|
|
 | | From: | Kent Paul Dolan | | Subject: | Re: AI will never work in 100 years !!!! | | Date: | 23 Jan 2005 13:38:11 -0800 |
|
|
 | "nobody" wrote:
> Prooving theorems fundementally takes no more
> "intelligence" than your basic string search and
> replace. Such systems are very restricted. That
> computers can play chess or prove theorems doesn't
> show the level of advancement of AI research but
> that such tasks are in fact mechanical in nature.
Unfortunately you have fallen victim to the "let's
move the goalposts" syndrome.
_Of course_ there will never be AI if every time
some researcher achieves some AI goal, it is
promptly redefined to be something "mere computers"
can do, and therefore merely "mechanical in nature".
Beating the world's best checkers player,
challenging the worlds best chess players, solving
theorems that have baffled humans for generations,
designing aircraft engine turbines better than the
best human design, designing truss bridges better
than the best human engineering designs, scheduling
workflows better than the best expediter, playing a
competitive game of soccer, are all things computers
have already achieved, all examples of successful
AI, except when goalpost movers redefine AI to
exclude them as soon as they are accomplished, lest
their contention that AI is valueless be noticed to
be an invalid and hollow claim, full of species
chauvanism and signifying nothing.
_Before_ each of those was achieved, it was
considered to be an AI goal of great significance.
Unfortunately for your thesis, each still is, and
AI is splendidly successful.
FWIW
xanthian.
|
|
| | From: | nobody | | Subject: | Re: AI will never work in 100 years !!!! | | Date: | Mon, 24 Jan 2005 10:41:34 GMT |
|
|
| "Kent Paul Dolan" wrote:
>"nobody" wrote:
>Unfortunately you have fallen victim to the "let's
>move the goalposts" syndrome.
>
>_Of course_ there will never be AI if every time
>some researcher achieves some AI goal, it is
>promptly redefined to be something "mere computers"
>can do, and therefore merely "mechanical in nature".
>
>Beating the world's best checkers player,
>challenging the worlds best chess players, solving
>theorems that have baffled humans for generations,
>designing aircraft engine turbines better than the
>best human design, designing truss bridges better
>than the best human engineering designs, scheduling
>workflows better than the best expediter, playing a
>competitive game of soccer, are all things computers
>have already achieved, all examples of successful
>AI, except when goalpost movers redefine AI to
>exclude them as soon as they are accomplished, lest
>their contention that AI is valueless be noticed to
>be an invalid and hollow claim, full of species
>chauvanism and signifying nothing.
>
>_Before_ each of those was achieved, it was
>considered to be an AI goal of great significance.
You could also add "cracking codes faster and more reliably than any
human could" to the list . Numerical solutions to optimization
problems were known and used long before AI or even computers came
into being. It's just that as the speed of the hardware increased and
these algorithms were refined, computers can tackle more and more
complicated problems in a timeframe that's practical. Designing
aircraft engine turbines better than the best human design is merely a
scaled up version of finding the minimum of a simple polynomial
function and nobody ever thought software would need to be
particularly intelligent to do that, just well designed. It's not a
matter of moving goalposts but defining them. Since the 70's,
goalposts have been defined lower and lower while the power of the
hardware has grown exponentially. Specialized solutions are fine and
it's no secret that a number cruncher can optimize FEA or simulation
far better than a human.
The problem of AI is that interaction with "reality" is the
fundemental piece of the puzzle. Systems restricted to particular
structured IO will never be able to adapt. Unfortunately, current
computer architecture doesn't allow that.
|
|
|
