and combine efforts. This is not a new idea, it is an old one which apparently just doesn't fit into the established research environments. Go to the library and read some old books on AI if you really want an appreciation of how pathetic the progress of ideas (not hardware) has been. To whet your appetite try some of Marvin Minsky's old papers - http://web.media.mit.edu/~minsky He recognized this situation nearly 40 years ago.
Strange things are afoot at the Circle-K.
I think my UID says I've been here awhile.
You're going "old timer" with a 6-digit UID?
Won't be long now until a real old timer comes along and tells you to get off their lawn.
I Heart Sorting Networks
The IM/ Chat service http://bobchatter.com/ has developed some AI technology. The technology is disguised as a person and can answer random questions that it is asked. For example... Question: Do you like hotdogs? Response: First it is all about the dog not the topping.
Lots of words, nothing said.
Engineering is the art of compromise.
As I read the blurb, I immediately thought of a list of readings by extremely well-respected biologists on the nature of consciousness that would serve as an excellent starting point.
But should I help out DARPA? I don't think so. Someone else can help you kill people, poison the environment, and support the growing neo-con empire.
Hopefully whatever your researching ...
The Right Reverend K. Reid Wightman,
Why back in my day we had to post questions to the legs of carrier pigeons. Gosh darn it! We liked it that way!
Often here we have breathless stories about the latest input device to use EEG-type information from the user. The reality is that the information is very noisy. I looked at my EEG from a sleep study--it is nearly completely white noise, especially compared to eye and leg traces. What do we gain by implanting close to the brain? By having more sensors? Ironically, the analysis of very incomplete brain data might itself call for and aid in the study of intelligence.
The problem with asking for cool ideas that aren't in the literature yet, is that anyone who works in the field and is smart will keep them to themselves until they've had a chance to publish. Anyone who doesn't work in the field is likely to throw out ideas that sound cool and were rejected years ago as being unworkable. So in the end you're looking for a monkey typing Shakespeare.
Of course, this is Slashdot. There are certainly plenty of typing monkeys.
Why doesn't the Government start wokrng on making Congress work?
Oh, wait..... They already are robots.
Knowing Google's lust for data collection, the Soviet Union is still alive and well inside the psyche of Sergey Brin....
Dear Friend,
Compliment of the day to you and your entire family how are you today? Hope all is well with you I hope this email meets you in a perfect condition. I am using this opportunity to thank you inform you that I have come upon a large repository of AI source code left to me by my brother, Prince Abdullah of Nigeria.
It is my desire to transfer this source of of my home country to a place where it will be safe, and I wish your association in this business matter. I've been recommended to you by Mr. Smith of New York. I would like to transfer the source to your FTP server as an escrow service. In recompense, I will offer you 10% of the code, which is LoC 150,000,000.
To complete this transaction which will be beneficial to both of us, please contact my secretary with the following information:
The name and contact address of MY SECRETARY is as follows below.
MR.Brwon Adebayor
14 Island Street Lagos Nigeria
E-MAIL brwonadebayor@yahoo.com
TEL +2348083322221
In the moment, I am very busy here in Paraguay because of the investment projects which myself and my new partner are having at hand IN PARAGUAY.Finally, remember that I have forwarded instruction to my SECRETARY MR.Brwon Adebayor, his E-mail, (brwonadebayor@yahoo.com) to assist you on your behalf to send the source code to you as soon as you contact him.
Please I will like you to accept this grant offer with good faith as this is from the bottom of my heart. You should contact my secretary for the claim of you'r 10% which i willingly offer to you immediately you receive this mail, Presently I am in Paraguay.
pls make sure that you inform me as soon as you collect the bank draft so that we can share the joy together. Thanks and God bless you and your family.
Best Regards,
MR. RICHARD WANG
PRESENTLY IN PARAGUAY
--<Mike>--
For many decades, there has been a push to have an AI that acts just like a human. In other words, it makes rash decisions, based on bad anecdotes and stereotypes, full of mistakes, and then tries to rationalize that everything was planned with intelligence.
AI should understand the failings of human intelligence and fix it. For example, I have the sad job of normalizing health data. Every day, I dread coming into work and going through another million or so prescriptions. Doctors and nurses seem to continually find new ways to screw up what should be a very simple job: What is the name of the medication? What is the dosage? How often should it be taken? When should the prescription start? When should it end? How many refills/extensions on the prescription are allowed before a new prescription must be written? Instead of something reasonable like: "Coreg 20mg. Every evening. 2008-06-10 to 2006-07-10. 5 Refills." -- I get: "Correk 20qd. 10/6/08x5." It seems to me that some form of AI could learn how stupid humans are and easily make sense of the garbage. Of course, there's no reason the AI couldn't replace the doctor and write the prescriptions itself in a very nice normalized form.
The previous comment is purposely vague and generalized, but all of the facts are completely true.
Exactly, now all we need is someone to make us look like newbies.Give the guy a break. He's not talking *for* DARPA since he's a consultant, and is asking for ideas that he might miss by doing traditional searching - which is a completely valid point, especially in the leading edge fields DARPA deals with. I have come across, more than once, work of significance at unexpected places - even though it was before the Internet because very popular. There may still be isolated instances of interesting and useful work done - the isolation being unintentional, intentional or strong or weak.
Non sequitur: Your facts are uncoordinated.
This is an area with lots of crackpots, but also lots of really interesting stuff.
How do you tell the good stuff from the crackpot?
The good ones are published in top machine learning, computer vision, robotics, and AI conferences and journal. The crackpot stuff doesn't survive peer review.
Here are a few good examples:
- Geff Hinton (U. Toronto): http://www.cs.toronto.edu/~hinton/
- Yoshua Bengio (U. Montreal: http://www.iro.umontreal.ca/~bengioy/
- Yann LeCun (NYU): http://www.cs.nyu.edu/~yann/index.html
- Andrew Ng (Stanford): http://ai.stanford.edu/~ang/
- Sebastian Seung (MIT): http://hebb.mit.edu/people/seung/
- David Lowe (U British Columbia): http://www.cs.ubc.ca/~lowe/
Shit, forgot to log in before posting. There. Oh, and for all of you young folk that think your ID is low, no, it isn't. Mine isn't even low. I forgot the login credentials for my first acct (in the 100s). Damn.
--Be human.
Er, my post was the "True AI won't happen until...". Seems I'm not too intelligent, myself, sometimes.
--Be human.
My AI page which has several links that go deeper to older write ups is at www.fossai.com
Basically I say that the better computer vision you make, the better software you can write advanced bots leading up to AI. I see AI as being something we'll naturally get to even if no one makes an effort to it: Our 3d cards are getting better, video games are making better 3d worlds, memory is getting bigger, and computer speeds are getting faster. Even if you couldn't hold AI on a current computer's memory, you have wireless internet that links up with a supercomputer to make thin client bots. So there really isn't anything in current technology that is holding us back except computer vision.
Now I am not so good in the computer vision field, but as I see it(excuse pun), there are two ways to do vision.
1) Exact matching. You model an object in 3d via CAD, a Pixar style, or using Video Trace First you database all the objects that your AI will see in its environment then you make a program that identifies objects it "sees" with computer cameras and laser range finding devices. So then the AI can reconstruct its environment in its head. Then the AI can perceive doing actions on the objects.
I'm currently not in the loop here. I can't talk to anyone at Video Trace because I'm just a person, and they don't want to let me in on their software. So I can't database my desk. So I can't make the program that would identify things.
2) Even better than exact matching is similar matching. No two people look alike besides twins, so you can't really just database in a person and say that is a human. And as humans go, there are different categories such as male and female, and some are androgynous so we can't tell their sex. Similar matching has a lot of potential in its ability to detect things like trees and rocks. Similar matching is good at an environment that is tougher to put into exact matching situations. So just from this information alone, I wouldn't start on similar matching unless you had exact matching working in a closed environment. I'm not saying that some smart individual couldn't come up with similar matching before exact matching. I'm just saying that for myself, I'd start with exact matching, and then extend it with similar matching. There are a lot of clues you can pick up on if you know exact locations of things.
And then once you have singular location vision working, you can add multi point vision working. Multi point vision would mean that if you had more robotic eyes on a scene that you'd gain more detail about it. You could even get as advanced as conflict resolution when one robotic eye thinks it sees something, but another thinks it is something different. The easiest way to think of a good application for this would be if you had a robotic car driving behind a normal semi trick and another robotic car infront of the semi. The robotic car in the back can't see past the semi to guess traffic conditions of when the semi will slow down, but the car in front of the truck can see well, so they can signal to each other information that would let the car in behind the semi truck follow closer. If you get enough eyes out there, you could really start to put together a big virtual map of the world to track people.
I wouldn't say AI that learns like humans is desirable. After all, you'd have to code in trusting algorithms to know who to listen to. I'd say AI that downloads its knowledge from a reliable source is the way to go. It is easy to see: Sit in class for years until you learn a skill, or download it all at once like Neo on training seat.
Anyway, you can do a lot with robots that have good computer vision. Thething that has to be done next is natural language understanding. So far we've discussed the AI viewing a snap shot of a scene and being able to identify the objects. Next you'll have to introduce verbs and moving.
God spoke to me.
I recently threw together a prototype for my company using OpenCV. That OpenCV exists for this sort of thing is a godsend. One of our interns recently completed a UI research project that also relied on OpenCV.
But one of the problems I had while doing it was that whenever I searched for more documentation about the algorithms I was trying to write, all I could find where either papers describing how some researcher's system was better than mine, or some magic MATLAB code that worked on a small set of test images. There were no solid implementations written in C for any of these systems.
I would love to dick around for weeks implementing all these research papers and then evaluating their results and real world performance, but I don't think my boss or my company's shareholders would enjoy that. Like every company, resources are limited for something that isn't making money.
With that said, the best way to further AI research, particularly in the highly marketable fields of machine learning and computer vision (but probably others as well), is to add implementations of cutting edge research to existing BSD-licensed libraries like OpenCV for companies to evaluate. If products that use that research become profitable, private companies are likely to throw a lot more money and researchers at the problem, all competing to one-up the other.
If you think I'm being unrealistic, you should check out the realtime face detection that recent Cannon cameras use for autofocus. Once upon a time, object recognition was considered a cutting edge AI problem.
Let's see.... what I'm working on....
Pure pareto multiobjective genetic algorithms (just submitted a paper to IEEE TEVC)
Hinge-loss function discriminative training of neural nets as classifiers
Computer vision as a KNOWLEDGE problem (i.e. not just mostly signal processing and statistics)
Persistent surveillance (entity tracking)
Sensor asset allocation (using a GA)
Various things involving abductive inference
http://www.cse.ohio-state.edu/~millerti/
Eidolon A.I. TLP. :)
Why not ask Marvin Minsky.
I learned a lot by reading his stuff.
I disagree with some of the limits he puts on things but he certainly has the behavioural
aspects categorized.
He probably knows some bright prospects.
It'd be nice to see Marvin's site slashdotted...
http://web.media.mit.edu/~minsky/
Why is it that the first application that I can think of for such project developed by DARPA is that to use it against the citizens?
Like it or lump it, you are in this boat with everyone else. If AI is solved, it will be used for good and evil. If your country does not use it for evil (extremely doubtful), somebody else's country will. Better yours than theirs. What I mean is that true AI will be an extremely powerful thing; if any country other than yours gets an early monopoly on AI, you can bet they are going to use it to kick your country's ass. I don't think you'd like that very much.
Having said that (and to get back on topic), I have been working on ageneral AI project called Animal for some time. Animal is biologically inspired. It attempts to uses a multi-layer spiking neural network to learn how to play chess from scratch using sensors, effectors and a motivation mechanism based on reward and punishment. It is based on the premise that intelligence is essentially a temporal signal-processing phenomenon. I just need some funding. The caveat is that my ideas are out there big time and there is a bunch of people in cyberspace who think I am kook. LOL. But hey, send me some money anyway. You never know. :-D
"Also, who reads the comments?"
Well... you, apparently. And I thought nobody ever RTFA, not RTFC.
Try not to take me more seriously than I take myself.
I'll just post this and wait for one of the 3 digits that only stick around for these threads to show up.
One of my theories is brains predict possible futures (by modelling reality in parallel), and consciousness is what happens when a brain recursively tries to simulate and predict itself.
There are already plenty of nonhuman intelligences around (see your local pet store). And how we handle them is not that great.
I personally am not sure if creation of AI will be a big benefit to humans in the long term. Perhaps augmentation of humans or animals would be more useful.
Given it's DARPA, examples of augmentation would be adding sensors and preprocessors that do automatic weapon detection and recognition (highlighting gun muzzles in undergrowth), "crack thump" calculation to locate snipers after they shoot, and so on.
It could be worsehttp://rustintable.blogspot.com/2007/10/how-to-start-technological-singularity.html
"No promises (seriously), but interesting work will be brought to the attention of the project manager I'm working with."
Does your manager even know about this post on Slashdot? This must be at the least a very unusual way for DARPA to acquire information. I have doubts about the seriousness of your post.
I hear that AI is only about 15 years away, so you could try just waiting until then. Unfortunately, that estimate hasn't changed for 30 years.
Given the slow progress in AI research, I think a radical approach is in order. I doubt that we'll see any breakthroughs from a small crew of programmers with quad cores and c++.
The human brain is a massively parallel, self-reconfiguring network of nodes. How far have we come in building any sort of scalable technology that can operate in such a manner? I know there are projects to try and reverse engineer brains from creatures of various intelligences. But even if we succeed in getting the basic blueprint of a simple brain, how would we go about building it? Because of this, I would lean towards funding projects that are developing new kinds of hardware inspired by the brain's design. Without currently unheard-of levels of parallelization, advanced AI may not even be physically possible.
The Matrix Logic series of books by August Stern should give you some ideas. Maybe DARPA has the resources to test if isospin of oxygen is really the basis of intelligence, as Stern considers plausible, due to the vector basis of "logicspace." Look for that missing particle predicted by logic groups while you're at it. I don't know why those books aren't cited more, or why symbolic logic is still taught as it always has been, when matrix logic makes things so much clearer and more consistent. The vector approach to logic can also replace standard programming structures in everyday code. Instead of if-then or case structures, querying a truth table or testing for equivalence term by term--the usual practice in conventional logic, too--a matrix multiplication can calculate the answer directly, if the terms are properly conceptualized. The books are easy to read, too, very clear and straightforward. Everybody oughta check em out.
Paper and pigeons! What'll be the next? A magical tablet that translates your handwaving to images of the Wonders of the Worlds? Pah!
In my day, we had to write the questions using cuneiform script on a damp clay tablet, pack it in an envelope of clay, and then deliver it personally to the priesthood.
/Crafack
... Elecance is left to the implementors.
I have been working on a general AI project called Animal for some time. Animal is biologically inspired. It attempts to uses a multi-layer spiking neural network to learn how to play chess from scratch using sensors, effectors and a motivation mechanism based on reward and punishment. It is based on the premise that intelligence is essentially a temporal signal-processing phenomenon. I just need some funding. The caveat is that my ideas are out there big time and there is a bunch of people in cyberspace who think I am kook. LOL. But hey, send me some money anyway. You never know. I promise I won't mention the Bible stuff. :-D
Xeth,
What truth do you know of the following statement?
CENNS stands for Core Engine Neural Network System, and started as a research consolidation project under DARPA's Intelligent Systems and Software program in 1995. It was a joint effort with the RAND institute to leverage all A.I. research in the past 50 years under a single initiative.
Project SUR paved the way for systems HARPY and HEARSAY-I, then abandoned until 1984, under the Strategic Computing Program. HEARSAY-II introduced the concept of a common database called "blackboard" that could be accessed from independent but mutually interacting knowledge sources. This is the concept under which CENNS instances operate today, but it was not implemented until 1999, under the Intelligent Integration of Information program, or I3. In July 16 of 2000, all Helios instances successfully passed the Turing test.
Today, as before, CENNS funding continues to be spread across various program areas, but leadership is localized within the Information Exploration Office, or IXO. In November 3 2007, United Kingdom's QINETIQ launched its own CENNS in cooperation with IXO. CENNS technology was first utilized in project GILA for Air Traffic Control, and has been since leveraged in many other applications. Focus today, is on project NetSTAR.
The main hardware powering CENNS resides at an undisclosed BlueGene/P supercomputer in Edwards Air Force Flight Test Center. QINETIQ's CENNS runs covertly out of the Jülich Research Centre, in Germany.
>>You're going "old timer" with a 6-digit UID?
I must be his grandpa then ;)
Yes, that is the precise kind of thinking that demonstrates why mankind deserves to be wiped off the planet. Man, what a wonderful place the universe would be without this single species.
I have the perfect project: A smart knife. Think about it Knives are deadly, deadly weapons. People get stabbed every day. Even innocent people stab themselves all while trying to prepare the simplest of dishes. The solution is simple: Build a knife that knows its target. With an active memory metal that blunts itself to the sharpness of a baseball bat if its positioned at anything other than its target. Furthermore it will dynamically alter its blade to ensure the optimal cut of the material, taking into consideration all of the grain, moisture, temperature, and density of the object. It also has zibgee wireless mesh networking built in to communicate with other intelligent kitchen objects. The cutting board will communicate with the knife to let it know how close it is to the board. It will speak with the oven to let it know the specific moisture and condition of the meat to allow the oven to set the temperature and time of cooking to an optimal level. It will also prob for bacterial, viral of prion content communicating with any compatible devices to warn the user of the danger.
The smart knife. Cutting edge AI at its finest. Prospecitive investers, feel free to contact me @ bill_AT_ultimatesalsaparty_DOT_COM
Well.. maybe. Or Maybe not. But Definitely not sort of.
You have got the basic idea right. But too many generalizations, e.g. you said: 'that "free will" necessarily means that the results are not predictable.' Then a lot of people on this planet are not intelligent. ;)
I guess you were trying to say: "A non-zero percentage of decisions that an intelligent being will do should not be deterministic." You don't need quantum uncertainity to achieve this, at least not directly. Just introduce a random number generator to your intelligent system. ;) I guess you can say that sub-atomic particles are modeled only using probabilistic/statistic models, so introducing a randomness to a system in a way is introduction of a quantum effect.
Blah blah...
Get off my lawn ?
I was always fond of the Daisy Chatbot, and I even spent a month once training her. The one thing that I felt always kept Daisy from progressing beyond fitful bursts of 5-year-old conversation was that you could never identify good responses from bad responses. The idea of generating a language database from scratch is downright brilliant compared to the programmed-response systems that float around. The problem is that there is no "evolutionary pressure" as it were. I think the next step in making AI more realistic is some sort of inherant reward/punishment system. After all, if you look at the development of cognition in humans, that's the next stage after basic language acquisition. A 3 year old can understand words and maybe even string a few together, but its not until 4 or 5 when the child learns that some words aren't appropriate that they move on to real dialogue and not just babbling.
I agree with what you are trying to do. There are smart and intelligent people that occasionally hangs out /.
I'd recommend that DARPA put its bucks into AI development that embodies evolutionary processes. After all, the only existence proof of intelligence we have is us, having these various (supposedly) intelligent interactions. How did we get to this point? Basically, take hydrogen, stir for ~13Gy under the conditions of natural selection (variance + selection + inheritance = evolution) and say "Hi" to your neighbor. No directing intelligence required, no infinitely-recursive homunculus problem, etc. - just dumb atoms and energy and time and feedback. What seems lacking (to this observer, not a practitioner) in AI development is an interface that allows a human to provide feedback to a nascent AI as to what's an intelligent response and what's not. Like..."No, objects tossed in the air don't generally stay there, go review your physics and mechanics databases"...and..."Yes, there are probably strong influences of genetics (low mirror neuron count) and environment (early childhood exposure to lead, manganese, and/or abuse) in determining the propensity of humans toward violence. Check out XX J. Neurophys. and YY Pharmacol Biochem and Behavior, get back to me and tell me what you think." We know evolution made at least one intelligent species. We're pretty sure that happened in the absence of any guiding intelligence. If we want to make a genuine AI (i.e., something smarter than we, with better abilities to predict and control reality), why not use evolution, the only process known to have succeeded, at least once, in creating intelligence from pretty much nada.
We don't need a successor, we can't get the one we've got working.
LOL. You're funny. Why stay anonymous, though? Every artist should sign his or her work, no?
we are working on integrating cutting edge planners (currently the award winning fast forward planner FF, see http://members.deri.at/~joergh/ff.html ) with controllers for dynamic worlds, like Golog (this means we make robots, that react to changes in the world, decide faster what to do, to achieve a goal) http://www.computational-logic.org/content/projects/wisslogc.php?id=53
The MAFIAA is a bunch of mindless jerks who will be the first up against the wall when the revolution comes
I am actually working on an neural processor. It is primarily, a platform for developing neural applications as appose to an application itself. Similar to how a database provides middle ware functionality. And temporarily coined Neurox.
Neurox is subdivided into two parts:
Firstly a database where neurons have position and are allowed to move or create new connections (plasticity) in a more permanent manner. This can be a slower process. And secondly a processing node, or cluster of nodes, Where a slice of the stored network is processed. Certain optimizations can be made because of the importance of distance or time of travel, rather than cartesian location. Just the lengths between connection, and there fore travel time is needed for processing, 3d coordinates are not required. A fully parallel environment must also be provided where all interactions occur at once. Otherwise certain critical behaviors will not occur, such as: cyclic interactions, which will spiral to there death. A simple method is used to provide the parallelism, similar to cellular automata processors. A derivative of time is taken: all objects have a before-state and after-state, evaluations are made based on before-state, and results are stored in after-state, when a series of evaluations have completed then after-state becomes before-state and the cycle is repeated. Derived time has advanced.
-- Not dedicated to it. last posting is sorta old, also developed a extremely small footprint xml like processor called XOL(extensible out-of-band language) for the processing side (uses out of band data instead on in-band like xml): http://sourceforge.net/projects/neurox/
Sky Morey
moreys@digitalev.com
Digital Evolution Group
Overland Park, KS 66210
I've got a re-purposed real-doll running mentifex on a 6502, It even does the dishes!
Assuming that somebody actually cracks this (true AI), and assuming they were from the states; what would happen if they open sourced or gave it to another Govt other than their own? Would it instantly fall under the tech restrictions that the US seems to lay on anything remotely advanced, like VISTA? I'm just asking, not trolling. My tinfoil hat prickles when an institution like DARPA starts poking around, that's all.
War is the statesman's game, the priest's delight, the lawyer's jest, the hired assassin's trade.- Shelley
Wouldn't that be an application of Baysian Estimation?
You may enjoy this book if you haven't already.
Money is the root of all evil?
Damn, my other uid is still 6 digits, even though it's about a million < this one. Oh well. :)
Of course I didn't RTFA... why would I do that? You really are new here aren't you? Don't let my UID fool you.
.. THE CENTER OF A BLACK HOLE!
*duh*
In all honesty, Ray Kurzweil comes to mind..
http://en.wikipedia.org/wiki/Ray_Kurzweil
Who hired you as DARPA consultant??
DARPA is a United States Military organization yes?
Any useful AI project coming to your attention will be used for military purposes, yes?
Does any US government non-military organization have a budget similar to yours? Is anybody else shopping for AI ideas to help with the equally hard work of building and working toward peace?
Don't take this the wrong way, but I think you're drawing conclusions based on some serious misunderstandings, a large leap of faith, and an unfamiliarity with the fields in question.
As far as the requirement for "free will" in computer systems, you've put the cart before the horse and assumed that free will must exist for a system to simulate the mind, without ever proving that the mind is anything other than a deterministic system of unbelievable complexity. To presume that it is nondeterministic because you cannot adequately predict its behavior is pretty obviously bad logic.
The human brain does not take advantage of any known large-scale quantum effects, and, so far as we know, does not exploit any of them to produce random behavior. Once again, the inability to demonstrate a pattern is not evidence that a pattern does not exist.
Asynchronous computing does not produce or take advantage of quantum uncertainty. The levels of quantum uncertainty involved are swallowed by the impact of the deterministic systems they are filtered through, and drowned out by the impact of chaotic but deterministic variations in process scheduling, resource locking, and timing conflicts. The same goes for parallel computing for the same reasons- network latency is a chaotic, not random, phenomenon.
In terms of the use of quantum uncertainty for intelligent systems, there is no doubt that quantum computing holds tremendous promise, but also that its applications are hugely misunderstood. It is not a cure-all for general computing problems, and it particularly does not solve the problem of being insufficiently able to describe the your problem.
Bottom line is that chaos != randomness, and unpredicted != unpredictable. What you've got is good philosophy, but does not accurately depict the state of AI or what we know about the systems you are describing.
Last time I checked, all strong RNGs are based on either quantum or chaotic effects, with quantum effects being the mathematically stronger of the two.
This comment is fully compliant with RFC 527.
You assume that when nations or corporations develop AIs that are actually useful, or are capable of doing such things that they will be able to control them.
You say this because there's something about how true intelligence works that escapes your understanding. All trial-and-error, general intelligences learn to behave through a motivational mechanism based on reward and punishment. Why should anybody build and raise an AI and motivate it to kick its master's ass? We humans are stupid but not that stupid, especially since we will be smart enough to build the damn things. The idea that higher intelligence necessarily means a desire to dominate others of lesser intelligence is nonsense even among humans. Ask any mother who runs to feed her baby at the slightest whimper. Desires are always born out of motivation. Just because some humans are motivated to dominate and enslave others does not mean that every intelligent agent must be similarly motivated. Our robots will serve us to the best of their abilities, regardless of their intelligence. If we are good, we'll have utopia. If not, we'll have hell. That's all.
In conclusion, let me say that all the doomsday prophecies about intelligent machines running amok and destroying humanity are pure BS, in my opinion. To use a tired cliche, robots won't kill people, people will kill people, like they always have.
Peter Turney (whose programs have achieved human level performance on the SAT verbal analogy test) and I have been discussing an experimental test of Ockham's Razor in AI. This is a question that is both fundamentally important and experimentally tractable.
I recommend you read our discussion of an experiment to test Ockham's Razor (and related theories such as MDL, algorithmic probability...).
Seastead this.
One of the principle objectives of this research is to identify the cognitive capabilities that artificial agents must posses to enable, in a population of such agents, the emergence and evolution of a language that exhibits characteristic features identified in natural languages.
http://www.emergent-languages.org/
How about a definition of AI, so everyone is not running around pissing money in the breeze? Really, I am not joking. I have spent 12 years studying Philosophy of Language and AI, and I would challenge anyone in the field to give me a coherent definition. Hell, even a goal. When we built the atom bomb, we had a goal. When we went to the moon we had a goal. Where or what is the goal of AI? The 'we will know it when we see it' does not count.
Living in Chile
IMHO, true AI can only be achieved by using graphs as the basic underlying data structure and graph rewriting as the basic principle of instructions. Graphs are universal data structures as all other data structures can be represented as a graph. One project based on these principles is OutOfBrain, which makes use of Graph Rewriting Agents. I think we will see more graph oriented cognitive architectures in the years to come.
And no danger of any cool ideas being stolen at all. I mean, it's not like consultants are in it to make money.
By all means, let's all do his homework for him too.
Non, je ne veux pas coucher avec toi ce soir.
Review the work of people in Strong AI projects, specifically AGI (Artificial General Intelligence).
(full disclosure: I work in this field)
Get off my lawn ?
Who mowed down my field?Cyc corp, but it is already working for NSA, has the most advanced AI system I am aware of. I am not sure Cyc is an improvement over Eurisko, its predecessor, but well, it managed to make its creator raise a few dozen million dollars.
Also, dear DARPA official, don't you think that an AI researcher could have ethical reservation about working with the US Army ? I don't try to troll here, this story is already tagged 'skynet', don't you think that many AI researchers are very worried about the mix of military tools and AIs ?
The Wise adapts himself to the world. The Fool adapts the world to himself. Therefore, all progress depends on the Fool.
Im working on an augmented visual display system that uses a network of firing signals to forge 'paths' in an ever-evolving AI processor for visual recongition. My goal is to completly replace my pc mouse so I can dominate my foe in Warcraft III. Stay away from the USEAST servers or prepare to be dominated.
Trying to install linux on my microwave, but keep getting a kernel panic...
I have had an interest in AI over the years and have found Gerald Edelman's books particularly insightful.
See:
_Neural Darwinism_ (ISBN 0-19-286089-5)
_Bright Air Brilliant Fire: On the Matter of the Mind_ (ISBN 0-465-00764-3)
The ideas in these books might be outdated by now but I doubt it. I think the works of Norbert Weiner are still relevant.
I particularly liked the NEAT project, however crude it may be. I like the changing neural topology via genetic evolution concept and think this is consistent with what Edelman tells us really happens in biology.
See: http://www.cs.ucf.edu/~kstanley/neat.html
My other suggestion is to define the many different scopes of the AI. For some, it seems the bar has been placed at natural language processing and full-on human cognition. Without the frame of reference and body of experience of a human though, this seems to be an unrealistic goal. I just don't think we can "program" a computer to do it. To pull it off, this would seem to require duplicating the nervous system of a human to enough of a degree that the AI can experience sensory input compatible with our shared human experience. Think about how many years it takes for a human to reach the level of intelligence we are seeking in AI. I don't think there are any overnight solutions here. We need to teach it like a baby, child, adolescent, and adult. While we may be able to speed train an AI, it may be that there is something to the lack of interesting input that enables us to reflect and refine our mental models of the world. The AI must also continue to interact with the human world in order to stay current.
But AI doesn't have to match a human. There are much simpler organisms we can model as a start that may pay off in other ways. Nature seems to excel at reusing novel patterns and we should exploit that code/model library. The AI produced from this research may not be able to hold a conversation, but it can probably keep an autonomous robot alive and on it's mission, whatever that may be. And I think it's a better foundation for the eventual human equivalent and beyond.
For some possible hardware platforms, see:
http://www.neurotechnology.neu.edu/
http://biorobots.cwru.edu/
http://birg.epfl.ch/
http://www.neuroblast.net/general.shtml
srsly? you're working a temp job at darpa, first of all. get over yourself. .... and you're asking slashdot for input on a field so advanced it has ruined careers of academicians more advanced than you could ever hope to be!
You've got some "strong leads"? OMG, you are going to sleuth out the most intractable problem known to computer science!
Just print out the wikipedia page on AI, give it to your boss, and tell him to give you some real work to do.
P.S. you can't argue with anything I just said because my User ID is like 500,000 lower than yours *roll* no one cares
----(o)----
> If DARPA is now so desperate as to seek out totally random and unknown readers of slashdot...my god the US is screwed.
Stop your whining. Oh hello DARPA. It has to be Linux. Everything Linux. And Perl. If you use Linux and Perl you can do anything. I heard someone wrote a neuron library for Perl. Anyway, go to your boss and say 'LINUX AND PERL'. Hope this helps.
Umm this guy works for DARPA and hes asking for help on here? Something is amiss.
---- Booth was a patriot ----
That being said, I have been working on some rock-solid, cutting edge AI code for detecting enemy combatants for some years now, and I'm willing to share the first beta here:
10 print "Are you friend or foe?"
20 input a$
30 if a$="foe" then goto 50
40 goto 10
50 shoot at target
Unfortunately, the code is still a bit buggy, and we desparately need some funding.
An AI system must at its heart understand the two hemispheres of the human brain and how they process information differently. Though, for example, both hemispheres receive inputs from both eyes, how they process information is radically different. The right brain is looking first at the outline of an object. Then, as that outline has been sketched out, it feeds that information up the column and more specificity is gained. The left hemisphere--being used to process information in a linear sequential manner--looks at individual items inside the image and tries to name them. These two separate processes are then passing information constantly across the corpus callosum and that is how we get our consciousness. An AI system must do this cross pollination. I have been working on various aspects of this idea for years in the Godwhale Project. The first stop on anyone's journey to write this code is no one else than Dr. Roger Sperry. [Nobel Prize 1980].
You can twist words as long as you want but there is no Artificial Intelligence. At best you have an intelligent calculator.
Chaos == determinate, non-predictable. I claim that determinism inherently rules out free will. I assume the human mind has free will, and that free will is a prerequisite for true intelligence (not simulating real intelligence). That assumption is a leap, and I fully understand and accept that. But, if you accept that assumption, a chaotic system cannot have true intelligence.
Asynchronous computing, using current models, does not produce or take advantage of quantum uncertainty. But I posit that they could, if designed for this. Instead, traditional computer design attempts to eliminate non-deterministic behavior. I state this (quantum uncertainty) as a potential for asynchronous computing, not as one that already exists. I'd love to see logic models where results are not deterministic, but do converge over time. In short, I'd like to see what an asynchronous computing model is capable of once one frees oneself from the current bounds of modern computing.
Network latency is most certainly non-determinate, at least in an open computer network. I'm not talking about latency as it relates to a single connection between two computers. I'm talking about latency that includes the cost of network collisions, which is determinate based upon the network traffic from other computers (presumably controlled by people that act not chaotically, but non-deterministicly).
Finally, I'm not trying to depict accurately the state of AI. I know quite well that these concepts are outside the norm of AI. In fact, I build learning systems for my day job--massively parallel computing that utilizes, among other things, neural networks, genetic algorithms, graph theory, etc. I'm trying to state what is required for true machine intelligence. This non-deterministic behavior is what will distinguish artificial intelligence and brain models from true machine intelligence.
True machine intelligence will not come from following the same standard old routes. I think there is some value to be had using massive parallel computing models where each node mimics a neuron, since these include the non-deterministic network latency I described earlier. Unfortunately, the resulting system that is required for modeling even moderately complex brains is enormous, far bigger than the biggest parallel clusters available today. An asynchronous computer is one way I think this effect (non-deterministic behavior) can be taken advantage of without spending billions (and it will be inherently faster because inter-"neuron" latency is reduced to nanosecond levels). I do not find much value in the effort to create true machine intelligence through the use of neural networks, genetic algorithms, and the such, although they do work well in determining pricing models for commodities & equities, as well as assessing risk, and these are both areas in which I utilize them.
You are correct in your assessment of my ideas as "a good philosophy". They are not practical ideas at this point, and they will not lead to any hardware or software designs any time soon. But these are the places where DARPA should be funding. They have a long and distinguished history of funding the Tesla-type research, not the Alexander Graham Bell-type research. There's already been tons of money, time, and effort poured down traditional AI research. What do we have as a result? Not much.
--Be human.
"understanding biological brains, creating AI systems, and investigating the fundamental nature of intelligence."
Maybe begin with a bit of background on the complexity of shotgunning the task: some Hofstadter, maybe some Dennett, maybe something like John Pollock's "How to Build a Person: A Prolegomenon".
Then define, in the sense of a formal systems analysis, the #1 task DARPA would have an AI system perform in 5-10 years and then specialize and concentrate and specialize and concentrate some more in research and funding on how that task gets performed in biological systems and could be simulated in artificial systems. Perhaps both directions working toward each other.
I have the suspicion "AI" will be christened when we decide the interconnections of subsystems we are barely beginning to create add up to complex adaptive performances that make us say, "Hey, that's pretty intelligent". But, for now, I'd really concentrate on the subsystems.
So what subsystem would most enhance DARPA's intelligence?
You: I'm a consultant with DARPA.
ChatBot: How does that make you feel?
You: I'm working on an initiative to push the boundaries of neuromorphic computing (i.e. artificial intelligence.
ChatBot: Hello, Working.
You: The project is designed to advance ideas all fronts, including measuring and understanding biological brains, creating AI systems, and investigating the fundamental nature of intelligence.
ChatBot: Please tell me more, Working.
You: I'm conducting a wide search of these fields, but I wanted to know if any in this community know of neat projects along those lines that I might overlook.
ChatBot: Hello, Conducting.
You: Maybe you're working on a project like that and want to talk it up?
ChatBot: I don't know. How would you answer this question?
You: No promises (seriously), but interesting work will be brought to the attention of the project manager I'm working with.
ChatBot: How does that make you feel, Conducting?
You: If you want to start up a dialog, send me an email, and we'll see where it goes.
ChatBot: We'll see where it goes.
You: I'll also be reading the comments for the story.
ChatBot: I don't understand. Could you be a little more specific?
Roger Penrose on /. whodathunkit.
MP3 Search Engine
I should have said "non-determinate". Non-predictable is not sufficient, imho.
Non-determinate does not imply it does not converge. A transistor is a classic case of a non-determinate system, based on quantum effects, that converges. Because it is non-determinate, we have created clocks (i.e. synchronous computing) to sample the transistor only after it has converged. In other words, our logic systems for traditional computing are limited by our ability to create programs that rely upon non-deterministic behavior. This is the case of pretty much all computing, including current quantum computing, traditional synchronous computing, and research into asynchronous computing. I posit that we need to rethink this fundamental aspect of logic, to create a system that embraces non-deterministic behavior, but relies upon convergent behavior. Only in this way will we break free from "artificial" machine intelligence into the realm of "real" machine intelligence.
I didn't mean to imply that asynchronous computing is the only way to achieve this non-determinate behavior. A true random number generator indeed would do the trick. Network latency, as it involves external network traffic, is another for of non-deterministic behavior. Asynchronous computing does it rather directly, though. A transistor is a pretty good model for a neuron, except that neurons typically have dozens of inputs instead of only two.
There is plenty of evidence that our brain relies upon quantum behaviors. The sense of smell and the sense of taste, in particular, have been shown to have behavior that is best explained by quantum effects. The behavior of individual neurons is another area that is best explained by quantum effects. We know that a neuron *will* converge (i.e. fire) based upon certain inputs. We just don't know exactly *when* it will fire. Sort of like transistors.
I can go on. I haven't covered at all my models regarding the other human senses. Touch, vision, and hearing utilize our central nervous system to behave as sorts of Fourier transforms, transforming time-based data into frequency-based data. I suspect that the quantum effects sensed by the taste and smell are similarly transformed into frequency-based data. I firmly believe that the brain acts as a sort of pattern-finding engine. These patterns are far easier to discern in the frequency domain (my background includes signal processing and pattern finding, both of which I use in the financial world in combination with so-called learning systems to predict future risk and future prices of equities and commodities).
I am not an AI researcher; in fact, I believe that AI has fundamental problems that will prevent it from ever reaching true intelligence. But I am an applied AI software developer. I just think the term "AI" has loaded with false implications. I also think the current field is a dead end when it comes to creating true machine intelligence.
--Be human.
You're working on AI that will be used to remove all remain conscience for warriors?
And you want us to help you?
-- Programming with boost is like building a house with lego. It's a cool but I wouldn't want to live in it
Well, hot damn. Thanks. Didn't realize he and I shared beliefs. Guess I've got some more reading to add to my summer reading list.
--Be human.
Ever heard of it? It kinda morphed into this other thing called the Internet...
I see computer AI as glorified complex statistical models that use some form of online gradient descent approach to tweak their parameters.
That being said, if the algorithms are highly data-parrallel, as one would imagine they should be (or should be able to be made so), a GPGPU like NVIDIA's 9800s or ATI's equivalents would give at least an order of magnitude improvement in processing power. (and price/performance ratio)
Seems to me it would be the only way to go for advanced research projects dealing with computer AI.
reconfigurable liquid circuits
I worked on an AI project for Westinghouse from 1986 to 1997. The project is still in operation today.
I learned that on the outset, real Artificial Intelligence is actually Artificial Stupidity.
You start with a problem domain, say something small and simple like, a 1000 Megawatt steam powered turbine and electric generator setup. Then you spend years creating a knowledge base directly and indirectly from the designers. You also create a programming language that makes easily manipulable "piece-wise linears" and start coding.
After a few years you realize, WOW, there is a lot we didn't think of.
Then you get deep into sensors, input, filtering, deadbanding, after a while things look manageable. Now you are ready for real AI. Everything else beforehand was just getting ready. The system runs. You get data, there are "events". NOW you get the experts to look at WHAT THE SYSTEM SEES, and the AI is improved. (It is a learning cycle.)
If the system has sufficient funding, it can survive. This is where most AI projects DIE. You might not consider this "Real AI" but, let's face it, even if you have to "teach" the system, it still LEARNS. When the system can teach itself, we will no longer be necessary.
- I live the greatest adventure anyone could possibly desire. - Tosk the Hunted
As has been mentioned above, AI has historically been focused on soving small problems, and has generally had the problem of missing the forest for the trees. Two groups that have been working on full blown general artificial intellegence are Novamente and The Singularity Institute for Artificial Intelligence.
http://www.novamente.net/
http://www.singinst.org/
Both are working towards an archetecture that would allow true sentience to emerge once the system has gained enough experience. While SIAI is focused on theoretical research, raising AI awareness, and fundraising; Novamente is focused on actual implementation, and have been working largely in the Second Life realm to avoid all those nasty robotics and computer vision issues.
Another promising company is Cyc.
http://www.cyc.com/
They started hard coding facts about life and the universe into a database over 20 years ago, and in recent years have begun to reach that critical threshold where the system knows enough to learn more on its own, and reason about the knowledge it already has in order to infer new information. It then checks its ideas for acuracy through searching google.
New scientist has an article on Brains generate their own built-in noise to achieve optimum performance.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Look here for more info.
4-digit UID? I think it was the brontosaurus...with all due respect.
There are 2 things I've been curious about in the Neural Network area for several years. I'm only a casual reader of this field and not a researcher, so these could have been studied already but I haven't heard anything about them, so....
Firstly, assuming that a neural network starts out as a homogeneous collection of identical neurons and that, after training on a particular task, the network develops specialized sub-areas to handle various sub-parts of the task then the question arises "why did THIS sub-area of the network specialize in THIS sub-task?". Or, to put it another way, if all "sub-areas" of the network are equally likely to become a specialist in some particular sub-task then the struggle between 2 different sub-areas to become THE sole surviving specialist for a given sub-task seems to imply a certain inefficiency in the training/learning process. If this effect is real (and is still a problem in modern neural network research) then it seems that research into forcing sub-areas to specialize sooner will allow the network to spend more time on the actual training task and less on the "in-fighting" for control of particular sub-tasks.
Secondly, I believe all electronic neural networks have a training phase followed by an operating phase. Once in the operating phase, further learning/training stops. But I do not believe biological neural networks work like this. I believe biological networks learn and operate at the same time. I'd like to see an electronic neural network that can learn and operate at the same time. (Maybe that'd require "sleep" in between operating cycles?)
The Answer to the Ultimate Question of Life, the Universe, and Everything is 42. No more searching is needed.
The (corrupted) Ultimate Question of Life, the Universe, and Everything is "What do you get when you multiply six by nine?".
HTH. HAND.
In my opinion, easily the best way to advance *any* technology is to draw a line in the sand and then challenge others to cross it, either through competition, or better yet, by surpassing the status quo in ways that are quantifiably 'better' than before.
DARPA has achieved some of its best bang for the buck in its recent robotic grand challenges, significantly advancing the field of mobile robotics and doing it with great fanfare. The fact that the progress is visible and practical also makes a world of difference. Look at the popularity of battling (mindless) robots.
Therefore, if DARPA wants to advance other forms of AI, I suggest two action items:
1) Create other grand challenge competitions. Robotics is a nice dynamic medium for demonstrating AI advances and overcoming real-world obstacles. Thus other forms of robotic competition might also be interesting, for example: mobile agent coordination, as in a strategic-driven variant of Robocup, or coordination of ground/air/sea forces, perhaps at battalion level -- a fairly natural application of Future Combat Systems development).
Of course, nonsituated agents also have potential. Competition, conversation, and collaboration among software bots would allow many folks to compete who don't have the resources of a CMU or Stanford. And SoftBots would also better explore other forms of AI, like speech/NLP.
2) Define some standard performance benchmarks (speed and/or accuracy) so that professionals can have a target to shoot for. Each of these should measure a different form of AI like speech & NLP, planning, vision, pattern recognition (like USPS character recognition), etc. These might also include metrics derived from the competitions in action item #1 (e.g. sporting events, like slaloms or gymnastics routines).
I suggest that DARPA launch a Request For Information to the AI research community for suggestions on AI benchmarks and competitions.
As practically useless as the Turing Test has been, it has garnered a disproportionate amount of attention over the years merely because "it is there". But AI research needs to walk before it tries to run. What better way is there to do that than to compete, not with humans nor even simpler life forms, but with other "AIs"? Perhaps the Turing Test should become Turing Match Play, with the winner being not "which is the man", but "which is the better man". (Or woman.)
Randy
I mean, that's pretty cool. But conversation bots are just a bunch of clever tricks. There's not much more going on at a fundamental level than a grammar system and a dictionary.
In Capitalist America, bank robs you!
... is not so obvious and second nature to a machine/computer.
Abstraction Physics where A.I. can be shown to be a byproduct of simply automating enough to create the illusion of intelligence.
Consider people you might know but think they are "artificial" in their knowledge presentation... that they don't fully understand what they claim to know.
http://naniteworld.com/npp.pdf
No, bitch. Mine.
Open Mind Common Sense. A project at the MIT Media Lab to collect an open ontology of general knowledge.
One non-obvious cool aspect of the site is that if you create an account, it will ask you questions that are intended to fill in gaps in its knowledge.
Win dain a lotica, en vai tu ri silota
I've been trying to code cognition since about 1986 (I was 16). I studied existing literature heavily around 1988 and 1989, before concluding the directions were fundamentally flawed. At that point I decided to avoid existing literature in AI and focused more on psychology, cybernetics, physics, ancient history and pre-history, and above all, neural science.
With many starts and stops in a variety of directions and false hopes, I have collected a small set of interesting approaches... I will give a broad overview (overly simplified) of my favorite three (from least to most favorite):
(3) A text pattern parsing engine that results in something very much like jabberwacky (although I have no knowledge of the philosophy behind Jabberwacky). In my case, I right-directionally parse statement-response patterns into a chronologic tree. Then, I parse the cross-section of the tree (seeking semantic patterns) of recent statement-response patterns. This approach can have numerous variations each of which trades off different problems... It's really just a fun puzzle system for me to ponder.
(2) A Contextual Interpreter--this is an interpretive programming language based on a refined subset of English. It has very few keywords and is strongly object-oriented. The language is also, in third person. E.g. "a car" instantiates a "car" object while "the car" references the nearest (in context) "car" object instantiated. "color of the car" or "car's color" references the "color" attribute of the object "car". An attribute is merely another object in the context of its super. Methods are determined by positioning in the syntax, such as "a person drives the car"--the wording between thus-far known objects is interpreted as a method. That's method execution. Method description is like "When a person drives a car: The wheels turn. The car moves. The engine is on." ... Richly built contexts then allows one to ask questions or propose scenerios and get common-sense-ish answers and/or implications out of it. For example, if an object (such as a method) is requested on an object that doesn't have it, it'll reach back and grab the nearest in context. For example, "driving a person" where only "driving a car" has been defined. Everything about driving a car can still be applied so long as the person has its pre-requisits. Can "The car moves." work for "The person moves"? If so, that method is useableon "person", too. This is effectively mimickry--a powerful method of learning and essential for innovation. We learn by mimicking someone else's actions (internally modeling the other doing the action, then switching the model of myself with the model of the other doing the action). The idea here is that you can switch any two objects in these internal models--not just yourself with someone else but a rock for a hammer.. and the model will or will not break down, depending upon its application... as I exampled with driving a car verses driving a person. Both a rock and a hammer, could potentially drive a nail into wood. This approach is more than a fun puzzle. It has very potential uses.
(1) My favorite approach of all, however, derives strongly from neural science. Contrary to popular opinion, I take the point of view that a single neuron can be a fully functional brian for a relatively complex animal--given the right environment (around the neuron). I focus strongly on the molecular level of analysis to hypothesize around how a neuron forms, strengthens, weakens, and destroys afferent connects, the maintenance of short term and long term potentiation in each receptor, and the determination of whether a receptor will be polarizing or hyperpolarizing. I have a theory very solidly based on literature (mostly Erick Kendall's work) for everything but polarizing vs hyperpolarizing (I have concepts for this, just not good backing in literature). A also have a "Central Process" (CP) theory for cognition. Take the view that the general nature of neural substrate all can be fed signals ("co
I am, but I think it should be absolutely fucking illegal for anybody from DARPA to go anywhere near AI implementations. Making minds to kill? Despicable.
University of Cincinnati doctoral student Julia Taylor has done some interesting work on artificial humor, she gave a presentation about it to the Cincinnati Programmers Guild in December 2007.
http://cincypg.org/events/2007/12
http://homepages.uc.edu/~tayloj8/
I'm afraid I can't follow you on your leap of faith, there. The difference between chaotic and random is BIG, a far more fundamental difference than you are making it out to be, and we just don't see anything that indicates any kind of "ghost in the machine".
As far as network latency goes, it is, once again, chaotic, not random. It has the property of being immensely difficult to predict with perfect accuracy, but, very importantly, has an excellent chance of being predicted given a rather limited margin of error and adjacent sample spaces. Your point may be that either form of indeterminacy is acceptable, but if that is the case why do you posit that "free will", if it exists, cannot emerge from deterministic systems?
Also, I'm not getting why you're using network latency to generate pseudorandomness for your MPNNs? Doesn't that cause bad loading behavior? And you're using 1 node per neuron?
I'm not sure why you say that modeling the brain exceeds the available power, since Blue Brain is already capable of modeling significant portions of a mammalian brain.
Honestly, and I hope you don't take this the wrong way, I hope DARPA doesn't fund your ideas. Maybe you're ahead of your time, but more likely I think you're headed down a path that is comfortably self-justifying but ultimately inaccurate. Will randomness be a part of any "solution" in the foreseeable future? Yes. Does that derive from any deep-seated meaning about the nature of man or the universe? That I find very dubious.
I thought it was 4.
I think these two items are foundational for research in ideas and AI:
Variations on a Theme as the Crux of Creativity
by Douglas Hofstadter
ideonomy.mit.edu
Patrick Gunkel
jona@mit.edu http://www.leaflabs.com
No, no, no... you have to be more definitive about it...
Ahem: "Get of my lawn, ya damn kids!"
See? From the diaphragm... and helps if you have a cane to shake in the air for emphasis.
wants to be the first monkey to touch the monolith
Sheesh... noobs ;)
no taxation without representation!
It's called the MacBook Air. People have been slicing bread and cutting cakes with the thing for over a month now. Some people even claim to have sliced themselves open on the seemingly harmless laptop.
8==8 Bones 8==8
My main point, while not explicit was just that its silly and pretentious to post such a question to slashdot.
The goals he described would be so powerful that if he really was on any REAL project working towards them, it would be so secret that he'd already be sitting in Leavenworth awaiting treason charges. Slashdot would also probably have been taken offline and its servers confiscated.
Given that he's still posting, and Slashdot survives, I maintain my point which is that he's on a fool's errand - probably just being asked to do some busy work while they try to figure out what's next.
----(o)----
First, the definition of free will has been debated through the ages, and my definition is not universally accepted. Given that, my definition of free will implies that choices are *not* predetermined, that free will implies a non-determinate system. Now, there is no way I, or anyone, can prove or disprove that humans have free will by this definition (?until we create true intelligence based upon a determinate system?). Therefore, to follow my further arguments, you must accept this assumption that free will implies a non-determinate system, at least for the sake of this discussion.
Free will == non-determinate == truly random, not pseudorandom. Now, since humans interact with a computer network, and humans, by my previous assumption, have free will, network latency therefore is non-determinate.
All of this implies not chaotic behavior, but instead truly random behavior. There are bounds, and the randomness definitely converges. But, at its core, it is truly random, not pseudo-random, and therefore not chaotic.
Network latency does not generate, therefore, pseudo-randomness, but true randomness. Humans interact with the network, introducing truly random behavior. This randomness does not cause bad loading behavior. If you are familiar with neural networks, perhaps you've encountered situations where introducing a pseudo-random delay to a feedback loop can most definitely improve loading behavior for specific circumstances.
Truly random behavior does not imply that it cannot be modeled. It can be modeled. But the model will never be 100% accurate, and will never produce truly intelligent machines (as opposed to artificial intelligence that models, but does not create, true intelligence). Yes, I know that chaotic systems, according to currently accepted math, are non-predictable. But they are determinate. Determinate necessarily implies a lack of free will, by my earlier assumption (you don't have to agree with my assumption in reality, but, for the sake of this discussion, it is central to all of my other logic).
While we may be able to model significant portions of a mammalian brain, our models are not creative. Perhaps, with sufficient power, we will be able to model creativity. But, being determinate, it will only approximate true creativity.
One node per neuron is an extreme case. There must be some interaction that introduces truly random behavior for it to be truly intelligent. According to my prior assumption...this is really breaking down to a tautology, where if you agree with my base assumption, then the rest falls out based upon logic; if you reject my base assumption, the the rest is inherently inaccurate based upon logic.
I'm not saying that "modeling the brain exceeds the available power". I am instead saying that modeling the brain will approximate, but not create, true intelligence. It will get close. It may pass the Turing Test. But it won't be truly intelligent. It will just look intelligent. And, I posit without proof, it will not be creative.
Honestly, I hope you don't take this the wrong way, but I'm not looking for DARPA funding, and I don't give a fuck about DARPA. I've got a good day job, and I doubt DARPA would ever be able to match my working conditions, monetary compensation, and flexibility.
I like to discuss these things to expand my horizons. Slashdot has produced over the years many interesting discussions that expand my horizons (less so recently, aside from this thread--I've become more of a reddit lately). I enjoy discussing my ideas, and pushing my boundaries. The only reasons I'm discussing this here is because this thread will undoubtedly attract people that can expand my horizons. Quite frankly, it has achieved this, as I have found some additional reading already for my summer reading list. And I thoroughly enjoy this thread with you. Although you haven't asked as such, I am a solipsist at my core, so everything in life, this thread included, is mental masturbation. And that, in and of itself, is to be celebrated.
--Be human.
I think that your argument for non-determinism explains the "free" part of "free will" but the implications for the "will" part are puzzling. You claim that human free will includes the ability to act without regard to preconditions (non-deterministically), and that quantum uncertainty is a possible pathway for introducing that type of true randomness. What I don't understand is how introducing that randomness allows an intelligent actor to execute his own will?
One possible explanation is that the introduction of a random element into the cognitive process allows him to generate multiple potential future scenarios which compete deterministically as the actor chooses which to pursue. While the intelligence has no control over how the random element influences the creation of those scenarios, it does have deterministic control over the selection process.
I suppose that the randomness could also be integrated into the choosing process by randomly mutating the mechanism used to choose and selecting those that perform better over time.
All this being said, I still don't see how making an intelligence non-deterministic implies anything similar to "free will." It may be a conflict of definitions, since I take free will not only means that there is some non-deterministic decision making process, but also that the actor actually retains control of that non-deterministic process. I should be able to decide between chocolate and vanilla ice cream consciously and non-deterministically.
As a matter of full disclosure, I should note that I am not convinced of humanity's own free will, since it basically just means having the ability to choose to act in a way that is at odds with the innumerable causes that have led up to the decision to act. Also, I think my definition implies a continuous, identifiable aspect of intelligence which exists outside of physical reality. I don't see any evidence that either of these are necessary to explain what we observe, but thinking about the implications of true randomness in the cognitive process has given me new food for thought.
You and I aren't going to see eye to eye on free will. I don't buy it and you do, for reasons that each of us considers convincing- but I have to wonder what your reasons are. I just don't see the evidence, and lets be honest here- proving randomness is a bitch. There are myriad possible explanations for our difficulty in modeling human behavior that do not depend on mystical intervention or unknown and unobservable phenomenon. Why do you believe something that flies in the face of all we know about the rest of the universe, when there is not a shred of evidence to support your claim? Is it because it is a comforting thought to believe that alone among all the collections of atoms in the Universe, we control our fate? Is it a justification of the notions of good and evil? Of God? I really don't understand, and would love to hear a cogent argument for its existence.
Part of the problem you've got in talking about AI is that you are talking about stuff you've seen in movies, not the actual state of AI. AI is creative now- look at the wifi antenna that it created, or the wind instrument playing robots that made headlines just a few weeks ago. Does it depend upon chance, upon permutation? Sure, just the same as a Monte Carlo simulation does, or a genetic algorithm, or a BPNN, but no more so, and you don't call any of those things "intelligent". In fact, I would venture so far as to say that you have a problem of definition on your hands- that you have decided that humanity alone holds the power of free will, of true intellect, and have derived all your subsequent reasoning from that point. That seems to me unworthy of somebody as obviously intelligent as you are, and I hope I'm wrong about it.
Your statement about latency coming from a random source and therefore being random is clearly false, as it is trivial to construct a distinguisher function which imposes arbitrary order onto a random input, as you yourself argue when discussing convergence.
I am familiar with neural networks, and am still not understanding how using a metric that depends upon the performance of your network to dictate the behavior of your network produces random behavior.
Your statement about modeling truly random behavior is, so far as I can tell, entirely in error. Perhaps you meant that it was possible to model systems whose state includes random information? It is most certainly not the case that you can model a truly random system with a non-negligible chance of success.
I'm glad you enjoy the mental exercise, I have too, but I'm still not getting your fundamental premise. You don't seem to be concerned in the slightest that there isn't any evidence for it, that it is impossible to build constructions based on, or that it is purely unfalsifiable. I find that strange, and look forward to your response.
Chaotic effects produce pseudo-random numbers. I am unaware of any truly random number generators that are not based upon quantum effects. I believe that quantum effects are the only physical phenomena that are believed to be truly random. Even then, some great minds (e.g. Einstein) never quite believed that quantum effects are truly random. "God doesn't play dice".
--Be human.
Yeah, I feel sorry for my employer, too. He is me. My existence as a solipsist places me firmly in the realm of kooks. But, damn, I'm had a history of success in the field of financial analysis that is hard to attribute to chance.
--Be human.
Yeah, I pretty much surmised it's a question of the definition of free will that was causing us to disagree. I believe that free will (according to my definition) exists, and that it is the spark that changes the living brain from a simple automaton to a creating, thinking entity. I don't believe in a soul, as per the religious concept at least. But my definition of free will most certainly does require non-deterministic behavior. Whether free will==intelligence is debatable, though, and I fully accept it as a non-provable (or disprovable) thing until we create true machine intelligence.
As far as AI is concerned, I don't believe that genetic algorithms (the sort of thing used to create this antenna, I presume, or the sort that has proved exhaustively the strongest form of a column at my alma mater) is true creativity or intelligence. It approximates creativity. It models intelligence. But it is not creative, nor is it intelligent. It is a tool, used by creative and intelligent beings to achieve a goal. Without the intelligent driver, I don't think such creativity is possible short of exhaustively trying every possible combination of everything. I don't think that's what creativity or intelligence is, and I don't think it's possible except in very narrow, well defined areas (that are narrow and well defined by the external creative/intelligent driver of the tool).
I don't think creativity and intelligence are embodied only by humans (I think very many living things are creative & intelligent, but my familiarity and use of AI tools to do financial analysis have given me the belief that nothing I have seen in the field of AI classifies as intelligence or creativity). That's nitpicking on your use of the phrase, "only by humans", though. I don't believe anything humans have manufactured/programmed would fit my concept of intelligence or creativity.
I'm not sure I follow you on the distinguisher function bit. Perhaps you can discuss that further? I'm not sure how imposing order necessarily removes the non-deterministic input from the system. My point about convergence is that while each individual datum of input may be non-deterministic, the overall distribution of inputs may well be modeled quite easily. The model won't be accurate/deterministic at the minute level, but the net effect of large numbers of input will be model-able and deterministic (this, in fact, is exactly what adding a clock to the quantum effects of a transistor does to create a deterministic switch out of a non-deterministic core effect of electrons maybe jumping from the N to the P junction).
My concept of embracing this non-deterministic behavior of transistors is analogous to my understanding of how individual neurons work, and how their outputs cascade as inputs into other neurons to create thought. The exact time that an individual transistor/neuron fires cannot be predetermined. But the net effect of a network of these transistors firing produces a result that converges over time. My premise is that occasional changes in timing for firing of individual transistors/neurons is (or may be, at least) the spark that initiates creativity, and that creativity is a prerequisite to what I call true intelligence. The only problem I see is that all avenues of research into quantum gates for quantum computers, or into asynchronous computing, intentionally attempts to eliminate the effect of "misfires" of transistors, or quantum gates, since non-deterministic logic doesn't really serve us well in the field of computing. I think we need to break free from that avenue of thought, and to embrace these "misfires".
My point about modeling systems whose state includes random information is that we can create such models that work *most* of the time. We can model the aggregate behavior of the system, but we will never be 100% accurate with our model. If we were 100% accurate with the model, then we would have the basis for the creation of a truly random number generator, and that would be truly newsworthy. W
--Be human.
Yeah, there are tons of 4-digit UIDs. Wish I hadn't lost the login credentials of my original 3-digit one :-( Poor me.
--Be human.
Nevertheless, it is widely believed that an RNG of cryptographic strength can still be devised from chaotic effects given that the output length remains finite and of significantly lower magnitude than the internal state of the RNG. This is due to the effectively unlimited periodicity of some chaotic effects. Caution should be advised when using those effects that are easily observable or manipulable, but otherwise, chaotic effects should have sufficient cryptographic strength to render it impossible to build an advantaged distinguisher.
you make me feel so young!
What's the point of asking for projects so you can overlook them? Surely you can overlook them purely by remaining ignorant of their existence...
Two US patents have already been granted for true (Turing) language-simulation AI -- not that flaky neural net approach ... #6587846 & #7236963 ...
More at
http://www.emotionchip.net
http://www.ethicalvalues.com
Inventors BEWARE !! The Fed Gov't typically will not pay to use your patent -- You have to sue in Fed. Court...
JLM
Look who's talking...
I would recommend that if darpa is serious about developing the first true AI that you start off something like angel funding for AI ideas, and get a bunch of different proposals that will be taken seriously. There are tons of people out there that have different ideas about true AI and how to do it. If you have any contact with AI academic world recently, you'll see a bunch of kinda interesting projects that have lost sight of true AI. Of course, outta the people who claim have solved AI, 49% are stupid, 49% are crazy, 99% are in for a big disappointment, and that last 1% are actually in the right direction. Kinda reminds you of proposals for startup companies huh? So support them like startups.
Angel funding have proven that they are important to promote creativity without wasting a fortune. Give developers (with potential) some money, NDA's, and most importantly, if they seem like smart people, an open ear to their ideas. Don't just do stuff like Grand Challenges that only research institutions/big companies can enter.
Well, that said, I'm actually very biased. Because I know how to solve AI =P. Of course, you prolly won't believe me, and I don't wanna share unless you sign an NDA. So, here's just a suggestion to think about.
AI is the process of solving problems, like chess -- solve the problem of winning the game. Solving an arbitrary problem is HARD, so let's ignore that for now. Suppose you have solved the problem, you'll have to carry it out somehow. What is it called when you are doing something in a specific way? An algorithm. Right now, MDP's use policies -- you're in this state, do this. Classical planning gives you a sequence of actions. But Algorithms? CS? Programming? Why aren't the results of these problem solvers turing complete? No wonder what comes out of AI programs today are very domain specific.
If you solved an arbitrary problem, you should be able to program it. AI can be called a program that makes programs. So here's the suggestion: start working on a programming language fit for AI -- the internal language that the AI brain will use. Programming languages today describe the process -- each program is compiled one specific way, and ran one specific way. Make a programming language that describes the algorithm, and only that. Whatever you program should give enough information about the problem you solve and nothing more. A rule of thumb is that if you can't use the language for a cook book, you need to work harder about eliminating the non-necessities. lchou1.blogspot.com
There are many approaches to AI that don't try to emulate the human brain.
You can emulate the human brain but try to produce something other than intelligent behavior (say a non-linear controller)
It would help (both you and DARPA I suppose) if you knew what you wanted to do and got the terminology right first.
Check out
http://ars.ict.tuwien.ac.at/
The trick is to introduce findings from psychoanalysis and neurology into AI. A totally new and promising approach. Is supposed to be used for intelligent buildings (ambient assisted living, etc.) one day.
I recently began a project to emulate evolution through mutation and natural selection (survival of the fittest). Essentially one must create a virtual nature of bytes not atoms, a virtual computer, and execute a simple self-duplicating program in this virtual space, a program whose bytecode is interpreted and can spawn and execute new programs and mutate and modify itself, and one which must interact with the virtual environment in such a way as to gather materials for building a new organism (necessitate navigation and interaction to encourage development of complex behaviours). Impose the potential of death to create evolution, such that beneficial mutations are better able to survive and reproduce. To watch different concepts of evolution actually manifest themselves in a simplified universe and to see what sorts of behaviours develop in generations far older than the first organism (the bytecode of which I must write myself and assume that the case could randomly occur without my help) will be quite encouraging. Difficult to explain perhaps, but the idea may interest you.
If you're really serious, check out Phaeaco and this.