"Roko" serves up the usual premature dismissal:
If you have a fact-based argument as to why smarter than human AI is not possible then please tell me.
Just what assumptions and frames are embedded in your notion of "smarter" here, and are the implications of those assumptions matters of fact? Are differences arising from these assumptions open to adjudication on the basis of what you consider to be facts?
People who have trouble distinguishing science fiction from science should be less cocksure that they always have the facts on their side, and that their skeptics are always ignorant or irrational.
Is "smartness" a matter of instrumental or formulaic calculation, are sensitivity, imagination, improvisation, criticism, expressivity dimensions contained in your notion of "smarter than human AI"?
Does it matter or not to your visions of post-biological smartness that intelligence has always only been materialized in brains, does it matter that performances of intelligence are always social, and that in some construals collaboration is already a form of greater-than-personal intelligence?
If not, why not? At what point is the trait you claim to be so palpably possible sufficiently remote from the actual phenomena denoted by the term "intelligence" that you might properly be compelled (by the demands of sense, I mean) to find some other word for what you are talking about?
What are the stakes of your attribution of "possibility" to the "arrival" of this smartness, whatever you happen to mean by it? Is it logical possibility? Is it theoretical possibility, however well-substantiated or not, however remote or not? Is it proximate practical possibility capable of attracting investment capital or demanding immediate regulation?
Do these distinctions figure at all in your determination of whether or not this question of engineering "smarter than human AI" is worthy of serious consideration?
If not, why not? Wouldn't these sorts of distinctions figure in most practical considerations of the kind you seem to think you are engaging in?
If you want to sell what looks to me like a faith-based initiative concerning the arrival of post-biological "superintelligence" you'll discover that skeptics you want to persuade don't have to meet your terms, you have to meet ours. It's the extraordinary claim that demands the extraordinary substantiation.
Your personal challenge to me is finally irrelevant, of course, since the challenge of scientific consensus is the one that confronts your claim and so far you have failed to attract that consensus. You may be able to find a cul-de-sac in which your claim passes muster for a marginal minority (that's the whole point of joining a Robot Cult, presumably), and you are surely able to best me or at any rate bamboozle me in some exchange on some technical matter I have neither the training nor the temperament to address the proper significance of, but all that is neither here nor there.
I pose my own challenges to you on the terms I am fit for, and those terms are relevant even if they are not the only relevant ones in a question like this, and even if you choose to demote them as not "fact based" and hence, apparently, unworthy of consideration. You'll discover that you live in a world with sufficiently many people in it who differ with you on the question of which concerns are the ones worthy of consideration that dismissals only ensure that you are dismissed. That, too, after all, is a fact.
> "Roko" serves up the usual premature dismissal:
ReplyDelete>
> > If you have a fact-based argument as to why smarter than
> > human AI is not possible then please tell me.
If you have a fact-based argument as to why the Flying Spaghetti
Monster (and his messenger, the Angel Macaroni) are **not possible**
then please tell me.
"Welcome to Roko Mijic's personal Homepage. I am a student of mathematics
and computer science with a keen interest in philosophy, ethics and futurism."
I guess a "keen interest in philosophy" doesn't get you very far
these days.
WOODROW WYATT: Do you think it is **certain** that there is
no such thing as God, or merely that it is not proven?
BERTRAND RUSSELL: I don't think it is **certain** that
there is no such thing as God, no. I think it's exactly on
the same level as the Olympic gods and the Norwegian gods --
the gods of Olympus and the gods of Valhalla -- they also
**may** exist; I can't prove they don't. But I don't think
the Christian god has any more likelihood than they do.
I think they're a **bare** possibility.
-- "Bertrand Russell Speaks His Mind"; Woodrow Wyatt interviews,
1959.
Gads this is beautiful.
ReplyDeleteIn my intro-level AI class, we spend a full 3 weeks JUST navigating claims of defining intelligence, and then we spend more or less the rest of the class filling in all the details traditionally left out of those accounts.
I'm tempted to point my whole class toward this post now that the semester is starting to wind down.
Shorter Dale: Define "intelligence." But definition problem aside, burden of proof is on those who claim "superintelligence" is possible, not the other way around.
ReplyDeleteAnd he is quite right. The problem is there's no such fact-based proof, just "reasonable extrapolations" (guesses), intuitions (influenced by the fact that "superintelligence" in one form or another is a common literary trope, but so are broom-flying witches), and philosophical inferences ("if it's material, it's buildable.") Really, that's all. If you think there are "scientific facts" in support of this, well, you either already have superintelligence you carefully observed and experimented with*, or you don't quite understand what sort of facts you can study using scientific methods.
* - of course, there might be prely theoretical proof of possibility of superintelligence, but that implies that you can at least perform much simplier task of exhaustively defining and making useful predictions of unobvious, verifiable, falsifiable, properties and behaviors of "merely human" intelligence. (Who again was that guy who mass-IQ tested scoolchildren in the 30s, and missed all 3 future Nobel Prize winners who participated? And just one of the thousand or so subjects who had highest scores did something notable, namely designed an Army ration widely used during WWII.) If you can't do much, much better than that IQ tester, then your theory is untested and thus worthless.
> Who again was that guy who mass-IQ tested scoolchildren in the 30s,
ReplyDelete> and missed all 3 future Nobel Prize winners who participated? And
> just one of the thousand or so subjects who had highest scores did
> something notable, namely designed an Army ration widely used
> during WWII.)
From _Broken Genius: The Rise and Fall of William Shockley, Creator
of the Electronic Age_ by Joel N. Shurkin
http://www.amazon.com/Broken-Genius-William-Shockley-Electronic/dp/0230551920
Chapter 1 "I've got dark eyes. I can frighten people."
p. 12
May [Shockley, William Shockley's mother] suffered one major disappointment
with her son. Since 1911, a Stanford psychology professor, Lewis Terman,
had been studying gifted children, hoping eventually to gather a sufficent
number of subjects he could follow through their lives to see how they
differed from other children. Terman hoped to prove -- at least initially --
that intelligence was genetic, and that the intellectually gifted
did better in life.
In 1916, he began testing hundreds of children in the Palo Alto, San Francisco
and Los Angeles areas using the Stanford-Binet IQ test he had recently
developed. Terman accepted as subjects only those children who scored
135 or higher, his definition of genius (100 being average). Teachers
initially selected who they thought were the two or three brightest
children in their classes.
It is not known how Bill was nominated but he was tested for the first
time at the age of eight, just before he entered public school, and scored
129. The next year he was retested and scored 125. (Having a small
decrease between tests was not uncommon.) He failed to make the cut.
He was still two standard deviations higher than average; he just was
not, according to Lewis Terman, a genius. Later in life Bill joked
often about how he could not qualify for Terman's gifted study, yet
could still win a Nobel Prize in physics. That he subsequently used
the same IQ tests as the basis for his unpopular beliefs about race
and intelligence never seemed to vex him, nor did the fact that he
was living proof the tests should not be taken too seriously.
The irony was lost on him.
Chapter 12, "Someday we may actually be terribly alone"
p. 229
Terman's kids were wonderful students. Eighty-five per cent skipped
grades at least once. Three-quarters of all their grades were As.
Fifty per cent could read before they got to school. When they
reached college -- and they did in unusually high numbers -- they
practically gobbled degrees. A third of the Termites admitted to
Stanford graduated Phi Beta Kappa. Terman found nothing in their
home lives (and his researchers inspected many homes) to suggest
the influence of environment to explain this success, he reported.
IQ is also a good predictor of success in life, at least by conventional
middle-class standards, and the Termites did splendidly, becoming
(at least for the males) doctors, lawyers, businessmen, and
scientists at a vastly higher rate than would be expected from the
general population. The 1,500 children grew up to produce at
least 2,500 scientific articles and papers, 200 books, more
than 400 short stories and 350 patents. And that didn't count
the output of the professional journalists. Terman was so proud
of them that his files bulge with their work. Three were
members of the National Academy (including his son, Fred); six
made the _International Who's Who_; 40 made _Who's Who in America_,
and 81 (including 12 women) made _American Men of Science_.
Terman's kids worked for the Federal Reserve, the Atomic
Energy Commission, the staff of the US Senate, the Department of
Justice, NASA and the United Nations. During the Second World
War, the men earned 90 valor medals, including 15 Purple Hearts.
By and large, they reported themselves to be happy people,
and they lived longer than the population average.
. . .[T]he Termites' children. . . also were exceptional, with a
mean IQ of 133, slightly less than their parents but still around
the 98th percentile. Sixteen per cent of the children of the
gifted were themselves gifted, an astonishing percentage -- in the
general population you would expect less than 1%. Less than 20%
of the second generation had IQs below 120, also an amazing
figure. Statistically, that is impossible to explain away.
Terman was convinced he was watching heredity.
Most of the critcism of Terman's results points out that the
children (and their children) started life with all the advantages
of prosperity and continued in that kind of atmosphere as they
grew up. But Terman included environment in his analysis, and
the premise is not entirely true to start with -- a substantial
number of Termites were not so financially advantaged, despite
where Terman found them. Environment seemed to make no difference
in their scores. It seemed to play the greatest role in his
analysis of why some of the gifted succeeded greatly in life
while others did not. It was unrelated to IQ.
A more valid criticism -- and this is crucial to understanding
the flaw in Shockley's argument too -- lies in what IQ did **not**
measure in Terman's study.
Most obviously, Terman missed the two Nobel laureates. Neither
Shockley nor Luis Alvarez had IQs above 135. Shockley was tested
twice and missed both times. Whatever talent they had went
unmeasured by Terman's questions. One hypothesis is that the
tests do not measure mathematical prowess very well, but is
that ability not a facet of what we mean by 'intelligence'?
One of the great mysteries of Shockley's story remains: how could
someone who was a living embodiment of the weakness of IQ tests
destroy his reputation on a theory based on their credibility?
Part of the answer may be merely reading the results with an
astigmatism, a lens bent to show what you want it to show; part
of it could be simply that neither Shockley nor anyone outside
the Stanford psych department -- with few exceptions -- had seen
the files, so they actually didn't know all that was in them.
For instance, Shockley often said the Termites won an uncommon
number of Nobel and Pulitzer Prizes. In fact, none of Terman's
kids won either prize. No one even came close.
Another gap is the issue of creativity, a deficiency that bothered
Terman so much that he sent his investigators to a school for
the arts in Los Angeles to test children, hoping to include that
variable. They all flunked. Terman found no link between IQ
and creativity, particularly in the arts. Except for Henry Cowell,
who was in a different study, none of the gifted subjects became
known as composers, musicians or artists, and there were only a
very few writers of any note: the science fiction author
L. Sprague de Camp and William A. P. White, who used the _nom de
plume_ Anthony Boucher. Only one, the actor Dennis O'Keefe, had
any reputation in theater or films, and another, Shelley Mydans,
was one of the few who earned note in journalism. The most famous
Termite was Jess Oppenheimer, the comedy writer who created
'I Love Lucy.'
. . .
The human quality problem and the controversy surrounding it became
a full-time obsession for Shockley. It released dark forces in him
that seemed always to have been lurking. Only by the greatest effort --
an effort he would rarely expend -- could he talk about or think about
anything else. Even when he visted Alison [his daughter] in Washington,
that was the topic of dinner conversation. Alison and her husband learned to
just sit and listen.
Shockley was oblivious to what he was doing to himself. There should
have been an apprehension at one point that he had gone too far with
his racial theories, a recognition that he had left a crucial opening
for his opponents. Yet there is no indication of it in his papers
and [his second wife] Emmy was quite clear he harbored no doubts, either
about what he was saying or the wisdom in saying it. He was impelled forward
by his own demons.
In the early 1970s, a Stanford psychiatrist told a reporter he thought
Shockley was suffering from the classic symptoms of paranoia. Indeed,
he began demonstrating many of the symptoms of what is now called
Paranoid Personality Disorder. Others have speculated that Shockley
was a high-functioning autistic or had Asperger's Syndrome, or that
he had obsessive-compulsive disorder. We'll never know.
Shockley had AT&T install recording devices on all his telephones --
home and office. Every conversation was recorded, and every one was
interrupted by a beeping sound every ten seconds. Conversations on each
cassette were separated by a countdown. Writer Rae Goodell, who researched
her PhD dissertation and subsequent book partly on Shockley,
recalls one conversation beginning: 'Goodell three, Goodell two, Goodell one,
Goodell zero. The time is now ten minutes to seven on Tuesday, the
twentieth. Goodell zero.' . . .
Every call, no matter how minor, was taped, indexed and stored. If the
Shockleys ordered take-out, the conversation is likely to be on tape
in his archives. If for some reason the tape recorder didn't work,
he would have someone, usually Emmy, listen on the line and take dictation.
They did this for every telephone conversation for the rest of Shockley's
life.
He said he thought the tape recorder was the single most important
application of the transistor. They made, he said, 'a profound difference
in honesty.' He also taped every personal conversation, usually --
but not always -- with the knowledge and consent of the person he
was speaking to. He could prove Henry Kissinger's aphorism that
even paranoids have enemies.
Chapter 9 "Really peculiar ideas about how to motivate people"
p. 163
Jim Gibbons walked into Shockley's office, sat across from him,
and was ready when Shockley pulled out a stopwatch.
'You have 127 players in a singles tennis elimination match,'
Shockley said. 'Obviously, you've got 63 matches and only 126
players can be in the first round so there's a 'bye. You can
put that next guy in so you have 64 people in the next round
and you have 64 matches. How many matches does it take
to determine a winner?
Click.
It was August, 1957. Jim Gibbons, a young physicist, like
every other new employee, had to take a little intelligence test.
Shockley knew perfectly well that Gibbons had a PhD from Stanford,
worked at Bell Labs and won a Fulbright scholarship to Cambridge
University -- a good sign he had something between his ears
besides lint. But everyone coming to work for Shockley Semiconductor
Co., had to take a battery of tests, either with Shockley
in Mountain View or with a New York testing agency.
Shockley had great faith in that kind of testing, feeling
increasingly that things like intelligence and creativity can
be quanitified. He had begun exploring their uses while
still at Bell Labs and became a firm believer. That the tests
had no real scientific basis never seemed to bother him.
Gibbons thought only a few seconds and said, 'Well, it must
be 126.'
Click. Shockley looked down at his stopwatch, his face reddening.
'What?'
'Well, it must be 126.'
'How did you do that?' Shockley asked, his agitation growing.
'There's only one winner and that means 126 people have to be
eliminated. It takes a match to eliminate somebody, so there must
be 126 matches,' said Gibbons assuredly.
Shockley pounded the table in fury.
'That's how I'd do it! Have you heard this problem before?' he
demanded.
'No sir,' said the young scientist, confounded at Shockley's reaction.
The Nobel laureate was coming unhinged.
Shockley gave him another problem, again clicking the stopwatch into
action. Gibbons thought about this one but could not figure out a
quick answer. As time elapsed and he said nothing, Shockley's face
returned to its normal color and he sat back. 'You could feel
the tension start relaxing,' Gibbons remembered later.
'That's enough, Jim. You're now at twice the average time for the
lab to solve the problem. Let me tell you how to do it,' Shockley
said, his equilibrium restored. Gibbons had missed the key.
'It was really tough for him, the fact I got the first one,' Gibbons
says. 'He'd set the damned thing up so you'd say 63+32+16+... --
you just don't sit there and say "126."
The possibility that this young man -- Gibbons was in his early 20s --
could compete clearly upset him. Gibbons only redeemed himself by
failing the second test. The thought that Gibbons might have been
as smart as he was ('Not even remotely close to being true,' Gibbons
said), seemed to frighten him. 'If I'd seen the next trick, the
guy would have been apoplectic,' Gibbons remembered.
Gibbons did well enough with the rest of the test and walked out
of Shockley's office to the laughter of the other researchers in the
building, all of whom had faced the same test.
p. 173
Shockley firmly believed that scientific advancement was the
result of a solitary genius or at most a small group of geniuses
who set the stage for an intelligent team of researchers
below them to break the necessary ground, a kind of trickle-down
creativity. The coterie of great minds running the Manhattan
Project stimulated the worker ants below them to great achievement.
He gave little credit to creativity from below.
Shockley had a model of how laboratories and institutions should
work taht very clearly involved that kind of hierarchy: The
workers take direction from above (their betters?) and progress
ensues. He, of all people, should have known better. The
invention of the point-contact transistor violated that model
(the men responsible worked for him without much direction),
and if he had paid attention to history he would have seen
that most innovation comes from motivated individuals, not teams
or hierarchical dictates. . .
In truth, he had no idea how to manage.
'He had some really peculiar ideas about how to motivate people,'
said [Gordon] Moore. 'First of all, he was extremely competitive
and controversial. If there were two ways of stating things,
one of which was controversial and one of which was straightforward,
he'd pick the controversial one every time. He just thrived
on stimulating controversy.' That stimulated conflict, not
originality. . .
He had trouble seeing people. . .
'He was very attractive to bright young people,' Terman later
explained, 'but was hard as hell to work for.' . . .
Shockley was often insulting, treating his employees the way
he treated his sons, with no glimmer of sensitivity. His
favorite crack, when he thought someone was wrong, was:
'Are you sure you have a PhD?'