According to an article
in TheStar Online, the problem of artificial intelligence has been
solved. Within 5 to 10 years, Terminators, Mr. Data, and other intelligent
robots will be a common part of daily life. Where have we heard this
before? This time the problem was solved by Sethuraman
Muthuraman, a Malaysian PhD student at Robert Gordon University who
has made a "major breakthrough" of "world significance"
that involves the application of evolutionary algorithms to artificial
neural networks. Like most folks who've solved the problem of AI, the
only thing missing is, well, the AI. The fine print says what they've
got is an ANN microchip that might "evolve in a modular fashion until
real intelligence emerged". General information on the school's AI
research program can be found on their website.
A more detailed explanation of Muthuraman's technique can be found in
his research paper titled, "The Development of
Modular Evolutionary Networks for Quadrupedal Locomotion" (PDF format).
I'd like to hear how these new "AI" solutions can be shown to be
successful in a real application.
I'm aware of the Turing test, but are there any "baby step" tests for
If not, I'm confident that we can look forward to many more
announcements like, "I solved AI! The proof (and application) shall be
left to the eager student!"
Gosh. Still no sign eh?
Still. At least this "solution" of AI isn't a totally mental one.
Modular evolution of ANNs seems perfectly reasonable. I even think I
may have read some papers by these folks.
As always, the question is why did this make it to the papers?
Yes, their approach actually sounds pretty interesting. I suspect the
media is partly responsible for the "AI Solved" thing. A headline like
"Amazing AI breakthough" just sounds cooler than "Interesting approach
to AI investigated".
Here's a link to an introductory article on their approach:
and Devolved Action: towards the evolution of systems (PDF format)
Ahem... AI solved, huh? I've heard that claim before. In fact, I've
made that claim
before. BTDT (been there, done that).
But did I, like Fermat, mention the solution in the margin of a
manuscript? No, I programmed it in Win32Forth. Now I have a simple request for robotics club
members, high school teachers, and neuroscientists: Download and run
Mind.Forth available on-line as http://mind.sourceforge.
net/mind4th.html and demo it to an individual, a class, or
yourself. Ask your audience whether they think that they are witnessing
real AI Mind. Let them use the new feature of pressing the Tab key to
switch to Tutorial mode so that they can watch the Robot AI Mind do its
internal thinking. If they are programmers, let them Tab even further
into Diagnostic mode and see if they accept the challenge of improving
Mind.Forth or of porting the AI to a
more modern language than Forth.
Here is a chance to make your
name -- either by debunking AI4U/Mentifex -- or by hosting and
MIND.FORTH IMMORTAL ARTIFICIAL INTELLIGENCE
FOR HUMANOID ROBOT CITIZENS
-- possibly the longest-running, oldest
artificial intelligence on Earth