How smart is the smartest AI?

This is a tricky question and I hope it doesn’t turn into a debate, but I’d like to get a handle on how smart Artificial Intelligences have gotten.
Defining intelligence is a mean task and I won’t pretend to do it.
Maybe an IQ score could be assigned or perhaps a comparison to an animal would suffice.

I dunno about intelligence, but the AI in Black and White is pretty cool. I taught my monkey to light its poo on fire and chuck it at enemy held villages.

Of course, he never really thinks for himself, he can only learn and all, but if you want to see some really cool AI and an example of how lifelike things are becoming, check out Black and White.

Somehow not the response I had envisioned. But good for a laugh.
Is “Black and White” a sim game? I could check it out with a little more info.
I want to know what the cutting edge is though, not just what’s commercially available

I have news for you, the programs in B&W are not a real artificial intelligence. That’s just what they call it. It is actually a “heuristic algorithm” which is just a fancy way of saying it is able to modify its behavior within a limited range of actions, based on inputs. The ability to learn does not mean a program is actually intelligent. Your monkey behaves as it does because someone programmed that possibility into the program. Your “teaching” only channelled the “behavior” into a different predermined algorithm.
I often cite a story from a famous AI researcher, he described a scene from “I Love Lucy” that he said fits his description of AI. Lucy is on an assembly line, packaging cupcakes that come out of a machine which puts on white frosting and a cherry on top. Someone drops their bowler hat into the machine, and it comes out covered with frosting, with 5 cherries symmetrically placed around the hat top with one on the middle. He said that this would fulfill his criterion for intelligence, an unfamiliar situation occurs and the machine uses its intelligence to solve the problem in a creative way, a way beyond its original programming.
Anyway, there was an almost complete stall in the AI field throughout the 80s and much of the early 90s, there really is nothing out there that comes anywhere close to even the intelligence of an insect. Some AI researchers have said that what we are now seeking isn’t artificial intelligence, it’s artificial instinct. If they could make a computer with just the instincts of an ant, that would be a huge breakthrough.

Cutting edge I do not know, but as Chas.E said, the “AI” in B&W isn’t really AI. I still think it’s really cool though. Especially when there are so many possibilities.

B&W is a god-sim in which you are a hand and your creature is your avatar, much like the Dungeon Keeper series. Totally awesome game, and hours and hours of fun. Required system specs aren’t that high, so if you bough your computer in the last year and i wasn’t a bargain bundle deal thing you should be able to run it ok. Developed by Lionhead in the UK, it’s available in all “big” countries. Just check out your local computer store, they should still have some copies lying around.

Anways, sorry for the little highjack, I just felt the need to share one of my funnier gaming experiences.

So how is instinct different from pre-programmed commands, which all computers have? Is instinct so much more than a set of if-then statements?

First off Black & White is a pretty cool game. While the programming may not constitute a true AI it does go beyond a clever branching system where all you are doing is guiding your ‘creature’ into canned responses.

Obviously the ranges of behavior are limited but the program allows them to be mixed and matched into a HUGE variety of combinations. Even the programmers at Lionhead occasionally are stunned at seeing their program do something totally unexpected and original (I think flaming poo might fall into that category).

True AI is a long way off. In short it seems like the goal would be Commander Data from Star Trek:TNG or perhaps HAL from 2001. Till then picking the ‘smartest’ AI isn’t really possible.

Deep Blue was the IBM supercomputer that first beat a Grand Master chess player (Gary Kasparov). AI? Not really. Just clever programming and brute force computing power. Either way that program couldn’t do anyting more than play chess. Sit it down to play Tic-Tac-Toe and it’d be completely lost.

Black & White and Deep Blue illustrate that most of what we call ‘AI’ today are merely expert systems targeted at doing one thing really well. Trying to compare which is smarter is not really possible unless you compare apples to apples (i.e. is Deep Blue’s chess algorithm better than the algorithm in Chessmaster?).

The current litmus test for AI is called the Turing Test after Alan Turing (creator of the first computer). Basically you sit in a room at a keyboard and type in questions which are responded to either by a human or a computer in another room (the person asking questions doesn’t know which). If the computer can fool a human into thinking they are talking to another human than that computer is said to have passed the Turing Test.

This test is performed as a sort of a competition yearly and to my knowledge no computer has fooled a human although some humans were mistaken for a computer (a Shakespeare expert was thought to be TOO knowledgeable such that people figured only a computer could keep that much info straight).

From the Artificial Intelligence FAQ, What has AI accomplished?

There is also a list of articles that may interest you at the “American Association
for Artificial Intelligence” site.
AI Overview

So far passing the “Turing Test” is out of the question even for the most advanced “Artificial Intelligence” machines. Comparing computers built for different purposes would be difficult at best. How can you tell if the computerized Leukemia diagnosis system is smarter than the a dishwasher’s fuzzy controller?

If by intelligent you mean “human like” AI it’s still neo-natal. Even the best AI programs extant would have trouble out thinking a gold fish in terms of dealing with random, adaptive, effective problem solving. AI in a more vertical sense (and I do think you can still use the word “intelligence” in this vertical context), however, is extremely powerful re ‘Kasparov v. Deep Blue Chess Match’ etc.

You might want to check out The Age of Spiritual Machines by Ray Kurzweil. He addresses what constitutes intelligence, and in what way we can design machines that will perform the “intelligence” work of humans (recursive routines, neural-network-based pattern recognition, etc.).

Interestingly, he postulates that most of the tasks which are seen by us as intelligent will be proven to require very little intelligence. It’s a very interesting book.

Also, he mentions that most people’s experience with AI is fairly limited to commercial products, and they they are way impressed by what is actually state of the art. Here at IBM, I know that there is research in speech recongition that is light-years beyond what is actually available as boxed software.

Of course, there’s some “intelligent” tasks at which computers excell. A computer has no problem, for instance, multiplying 17-digit numbers accurately. It might even be possible to program a computer to parse some word problems, and it could certainly have a built-in dictionary/thesaurus to look up words. You could probably build a computer to get a half-decent score on, say, the SAT. How meaningful would this be, though?

The Creatures games are another thing that makes a claim to IA, but really isn’t. Even with the extra tinkering, the creatures probably would only have an IQ of 30 or so in people terms(they really dig bouncing face first into walls). I guess in games, AI is a loose definition at best.

Chronos sdaid:

Well, this is sort of part of the whole isssue, isn’t it? What do we consider to be an intelligent task. One of the hurdles AI has faced in recent years is the moving target of intelligence. When we beat Kasparov with Deep Blue, some were heard to say that all it proved was that chess requires no intelligence. We have to define our terms before we can evaluate computer progress toward that goal.

I believe this was accomplished back in the 1970s.

Meaningful in what way? What would impress you? “Consciousness”? How do you define that? I’m of the mind that consciousness is a big illusion created by a very complex machine with many, many simultaneous connections being able to examine its own processes on a high level, without being able to understand the low-level workings. YMMV.