The
stubborn myth of intelligent machines
The opinion that
computers will become more intelligent than man and even take command of us is
neither technically nor philosophically likely, so why are so many of us frightened
by these science fiction speculations? If there is something to worry about, it
is rather the moral consequences of this superstition about technology.
In 1936, 24-year-old
mathematician Alan Turing published the thesis entitled "Computable
numbers and the problem of decisiveness". It aimed at trying to find a
logical method to determine whether an arbitrary mathematical statement was
true or false. Touring's conclusion was: There is no such universal procedure.
In order to prove this, he in theory designed a "universal machine",
later known as the "Turing machine". It is today seen as the
theoretical foundation of the computer science.
In the next decade,
development took tremendous steps. Turing himself was recruited to the secret
Bletchley Park project that built the machine that broke the Germans' Enigma
code. Already in 1948, mathematician John von Neumann proclaimed that computers
would soon surpass human intelligence. Two years later, Turing published an
article in which he claimed that machines would soon think like humans. To
determine when this happens, he imagined an "imitation game" - later
known as the "Turing Test" - where a man may communicate with a
machine and a human being without knowing who is who. When it is no longer
possible for him to determine the difference, the machine is intelligent. A few years
later John McCarthy coined the term "artificial
intelligence". The course was now set for a project of cosmic dimensions: Man was about to recreate the ability that through the millennia of philosophy
and theology was regarded as his most unique feature and even the bond to God: The intellect of humanity.
This also raised the question: What does this discovery mean
for the future of man? The literary and artistic answers arrived instantaneously.
Yes, in fact, the answers had already been prepared for a long time. In modern
literature, it became Mary Shelley's "Frankenstein" (1818) who
captured the fascination and horror of the technological era facing artificial
life. Victorian author Samuel Butler's book "Erewhon" (1872)
describes how a new breed of machinery in Darwin's spirit achieves
consciousness and takes over the world. But it was the Czech writer Karol Capek who in his play "R.U.R" from 1920 came to mint the term "robot" as the name of a machine creature who revolts against his masters (from a slavic word for serfs workers).
The technical
breakthroughs from the 40s and onwards gave the fantasies new momentum, also
among the researchers themselves. A fascinating person is Irving John Good,
colleague of Turing in Bletchley Park. In 1965 he published an article in The
New Scientist where he imagined the emergence of an "ultra intelligent
machine" that could generate new machines. In each machine generation, it
would expand its abilities, which would quickly result in an "intelligence
explosion," an event that von Neumann already referred to as the
"singularity".
When Stanley Kubrick and Arthur Clarke
together created the film "Year 2001 - a space adventure" (1968), they
used Good as scientific advisor. The movie’s (and the book's) supercomputer Hal
9000 came to be the paradigmatic design of a machine intelligence that, at a
crucial moment, does not obey man but turns against him.
Among philosophers, however, there were voices that
questioned not only the concrete predictions, but also the very idea of
intelligence within AI. One pioneer was
Martin Heidegger, who was early interested in cybernetics. In an influential
essay from 1953, "The essence of technology", he argued for how our
understanding of technology leads to a technical understanding of ourselves.
When everything is transformed into "information", we no longer see
what it means to exist in the world as finite historical beings. His way of
reasoning was partly the basis of the American philosopher Hubert Dreyfus
"What computers can't do" from 1972, which shows how AI research rests on an narrow image of intelligence that it is
merely abstract symbolism, without connection to body and life context.
However, it was not Dreyfus, but his colleague at Berkeley
John Searle who, above all, was associated with the
philosophical criticism of AI. In explicit polemics against
Turing's criterion, he meant that there is no reason to
attribute human characteristics to machines, other than
in metaphorical terms. A computer may perform fantastic
chess moves, accurate translations, or sensible responses
in a conversation, but ultimately no one knows (There is
no one there) who is involved in any of these activities.
Searle's criticism was about the basis on which we can
define someone or something as intelligent. For the
majority of practically minded engineers and computer
scientists, these issues were irrelevant. When it comes
to designing a machine that can do one or the other,
translate, diagnose, drive or play chess, it is about
delivering results. That the criticism nevertheless
provoked AI research shows that many of its representatives
wanted to feel that they were close to understanding
human intelligence, sometimes for genuine intellectual
reasons but probably also for economic reasons. To claim
that AI has a philosophical relevance for our
self-understanding or that it is about to produce a new
higher silicon-based life-style creates attention and
attracts money. That Ray Kurzweil, development manager
at Google, in his acclaimed book from 2004, "The Singularity
is Near", devotes several pages to trying to refute
Searle's arguments testifying to what is at stake.
During
the 1980s, new models were tried to improve the function of the machines,
especially so-called neural networks, which one thought better imitated the
function of the human brain. But even if the machine's performance increased,
the major breakthroughs and commercial successes did not materialize. In
retrospect, the 80s and 90s appear as two decades of no progress for AI. The
process is described in detail in the excellent overview book that was compiled
last year by the New Scientist magazine, entitled "Machines that think: Everything you need to know about the coming
age of artificial intelligence".
When IBM's Deep Blue defeated Kasparov in
1997, it was certainly daunting for the chess world and it helped to draw
attention to research again. But no one
could claim that the program was "intelligent" in any significant
sense. It lacked strategy and ability to learn or draw conclusions and it was
inconvenient for all other activities than to calculate chess moves.
Today, everyone is again talking about AI. It has its prophets,
evangelists as well as doomesday preachers. Designs of mutually benevolent,
sometimes threatening, intelligent robots are repeated in literature and film.
The commercial interest is hot.
What happened? The explanation holds a
philosophical screw. As long as AI was believed to recreate or explain human
intelligence, it arrived nowhere. The
real breakthrough came when we gave up the idea of building models of human
thinking and instead invested in creating machines that could "learn
themselves" by statistically processing huge amounts of growing data via
simpler algorithms.
The idea is perhaps best illustrated by the translation
programs. They were promised already in the 50's and great efforts have been
made to create algorithms for how an ideal human translator works, but in vain.
Instead, the breakthrough came in the 21st century through the creation of
programs that, based on a huge and ever-growing text database, generate a
statistically calculated probability that a certain construction in the target
language corresponds to that in the source language. These programs testify to
the impressive intelligence of their creators. But no one can reasonably argue
that these machines understand the languages they handle or that they would
be intelligent.
Nevertheless, dramatic scenarios of how intelligent machines
are about to take over the world are repeated. Among the
most renowned researchers in the genre, are two Swedes,
the philosopher Nick Bostrom and the physicist and
mathematician Max Tegmark. Their best-selling "Super
Intelligence" (2014) and "Life 3.0" (2017) both assume
that today's self-learning machines can generate an
"intelligence explosion" in the near future. By referring
to the brain's extremely complex structure of neurons and
synapses, they mean that super intelligence is only a matter
of when the machines approach a comparable complexity in
computational capacity. But how and why today's highly
specialized programs according to some kind of evolutionary
logic would generate a super brain with its own intentions
remains to be shown. Based on the technology we have today,
it is neither technically nor philosophically likely. It is
one of many "esoteric possibilities" to quote The New
Scientist, who believes that the whole field is in great
need of a "reality check". It does not prevent the threats
from having a deep aesthetic and religious sigh, which
probably explains their impact.
The apocalyptic 50's visions are now dusted off in the
light of new technological successes that ultimately risk
diverting attention from more urgent issues. One of them
is the political and ethical consequences of the imminent
robotisation of both work and private life. Another is
the control of the technology.
The country that focuses most on AI right now is China,
a communist dictatorship that obviously also sees it as
an instrument of economic and political governance.
Finally, it is important to also keep the issue alive
that Heidegger once posed, namely what the origin of
mechanical intelligence does to man's self-understanding.
An increasing expert and consulting culture is already
contributing to a weakening of the individual's ethical
and professional responsibility. If man believes himself
to be in the process of being replaced by machines that
are more intelligent than he himself, he risks more
easily renouncing responsibility and judgment.
Rather than worrying about "super-intelligence", we might
have to worry about the "super-stupidity" that threatens
when man no longer thinks he needs to think because he
believes his tools does it for him.
Todde
Also check:
