3 ms·
>> AI agents will control the future, and which ones we create is the only thing about our time that will matter in the long run. Every time there's a big stor
by YeGoblynQueenne 6y ago
>> AI agents will control the future, and which ones we create is the only thing about our time that will matter in the long run.
Every time there's a big story in the news about a new thing achieved by an AI
system, someone will lose their shit and go on social media to say how it's a
sign of the end times where Terminators will roam the Earth and humans will hide
in caves. At least the article has the common indicency of placing this
inevitable eventuality "in the future", with all the precision typical of palm
readers and popular entertainers.
For perspective, here's an article I found by following the wikipedia links on
Gary Kasparov's 1997 defeat by Deep Blue in chess:
https://www.washingtonexaminer.com/weekly-standard/be-afraid-9802 https://www.washingtonexaminer.com/weekly-standard/be-afraid...
The article is titled "Be Afraid" (!) and it includes passages like the
following:
In Game Two, Deep Blue passed the Turing test. Yes, of course, it was for
chess only, a very big caveat. But, first, no one was ever quite sure that a
machine ever would pass even this limited test.
And second, if a computer has passed the Turing test for chess, closed logical
system though it may be, that opens the possibility that computers might in
time pass the Turing test in other areas.
Besides the typical abuse of the Turing test to mean anything anyone wants it to
mean, as long as it means something as vague and fuzzy as it is irrelevant, I
remind the esteemed reader that the year is 1997 and people are raising the
alarm about AI going superintelligent and taking over the world a long time
before Eliezer Yudkowski or Elon Musk and OpenAI. And if you look further back,
you'll find similar predictions of AI being "just around the corner" since 1957.
Maybe we need to build guns that can shoot behind corners, to protect ourselves
from evil superintelligences, eh?
So hold your horses and don't lose your mind. AI today is just as limited as it
was in 1997 and in 1957. We have more data and bigger computers but we still
don't understand intelligence and our best bet to reproduce it on a digital
computer remains the hope that it will somehow arise, magickally, unexpectedly,
on its own, fully formed like Athena from the brow of Zeus, if only we throw
more big data and more big computers at narrow tasks in irrelevant domains- like
chess, Atari, Dota 2... or, indeed, protein structure prediction (not quite a
hallmark of intelligence that).