We may not have arrive in the dream 'promise land ' for AI but as you rightly pointed out, there is enough evidence now to tell us that we are getting closer. AI is definitely making things much more easier and fascinating for us: Envisioning a Future for Intelligent Software Agents.
|
Does Verizon's Voyager stack up to the iPhone? |
|
|
5 IT skills that won't boost your salary 
[1,407]
Women 4 times more likely than men to cough up personal info [589]
Japan's 10 funniest tech-related commercials [Videos] [407]
Throwing away a promo CD is "unauthorized distribution"? [1,265]
Adults too quick to dismiss educational video games [682]
Attack of the iPhone clones [Slideshow] [578]
10 things IT needs to know about AJAX [1,258]
This Year's 25 Geekiest 25th Anniversaries [Slideshow] [409]
|
|
Get out of the lab and look at neuroscience
The term AI is dying because it is associated with the old way of programming but the concept is still surviving under different names. In my opinion if it was for video games and MIT the AI style of programming would be obsolete. AI test the strategy of the programmer where as other programming techniques such as machine learning is based on statistics from real world data.
The first setback is due to code being written in the lab for the lab in what is best considered a toy problem. If you get out of the lab and into a real environment everything fails. We are now just getting out of the lab and we are now facing real world problems. Just look at the DARPA grand and urban challenge. Extremely large advances were made in a relatively short period of time.
There has been a slow progression because people first solve the toy problem then realize that the solution to the toy problem does not work. Then they go back to a more sophisticated toy problem. How long does that have to last before people realize that we are ready to solve real problems. By the way AI in games is an example of a toy problem.
The last suggestion is that we need to look at something that has already found a solution. Primates have solved this so called vision problem that people think is impossible to solve. So how come almost zero machine vision solutions incorporate strategies determined by the neuroscientest? Because they are focused on improving the speed. In the long run it will be better to share resources (a strategy in the brain) than to have several small isolated programs with one sophisticated arbitrator.
Out of the lab..
You are correct. Almost any problem is a toy problem alone but put any of those together and the trouble starts. When AI was high in 80's I wrote a computer configuration / predicting system. All cool and well as long it did the normal configuration problems but because it had a "learning" from new hw/sw, customer and benchmark results and got more accurate, people started arguing with it. So - add the human factor, hundreds, each one different - not so much liked anymore! The use died in 8 years and was replaced by a simple configuration report writer, even the report program needed much more upfront work and the results were much, much more inaccurate, never enhancing, BUT it didn't argue so the user, right or wrong, had the feeling making the decisions? And of course I should have added the business learning also - some, for customer, optimal configurations don't have as high markup for salesman than some less optimal system and there is no way to argue that, from salesmans point of view!
So, AI (in a way) can be done but I think not the way it was dreamed, otherwise there would be no place for political systems, wars, salesmen, or any other believe systems - if it is not AIs own system!
The False Premises of AI
Before you design Artificial Intelligence you have to define intelligence and there is no consensus and no scientific definition of intelligence. For example, we have the various IQ tests that have littered the past 100 years or so and all of them are contradictory and all of them are based on circular logic (the test defines intelligence, therefore intelligence is the ability to pass such and such a test) or models that are heavily reliant on subjective and often socio-political bias.
So mistake number 1 was that AI assumed that what the IQ test models measured was 'intelligence'.
Second mistake was to disembody and abstract 'intelligence' as an eternal singular entity, something that really should remain in the realms of science fiction. This meant that they could 'code' intelligence and then the mind, develop robots that could fight in wars (the original aim) and a whole raft of other goodies. This was complete nonsense and as critics pointed out in the early days this approach ignored the role of bodies in intelligence and the role of neuroscience (the real mechanics of how brains and minds worked).
So what did late 60s/early 70s AI offer? The most advanced computers in the 70s and 80s were designed by people that adhered to the identity of computers as 'discrete state machines' and engaged with engineering principles. This led to advances that enabled other areas to flourish - the advances in game programming, the automation of business processes, and interesting experiments in neural networks that actually took account of advances in neuroscience. These in turn led to the application of computing and technology to areas of healthcare that engaged with the body. In comparison AI is a 'failed research project' and we have seen a decisive 'paradigm shift' away from its nonsense. Despite this people like Minsky still claim that plaudits for anything that 'appears' to be moving towards intelligence like smart searches, the internet, etc, etc.
What happened??
AI alwasy appeared to be a solution in search of a problem.
Scam!
You still have to think!
http://fakesteveballmer.blogspot.com
AI is a disguise
"Advancements" in AI are really advancements in mathematical models and Complex models and rules that now have a chance to be deployed on devices and servers and imbedded in application programs. But the a more holistic development of a true AI model is what that has not been delivered. Take the "A" out of the "AI", call it something like CS-I and then everything will be right in place.
Gaskin misunderstands AI
Pre-fetching of applications by Vista and traffic routing based on accident history are examples of simple statistical analysis - not AI. In the 1960s the people who foolishly believed that AI was right around the corner had no problems with their definitions. Simply put, AI was a machine capable of doing anything a man could. At least intellectually. More than three decades later AI is as distant of a dream as ever (perhaps more distant now that we know just how lofty our goals have been), and the AI researchers make half-hearted attempts to pass an automatic vacuum cleaner for a Start Trek computer. Now they are playing with the very definition of AI. Search engines are "intelligent" because they can strip a word down to its root; cell phones are "intelligent" because they have voice dialing; and even vacuum cleaners are "intelligent" because they don't run into the same coffee table twice. So in almost forty years we went from the goal of human-like intelligence to simple adaptive algorithms feeding off rudimentary sensors. And the likes of Gaskin are now trying to convince us that this is exactly what AI is all about: avoiding coffee tables.
Point proven?
Put all of those together and you have the Jetsons' Rosie: a vacuuming robot that responds to voice queries while not running into the coffee table. Given that some people can't manage that same set of tasks, I'd say we have a winner.
AI definition is wrong...
Part of the failure in AI is using it to define intelligent behavior. Ants have incredibly intelligent behavior, but by most definition are NOT intelligent. A better definition for intelligence would say something about adapative, predictive algorithms. Using this definition nearly all of the advances described in the algorithm completely lack intelligence and are dumb ants (albeit with very useful intelligent BEHAVIOR).
Ants and managers..
Actually ants are proven to have intelligence, they have managers who make independent decisions! So, if you say that managers are intelligent so are the ants! I know, in IT world we still debate that but..
Jokes aside, many, almost all animal life has proven intelligence, it just is different what humans have. But learning of experience, teaching it to next generations, advising, inventing tools, and so on - you can see it all over the animal world. Of which we are part, arguably on top but still learning, I hope..
Post new comment