We
live in an age where science has ceased to reside in a quaint ivory tower. Even
analytic philosophy took a quantitative turn and became inextricably entwined
with computer science, as Bertrand Russell diagnosed and predicted already in
1945:
"Modern analytical empiricism [...] differs from that of Locke, Berkeley, and Hume by its incorporation of mathematics and its development of a powerful logical technique. It is thus able, in regard to certain problems, to achieve definite answers, which have the quality of science rather than of philosophy. It has the advantage, in comparison with the philosophies of the system-builders, of being able to tackle its problems one at a time, instead of having to invent at one stroke a block theory of the whole universe. Its methods, in this respect, resemble those of science. I have no doubt that, in so far as philosophical knowledge is possible, it is by such methods that it must be sought; I have also no doubt that, by these methods, many ancient problems are completely soluble."[1]
While neuroscience has made some rather remarkable inroads
into neurobiology
and behavior [2],
it is in many other ways still in its infancy – yet artificial intelligence (A.I.)
purports to mimic and surpass the human brain on its way to a “singularity”
that Ray Kurzweil, Google’s director of engineering, predicts
to happen circa 2045. And Stephen Hawking cautions that creating artificial
intelligence might be the biggest and yet last event in human history “unless
we learn how to avoid the risks.”
Certainly, algorithms and machines are likely to make most
knowledge-based work more data-driven, more rational, and it is also highly
likely that the coming decades will see machines created that will perfect
elements of reasoning and execution of conclusions better than we can. But if
robots replace 80 or 90 percent of a profession – those engaged in largely
repetitive and routine (however advanced) functions – that is not saying that
there will be no doctors or no lawyers under such conditions. Not so long ago,
the number of people in agriculture dwindled from, say, 90 percent into the low
single digits, so the effect of technological change on employment patterns is
hardly new. It is important to remember the fact that many types of
professional services are capable of being assumed entirely by robotic
functions – with significant benefits in the overwhelming number of cases, if
we think of robotic hip replacement or bypass surgeries with an increasingly
negligible number of failures where the machine is confronted by limitations of
its design and knowledge base. As soon as the incidence of such robotic failings
is surpassed by data reporting human malpractice, the ethical aspect of the
discussion will turn moot.
Because we have not really nor entirely understood how human brains
work, we cannot duplicate, much less create artefacts that surpass them.
Playful companion robots capable of speaking accent-free and
faultless texts with synchronized moving lips that remarkably approximate
the human appearance are produced and becoming affordable at the price of a
high-end laptop in Japan today. While they are presently used as guides in
Tokyo’s National Museum of Emerging Science and Innovation Miraikan or as exotic TV
announcers, such androids will imminently cross the threshold to behavior that
will appear to humans as flirting – but just like it would be a mistake to
think that a robot translating “understands” language and is, in fact, “translating,”
robotic flirting is devoid of any emotional quality and merely conditioned to
respond to certain observations and stimuli. The design of android robots
requires profound exploration what it means to be human.
While the U.S. Armed Forces are developing battlefield robotics
including killer robots,
the Swiss École Polytechnique Fédérale’s Laboratory of Intelligent Systems in Lausanne
already in 2009 produced the first robots that, on their own motion, use
subterfuge and deceit to achieve an objective. In a series of
experiments, engineers Sara Mitri and Dario Floreano and evolutionary
biologist Laurent Keller divided 1,000 robots
into ten groups with the task of locating certain resources. If they did, they
were equipped with a blue light alerting other robots in their group of the
location. High points were given for locating and sitting on a good resource,
and points were distracted for doing the same on a no-good resource.
Furthermore, good resources were limited so that not every robot could score
when one was found, and overcrowding could dislocate the original finder. Each
robot had a sensor and its own 264-bit binary code controlling its behavior.
After each round of experiments, the 200 highest-scoring software “genomes”
were randomly “mated” – allowing for “mutations” of their software – simulating
machine evolution. “Robocopulation”
as a means of modeling evolution has become a very hot area in robotics with
considerable potential. It took all of nine generations to perfect their skills
at locating the target resource as well as inter-group communication. But the
truly interesting pattern took a little longer to emerge: 500 generations
later, 60 percent of robots had learned not to turn their blue light on when
the target resource was found, enabling them to preserve the benefits all to
themselves. One-third
of them learned to identify liars by declining to react to the (often
enough false-positive) blue light, in direct violation of their original
“genetic” software programming.
Now,
there is a whole lot to say about this experiment. First off, it would be
absurd to speak of “lying” in the case of a machine. There is little question
that human and robotic intelligence would evolve differently in an evolutionary
computation system developing outside direct human involvement. Equally, motive
and moral base, both characteristic elements of lying when defined as an
intentionally false statement, are decidedly absent in a machine. Deception in
order to achieve a goal is a common tactic in many games, but also in the
animal kingdom. While it promotes survival of the fittest, it cannot be
considered lying sensu proprio.
Because certain behavioral moves are assigned specific costs for a robot, the
machine does what it does best – maximize the score. The same concept has already
formed the basis of chess capabilities, one of the early accomplishments of AI.
Intention and moral judgment are clearly absent here – but are conditioned
human responses really that different from software assigning costs to certain
behavioral responses?
In
Wittgenstein’s example, “if
a lion could speak, we would not understand it.” Functionally desired
outcomes notwithstanding, the parallels between complex human behavior and
machine responses only go so far. But do they – and are humans not a similar,
as of the present state of the art more complex, computational cybernetic
system translating inputs into outputs?
Hawking’s
caution, while unquestionably meritorious, still implies the disappearance of
humanity “as we know it” – either because superior machine intelligence will
exceed not only individual human intelligence but also aggregate human
intelligence in the world, and will dispose of us as homo sapiens did of homo
neanderthalensis: we are, after all, unstable, war-prone, hyper-armed,
and a source of computer malware; or else natural intelligence will be improved
through artificial means by adding neuron circuits and other extensions for
speed. In either case, “human nature as we know it” will vanish – in the latter
case peacefully and with some legacy, just like the Neanderthal genes that
survived in us. Yes, as of today, digital A.I. needs to evolve by quantum leaps
to play in the league of analog intelligence of present-day people. But in only
a half-century, and largely without any of the accelerating quality of
contemporary computational resources, IT has come a long way few even at IBM had
imagined.
Without
discounting critics who point to the precedence of resolving present-day
threats to human existence before worrying about more distant eventualities,
the advent of the singularity is an inexorable and inevitable fact of evolution
of A.I., and it is ultimately specious to distort critical social dialog by
mocking one or the other side’s time projections. This is why topics of transhumanism
cannot be sidestepped in reality as they will affect, and change, all areas of
life already in the relatively near future, starting with an increasing
percentage of prostheses, implants and “spare parts” no one will suspect of
altering the human condition – while they most definitely do, just as gay
marriage did for the notion of marriage commonly accepted only a generation
ago, and as the digital revolution did for the notion of privacy.
Analogy
will have it that, in similar ways, A.I. will be characterized by the law of
unexpected/unintended consequences for the majority of those affected by it. One
of the most significant ways technology revolutionizes our existence is the incompleteness
of our predictive abilities. In 1931, logician Kurt Gödel stated in his
incompleteness theorem for any computable axiomatic system that, if the system
is consistent, it cannot be complete, and that consistency of its axioms cannot
be proved within such a system. If it were provable, it would be false,
which contradicts the idea that, in a consistent system, provable statements
are always true. Rather, there will always be at least one true but unprovable
or undecidable statement.[3]
The very same is true of our attempts to foresee and pre-empt or control the
consequences of cognitive and technological change, which will arguably be most
striking in the area of artificial intelligence.
[1] Russell, Bertrand. A History of Western Philosophy. New
York: Simon &
Schuster (1945) 834.
[2] Kandel, Eric R, Schwartz, James H., Jessel Thomas M., Principles of Neural Science (4th ed.). New York: McGraw-Hill
(2000) ISBN 0-8385-7701-6.
[3]
Hawking, Stephen (ed.). God
Created the Integers: The Mathematical Breakthroughs That Changed History. Philadelphia:
Running Press (2005), ISBN 0-7624-1922-9, 1097 et seq.
No comments:
Post a Comment