Slashdot Mirror
U.S. Plan For "Thinking Machines" Repository
An anonymous reader writes "Information scientists organized by the US's NIST say they will create a "concept bank" that programmers can use to build thinking machines that reason about complex problems at the frontiers of knowledge — from advanced manufacturing to biomedicine. The agreement by ontologists — experts in word meanings and in using appropriate words to build actionable machine commands — outlines the critical functions of the Open Ontology Repository (OOR). More on the summit that produced the agreement here."
Shit, we really are going to have to start watching and learning from the terminator films now.
If computer history tells us anything, they will create more data then we can understand in a short amount of time.
The Kruger Dunning explains most post on
Wow, this can be scary. I hope the US is investing in a primitive non-computerized emergency plan to destroy this project, in case of the uprising. There has to be strict limitations placed on this sort of system, not just 3 rules. This is one time when the lessons learned from fictional books/movies would come in handy. I'm serious too.
42.
I for one would like to welcome our thinking machine overlords...
Singularity here we come!
But damnit, what was the question?!
The enemies of Democracy are
Just imagine how fast they could post catchphrases! They will hunt down low numbered users. AC is humanity's last hope for survival.
from TFA: OOR users, tasked with creating a computer program for manufacturing machines, for example, would be able to search multiple computer languages and formats for the unambiguous words and action commands.
from my experience, the ambiguous words is the documentation, followed closely by the comments.
"unambiguous words and action commands"? Is this what "experts in words" call a computer language syntax? now we're going from "you don't need to be no stinkin' programmer, all you need to do is point and click and connect the dots" to "it will create programs by searching computer languages for action commands".
wow. good luck with that. just what we need, another good AI boondoggle.
Forget about the "reasoning". The agreement is about creating standard ontologies in different fields (contexts). Personally, I think it will be very difficult because first they will have to gather experts in all those fields (may it be biomedicine or business processes) and define a way to express all this knowledge. Of course, OWL is the ontological language to use, but they will need a serious bunch of guidelines to keep the model consistent.
Somebody claims to be able to build a ''thinking machine''. All efforts so far have failed. There is reason to believe all efferts in the forseeable future will also fail. It is even possible that all efforts ever will fail, as currently we do not even have theoretical results that would indicate this is possible.
So why these claims again and again, and (I believe) often against better knowledge by those making the claims? Simple: Funding. This is something people without a clue about information technology, byt with money to give away, can relate to. Basically the same scam the speech recognition people have been pulling for something like 40 years now. Personally I find this highly unethical. When you confront these people, they typically admit the issue but claime that other good things come from their research. My impression is more that they are parasites indulging themselves at the expense of honest researcher that work on things that are both highly needed and actually have a good chance of producting usable results.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Thou shalt not make a machine in the likeness of a human mind.
It's merely intended as a convenient resource for programmers.
Caveat Utilitor
OK. I know, this prediction has been made before, but now it's for real, because the hardware capacity is well within the reach of Moore's law. To build a cluster of processors with the same data-handling capacity of a human brain today is well within the range of a mid-size research grant.
Unfortunately, they have cried "wolf" too many times now, so most people will doubt this, but it's a reasonable prediction if one calculates how much total raw data-handling capacity the neurons in a human brain have. Now, software is another matter, of course, but given enough hardware, developing the software is a matter of time.
When I first read the headline I thought it was referring to Thinking Machines of Danny Hillis fame. You know, the hypercubic CM series. "Do you know anyone who network three connection machines and debug 2 million lines of code for what I bid for this job?"
Yay, another AI bubble to be followed by another crushing AI Winter!
While it's a bad thing the first AI Winter unjustly tarred Lisp (a general purpose language good for lots of stuff) in many people's eyes with the same brush of "fail" as AI, a lot of the current AI weenies are basically just porting 20+-year-old Lisp-based AI stuff to XML+Java and passing it off as new. I predict this bubble will burst fast.
So why these claims again and again, and (I believe) often against better knowledge by those making the claims? Simple: Funding. This is something people without a clue about geography, but with money to give away, can relate to.
Is it somehow actually different and/or better than the BSD licensed WordNet that's been active since 1985 or is it a case of NIH syndrome?
http://en.wikipedia.org/wiki/WordNet
http://wordnet.princeton.edu/
Computer thought is probably no more advanced than that of a bug. Mars rovers etc can only executed canned move sequences and don't operate autonomously. Some robots etc are more autonomous, but are still pretty limited when it comes to any biological equivalent.
As much as people have been predicting thinking machines for the last 60 years or so, the reality is a lot less impressive.
Engineering is the art of compromise.
Let's take the example of a simple idea: a pun. This is a word that in a given context can have more than one possible interpretation. One can classify either one or both of the interpretations as the ideas expressed, but that would be incorrect. Often times it is the presence of both meanings that give the pun a new meaning that joins the two contexts.
It is the interconnections between contexts that generally give new insight into subjects. Repositories of existing concepts can only be used to explore the implications of the already known connections. I don't see how they can come across connections which can be formed, but which cannot be formed from what has already been stated.
The downside to forming such a repository and such an exploratory would be that it would discourage human-based exploration of the already known ideas. Human based exploration of the already-known ideas serves as a means to training people who may at a later day discover connections which cannot be currently formed. By discouraging such training of humans it would ensure lessened pathways to exploration in the future.
Any guest worker system is indistinguishable from indentured servitude.
I thought the article was going to be about Thinking Machines the company. I got to see a CM in all its blinkenlight glory when we toured Schlumberger's lab in high school.
They say the mind is the first thing to
This will mean a bit of pain for a while.. but the Butlerian Jihad will fix things.
Would be to create a computer-based system for assisted thinking. By that I mean something along the lines of what the visual thesaurus people have created only which would allow people to populate their own interconnections. Something that would allow people to form easy ways of presenting the data they think about as well as as interconnecting it. Currently we are sinking under the weight of the cross-referencing. It takes half-a-lifetime to train someone in some narrow subject because of interwoven network of cross-references. All this can be easily automated with a dedicated interface project. I am not sure that AI is even possible as thinker because AI is unable to go through human experience. But AI certainly is possible as a deducer of implications from what's already known... but still not of full implications.
Any guest worker system is indistinguishable from indentured servitude.
The guys at cyc (look for wikipedia entry too) are already halfway there. Last time i checked there were already something like 5 million facts and rules in the database, and the point where new facts could be gathered automatically from the internet was very close.
Many years ago i remember the founder (Doug Lenat) saying that practical purpose intelligence could be reached at ten million facts....
we'll see within the next decade, i guess.
We learn from history that we learn nothing from history - Tom Veneziano
What sense is there in trying to encapsulate "concepts" particularly when phrased in language? Both of these are fluid and evolving. Attempting to archive a particular static state is at best a waste. Ontologists above all should know this.
And maybe that's the point. For centuries ontology has existed primarly to serve itself and secondarily to trade favors with other branches of philosophy. The proposed project has the primary result of providing gainful employment outside the halls of academic philosophy for the first time. It has the secondary result of allowing ontologists to get their revenge for centuries of being ignored by getting us to pay them for something that had we only listened to them we'd know to be a monstrous leg-pull.
Let them have their fun and their Venn diagrams. Cognitive science already knows better, and is better suited to conceptual mapping with its cognitive mapping in conceptual space.
"I may be synthetic, but I'm not stupid." -- Bishop 341-B
You forgot to mention that it will fail for the exact same reasons that Good Old-Fashioned AI has always failed. All the classifications in the ontology, when actually applied to any real-world problem, will turn out to be unexpectedly and hopelessly fragile.
Are you adequate?
Building a standard "ontological repository" would seem to require establishing a structure within which its objects and relationships can be contained.
While this might seem to be of benefit to extending the capabilities of some tasks like machine translation into broader fields, I think this might cause problems at the cutting edge, that is: machine reasoning.
Reasoning about complex problems at the frontier knowledge (to paraphrase TFA) requires identifying new links and relationships between objects. Nailing this structure down would seem to hinder this. It might make lower level tasks (pattern recognition, etc.) simpler. But you need to continually 'break' and 're-sort' the knowledge database to accomplish this.
But then, what do I know. I'm just an EE that was working this stuff about 10 years ago. We got this far when the company boxed it all up and sent it overseas.
Have gnu, will travel.
Lots of people are making posts about this vs. skynet, terminator, etc. But there are some problems with that (overly simplistic and totally misguided) comment.
There are numerous formal logic solvers, that are able to come to either the correct answer (in the case of deterministic systems, for instance) or to the answer with the highest degree of success. The difference between the two should be made clear: Say if I give the computer that:
A)All Italians are human. B)All humans are lightbulbs.What is the logical conclusion? The answer is that all Italians are lightbulbs. Of course, the premises of such an argument are false, but a computer could work out the formally correct conclusion.
The problem these people seem to be solving is that there needs to be a unified way to input such propositions, and a properly robust and advanced solver that is generic and agreed upon. Basically this is EXACTLY what is needed in order to move beyond a research stage, where each lab uses its own pet language.
I mentioned determinism, because the example I gave contained the solution in the premises. What if I said, "My chest hurts. What is the most likely cause of my pain?" An expert system (http://en.wikipedia.org/wiki/Expert_system) can take a probability function and return that the most likely cause is... (whatever, I'm not a doctor!). But what if I had multiple systems? The logic becomes more fuzzy! So there needs to be an efficient way to implement it, AND draw worthwhile conclusions. Such conclusions can be wrong, but they are the best guess (the difference between omniscient and rational, or bounded rational).
None of these things are relating to some kind of 'skynet' intelligence.
IF you DID want to get skynet like intelligence, having a useful logic system (like what is planned here) would be the first step, and would allow you to do things like planning, for instance. If I told a robot, "Careful about crossing the street." it would be too costly to try to train it to replicate human thought exactly. But it records and understands language well (at this point), so what can we extract from that language?
Essentially, this is from the school of thought that we need to play to computer's strengths when thinking about designing human like intelligence, rather than replicating the human thought processes from the ground up (which will happen eventually, either through artificial neurons, or through simulation of increasingly large batches of neurons). On the other hand, if such simulations lead to the conclusion that human level consciousness requires more than the model we have, it will lead to a revolution in neuroscience, because we will require a more complex model.
I really can't wait to get more into this, and really hope it isn't just bluster.
Also:
'Thinking Machines' title is inflammatory and incorrect, if we use the traditional human as the gauge for the term 'thought'. It is a highly formalized and rigorous machine interpretation of human thought that is taking place, and it will not breed human level intelligence.
I, for one believe in this, and welcome my new artifically intelligent overlord.
If video games influenced behavior the Pac Man generation would be eating pills and running away from their problems.
Putting tags on your del.icio.us links doesn't make you an ontologist any more than using object oriented methodologies makes you a platonist. I think the correct label for those who misappropriate terminology from other domains (for no other seeming reason than to make them sound clever) is "wanker". Hell, call yourselves "wankologists" for all I care, just don't steal from other domains because "tagger" sounds so lame.
I called my cable company the other day and got an automated response that asked questions and responded, not only with words and instructions but also with a modem reset. The computer system could ask questions, determine responses and perform actions. Yes, it was limited, but decades past it would have been considered awe inspiring and doubtless would have been dubbed both a successful artificial intelligence and thinking machine.
What then is the proper definition of a thinking machine? We already have computers that can follow complex logic paths to arrive at unexpected results (bugs?) and offer solutions we would not have foreseen on our own. Similar in result to having a conversation with an expert in an unfamiliar field.
As machines, both hardware and software become more complex and capable, we are already raising the bar for what we consider an artificial intelligence. Doubtless we will continue to do so for quite some time, but when you can talk with a machine built on the ability to work with volumes of processable knowledge such as is being compiled in the OOR, how will we raise the bar?
Historically, humanity has considered people that they considered unlike themselves to be less than fully human. As the majority of our species progresses toward a more inclusive standard, our language and perception is becoming inadequate to differentiate a human from a very advanced machine. Already most of us consider the issues of race, language, geology, age and affiliation to be irrelevant to defining what makes someone human. Biology is even a wavering standard since we consider people with prosthetics to be people with human rights and human bodies with the inability to think (vegetables) to have none. We are left with the ability to think and biology as the standard, but the definition of thinking is somewhat hazy to say the least.
I think therefore I am, but what does it mean to say "I think" and how do you define thinking without biology in an external entity?
Back in my day when we chiseled our bits into stone and sent them by mule train from village to village...
The Stone Age lasted a few hundred thousand years. When we learned how to use metals, the Bronze Age lasted a few thousand years. Then came the Iron Age. We only learned how to make steel in an industrial scale in the nineteenth century, the Steel Age only lasted a hundred years, then we got into the plastics and composite materials age.
Technology accelerates exponentially, it's very risky to extrapolate from the past. We cannot work backwards and expect to get any reasonable predictions for the future.
Sounds very close to the Semantic Web (http://semanticweb.org) to me.
Does this remind anybody else of the prime number ontological schema talked about in the Baroque Cycle (by Neal Stephenson)?
Maybe it's time for MIT and other tech Universities to start a Mentat degree?
>> To build a cluster of processors with the same
>> data-handling capacity of a human brain today
>> is well within the range of a mid-size research grant
Nope. The brain is hundred billion neurons, connected by 100 trillion synapses. Sure the "clock frequency" is very low, but even taking this into account, those figures far exceed what could be built with today's technology. Not to mention that scientists today have absolutely no clue how major parts of the brain work, so even if hardware was available, it'd take decades of tinkering to get anything reasonable running on it.
Well, if you want to be picky about "ontology", why not your misuse of "methodology", which is rightly the "study of methods and techniques" but has been morphed by popular use to be synonymous with "method" itself.
This really ties in with this article: http://news.slashdot.org/article.pl?sid=08/05/28/2217230
So, we don't want to fund proper science, or proper education, but we want to build machines that can think for us, so we can concentrate on the important things, like believing that the war in Iraq is about bringing freedom and democracy to the poor people and that the world was created in 6 days (BTW, how can one even talk about days before the creation of Heaven and Earth, and crucially the sun?)
Not that this kind of research is bad in itself - we already have 'logic computers' that can construct mathematical proofs, which has made it possible to advance in some areas, where brute force seemed to be the only way forward and where the task simply was too overwhelming for a human to take on. But the danger is, of course that this kind of technology will make us intellectually lazy and incompetent, just like many people now are "psysically incompetent" because we have machines to the work for us.
In the last 15 years there was ZERO progress on the AI front. All we need is program that can learn, just basic stuff to begin with, on the level of 3 year old. No one has done that yet. I actually think now it is impossible.
Look at this this way. If I create a program that can do what I myself can't do... Or if it derived a fact that I did not know how to come to. Then I can instrument the program and obtain a log file of steps taken to solve certain problem... Remember, we still use von Neuman deterministic h/w.
Then, not only this software solved this particular (and probably insignificant) problem, but more importantly, it showed a way to algorithmically solve problems in general!!!
Chances of that happening a worse than that of a monkey typing War and Peace in 1000000 years.
Despite the capitalization, this is not a reference to the Thinking Machines Corporation, recently featured on the DailyWTF.
All this "agreement" is about is to have a repository for everybody's "ontology" data. It's like SourceForge, only less useful.
Most of what goes in is tuples of the form (relation item1 item2); stuff like this:
(above ceiling floor)
...
(above roof ceiling)
(above floor foundation)
(in door wall)
(in window wall)
The idea is supposed to be that if you put in enough such "facts", intelligence will somehow emerge. The Cyc crowd has been doing that for twenty years, and it hasn't led to much.
The classic paper on why this idea is bogus is "Artificial Intelligence Meets Natural Stupidity", by Drew McDermott. That was written in 1974, and it's still relevant. There are plenty of citations of this paper on the Web; if anyone can find the full text, please provide a link.
If something like this ever works, it will probably look more like Bayesian statistics than deductive logic.
Where can I go to get a degree in "wankology"
...taking this project seriously.
:>
Its name means "boob" in my first language
One that hath name thou can not otter
Computer systems already display some basic qualities of living things - especially when we look at the global system(s). For example:
1. Self preservation - as a whole and in part, systems do back-ups and activate redundant power supplies to preserve their 'state'.
2. Computers consume, process and exhale nutrition or energy.
3. Systems self replicate - some even develop new software more complex than humans could come up with.
4. Humans rely on and interact with these systems all the time... think of the software that keeps the power grid up, traffic signals, financial systems, inventory control, etc.
5. Computer software grows, learns and changes over time - as a whole there exponentially more working code now than just a few years ago.
6. Some software is very hardware independent - think self replicating software viruses and malware...
Consciousness, as humans perceive it will seem to develop as expert systems evolve - many humans will be fooled...
And here I thought this article was going to be about the Connection Machine which is one of the coolest looking computers ever built.
All you did was use the paper towel to wipe away the scent trail the ants were following.
Ant trails are created by scouts doing a random walk. When they find something tasty hey follow their own trail back to the nest and all the other ants follow that same trail and also strengthen it.
Occasionally an ant gets lost, starts a random walk and often run into the line again. If this new path is faster it will tend to displace the original line. Once the food is gone the ants will disperse where the food was and the return path will weaken and eventually go away.
By disrupting the path long enough you removed the scent and the colony reacted as is the food from that source was used up. Eventually another scout randomly found the food and started a new path.
I am very comfortable with computers running simple programs and always reacting the same way to stimuli.
When a computer can analyze itself and reprogram itself based on it's own interpretation THEN I will be worried.
Can you imagine a computer having any sense of humor? I cannot, since it is a very human characteristic. As much as the thought. This project may be renamed to something more related to the process of data, instead of reasoning and thinking. That's the only thing machines can do. And it is easier to be funny and humoristic than to conceive new ideas and reflections!
Submitter / article summarizer mistakes the difference between "ontology" and "semantics".
Was I the only one to think of Thinking Machines Corporation? All the connection machines in Australia (well the CM5 ones anyway) ended up being merged into one big CM5 in Adelaide, and finally turned off in early 2002. I last used it during 2001 to run some legacy code to get a baseline of what it did for a porting project.