By Eric Vandebroeck
and co-workers
Calling it, a Cargo Cult Wired was one of the first
high profile publications that pointed out
specific misconceptions surrounding AI, including that
intelligence is a single dimension. Most technical people tend to graph
intelligence how Nick
Bostrom does in his book, Superintelligence, as a literal,
single-dimension, linear graph of increasing amplitude. At one end is the low
intelligence of a small animal; at the other end is the high intelligence, of,
say, a genius, almost as if intelligence were a sound level in decibels. Of
course, it is then straightforward to imagine the extension so that
intelligence's loudness continues to grow, eventually to exceed our own high
intelligence and become a super-loud intelligence, a roar! Way beyond us, and
maybe even off the chart.
To which
Forbes added, we have fallen into this trap before. We have tried to turn
machines into automatons that mimic human behavior; these attempts failed.
Machines can do incredible things, but we need to design them in a way that
fits their strengths. Today, we struggle with building autonomous cars,
intelligent analysis tools, complex computer vision systems, but we are not the
first to automate our world. Rewind to the early 1800s, and the challenge of
the day was how to build a sewing machine. The world population was increasing,
and walking naked was not an option. There were just not enough free hands to
make the clothes to meet demand. We had to come up with a better solution for
making garments.
Extensive analysis
about AI was put out by UNESCO when it aptly stated that the success of
the term AI is sometimes
based on a misunderstanding,
Or more recently, MDN
explains how AI does not mean data will be fixed automatically, more data
will produce better outcomes, or that
the AI is ready to create out-of-the-box solutions.
Of course, the Cargo
cult idea is not entirely
true; AI is possible. As to that, the future of AI is a scientific unknown. The
myth of artificial intelligence is that its arrival is inevitable, and only a
matter of time, that we have already embarked on the path that will lead to
human-level AI and then superintelligence. We have not.
The path exists only in our imaginations. Yet the inevitability of AI is so
ingrained in popular discussion, promoted by media pundits, thought leaders
like Elon Musk, and even many AI scientists (though certainly not all), that
arguing against it is often taken as a form of Luddism, or at the very least a
shortsighted view of the future of technology and a dangerous failure to
prepare for a world of intelligent machines.
The science of AI has
uncovered a huge mystery at the heart of intelligence, which no one currently
knows how to solve. Proponents of AI have huge incentives
to minimize its known limitations. After all, AI is big business, and it's
increasingly dominant in culture. Yet, the possibilities for future AI systems
are limited by what we currently know about intelligence's nature, whether we
like it or not. And here we should say it directly: all evidence suggests that
human and machine intelligence are radically different. The myth of AI insists
that the differences are only temporary and that more powerful systems will
eventually erase them. Futurists like Ray
Kurzweil and philosopher Nick Bostrom, prominent
purveyors of the myth, talk not only as if human-level Al were inevitable but
also as if, soon after its arrival, superintelligent
machines would leave us far behind. The scientific part of the myth
assumes that we need only keep “chipping away” at the challenge of general
intelligence by making progress on narrow intelligence feats, like playing
games or recognizing images. This is a profound mistake: success on narrow
applications gets us not one step closer to general intelligence. The
inferences that systems require for general intelligence, read a newspaper,
hold a basic conversation, or become a helpmeet like Rosie the
Robot in Tlic Jetsons, cannot be
programmed, learned, or engineered with our current knowledge of Al. As we
successfully apply simpler, narrow versions of intelligence that benefit from
faster computers and lots of data, we are not making incremental progress.
Rather, picking low-hanging fruit, Hie jumps to general “common sense” is
completely different. There’s no known path from one to the other. No algorithm
exists for general intelligence. And we have good reason to be skeptical that
such an algorithm will emerge through further efforts on deep learning systems
or any other approach popular today. Much more likely, it will require a major
scientific breakthrough, and no one currently has the slightest idea what such
a breakthrough would even look like, let alone the details of getting to it.
Mythology about Al is
bad because it covers up a scientific mystery in endless talk of ongoing
progress. The myth props up belief in inevitable success, but genuine respect
for science should bring us back to the drawing board. Pursuing why this
is not a good way to follow the smart money or even a neutral stance. It is bad
for science, and it is bad for us. Why? One reason is that we are unlikely
to get innovation if we choose to ignore a core mystery rather than face up to
it. A healthy culture for innovation emphasizes exploring unknowns, not hyping
extensions of existing methods, especially when these methods are inadequate
to take us much further. Mythology about inevitable success in AI tends to
extinguish the very culture of invention necessary for real progress, with or
without human-level Al. The myth also encourages resignation to the creep of a
machine-land, where the genuine invention is sidelined in favor of futuristic
talk advocating current approaches, often from entrenched interests.
While we cannot prove
that AI overlords will not one day appear, we can give you a reason to discount
the prospects of that scenario seriously. For example, AI culture has
simplified ideas about people while expanding ideas about technology; This
began with its founder, Alan Turing, and involved understandable but
unfortunate simplifications one could call intelligence errors. Initial errors
were magnified into an ideology by Turing’s friend and statistician, I. J. Good
introduced the idea of 'ultra
intelligence' as the predictable result once human-level AI had been
achieved.
Between Turing and Good,
we see the modern myth of AI take shape. Its development has landed us in an
era of what I call technological kitsch, cheap imitations of deeper ideas that
cut off intelligent engagement and weaken our culture. Kitsch tells us how to
think and how to feel. Tire purveyors of kitsch benefit, while the consumers of
kitsch experience a loss. They, we, end up in a shallow world.
The only type of
inference, thinking, in other words, that will work for human-level AI (or
anything even close to it) is the one we don't know how to program or engineer.
The inference problem goes to the heart of the AI debate because it deals
directly with intelligence, in people or machines. Our knowledge of the various
inference types dates back to Aristotle and other ancient Greeks and has been
developed in logic and mathematics. The inference is already described using
formal, symbolic systems like computer programs, so a very dear view of the
project of engineering intelligence can be gained by exploring inference. There
are three types. Classic AI explored one (deduction), modern AI explores
another (induction). The third type (abduction) makes for general intelligence,
and, surprise, no one is working on it at all. Finally, since each type of
inference is distinct, meaning, one type cannot be reduced to another, we know
that failure to build AI systems using the type of inference undergirding
general intelligence will fail to progress toward artificial general
intelligence or
AGI.
The myth one could
argue has terrible consequences if taken seriously because it subverts science.
In particular, it erodes a culture of human intelligence and invention, which
is necessary for the very breakthroughs we will need to understand our own
future. Data science (the application of a to “big data" is at best a
prosthetic for human ingenuity, which if used correctly can help us deal with
our modern data deluge." If used as a replacement for individual
intelligence, it tends to chew up investment without delivering results. We
explain, in particular, how the myth has negatively affected research in
neuroscience, among other recent scientific pursuits. The price we are paying
for the myth is too high. Since we have no good scientific reason to believe
the myth is true, and every reason to reject it for our own future flourishing,
we need to rethink the discussion about Al radically.
Turing had made his
reputation as a mathematician long before he began writing
about A I. In 1936; he published a short mathematical paper
on the precise meaning of "computer," which at the time referred to
a person working through a sequence of steps to get a definite result (like
performing a calculation). In this paper, he replaced the human-computer
with the idea of a machine doing the same work. The paper ventured into
difficult mathematics. But in its treatment of machines, it did not
refer to human thinking or the mind. Machines can run automatically, Turing
said, and the problems they solve do not require any “external" help or
intelligence. This external intelligence, the human factor, is what
mathematicians sometimes call “intuition.”
Turing’s 1956 work on
computing machines helped launch computer science as a discipline and was an
important contribution to mathematical logic. Still, Turing apparently thought
that his early definition missed something essential. In fact, the same idea
of the mind or human faculties assisting problem-solving appeared two years
later in his Ph.D. thesis, a clever but ultimately unsuccessful attempt to
bypass the Austrian-born mathematical logician Kurt's result Gödel.
Though his
language is framed for specialists, Turing is pointing out the obvious:
mathematicians typically select problems or “see" an interesting problem
to work on using some capacity that at least seems indivisible into
steps, and therefore not obviously amenable to computer programming.
Gödel, too, was
thinking about mechanical intelligence. Like Turing, he was obsessed with the
distinction between ingenuity (mechanics) and intuition (mind). His distinction
was essentially the same as Turing's, in a different language: proof versus
truth (or “proof-theory" versus “model-theory" in mathematics lingo).
Are the concepts of proof and truth, Gödel wondered, in the end, the same? If
so, mathematics and even science itself might be understood purely mechanically. Human thinking,
in this view, would be mechanical, too. The concept of AI, though the term
remained to be coined, hovered above the question. Is the mind’s intuition,
ability to grasp truth and meaning reducible to a machine, to computation?
This was Gödel's
question. In answering it, he ran into a snag that would soon make him
world-famous. In 1931, Gödel published two theorems of mathematical logic known
as his incompleteness theorems. In them, he demonstrated the inherent
limitations of all formal mathematical systems. It was a brilliant stroke.
Gödel showed unmistakably that mathematics, all of mathematics, with
certain straightforward assumptions, is, strictly speaking, not mechanical or
for- realizable. More specifically, Gödel proved that there must
exist some statements in any formal (mathematical or computational) system that
are True, with capital-T standing, yet not provable in the system itself using
any of its rules. The True statement can be recognized by a human mind but is
(probably) not provable by the system it's formulated in.
How did Gödel reach
this conclusion? Hie details are complicated and technical, but Gödel’s basic
idea is to treat a mathematical system complicated enough to do addition as a
system of meaning, almost like a natural language such as English or German.
The same applies to all more complicated systems. By treating it this way, we
enable the system to talk about itself. It can say about itself, for instance,
that it has certain limitations. This was Gödel's insight.
Formal systems like
those in mathematics allow for the precise expression of truth and
falsehood. Typically, we establish a truth by using the tools of proof, we use
rules to prove something, so we know it’s definitely true. But are there true
statements that can't be proven? Can the mind know things the system
cannot? In the simple arithmetic case, we express truths by writing equations
like “2 + 1 = 4." Ordinary equations are true statements in the system of
arithmetic, and they are provable using the rules of arithmetic. Here, provable
equals true. Mathematicians before Gödel thought all of the mathematics had
this property. This implied that machines could crank out all truths in
different mathematical systems by simply applying the rules correctly. It’s a
beautiful idea. It's just not true.
Gödel hit upon the
rare but powerful property of self-reference. Mathematical versions of
self-referring expressions, such as “This statement is not provable in this
system,” can be constructed without breaking the mathematical systems' rules.
But the so-called self-referring “Gödel statements” introduce contradictions
into mathematics: if they are true, they are improvable. If they are false,
then because they say they are improvable, they are actually true. True means
false, and false means true, a contradiction.
Going back to the
concept of intuition, we humans can see that the Gödel statement is, in fact,
true, but because of Gödel's result, we also know that the rules of the system
can’t prove it, the system is in effect blind to something not covered by its
rules. Truth and provability pull apart. Perhaps mind and machine do, as well.
The purely formal system has limits, at any rate. It cannot prove in its own
language something true. In other words, we can see something that the computer
cannot.
Gödel’s result dealt
a massive blow to a popular idea at the time that all of mathematics could be
converted into rule-based operations, cranking out
mathematical truths one by one. The Zeitgeist was formalism, not talk
of minds, spirits, souls, and the like. The formalist movement in mathematics
signaled a broader turn by intellectuals toward scientific materialism, and in
particular, logical positivism, a movement dedicated to eradicating traditional
metaphysics tike Platonism, with its abstract Forms that couldn't be observed
with the senses, and traditional notions in religion like the existence of God.
The world was turning to the idea of precision machines, in effect. And no one
took up the formalist cause as vigorously as the German mathematician David Hilbert.
At the outset of the twentieth
century (before Gödel), David Hilbert had issued a challenge to the
mathematical world: show that all of the mathematics rested on a secure
foundation. Hilbert's worry was understandable. If the purely formal rules of
mathematics can’t prove any truths, it's at least theoretically possible for
mathematics to disguise contradictions and nonsense. A contradiction buried
somewhere in mathematics ruins everything because, from a contradiction,
anything can be proven. Formalism then becomes useless.
Hilbert expressed all
formalists' dream to prove finally that mathematics is a closed system governed
only by rules. Truth is just "proof.’’ We acquire knowledge by simply
tracing the "code” of proof and confirming no rules were violated. The larger
dream, thinly disguised, was really a worldview, a picture of the universe as
itself a mechanism.
Condensing the history of the AI myth
The story of
artificial intelligence starts with the ideas of someone who had
immense human intelligence: the computer pioneer Alan Turing.
In 1950 Turing
published a provocative paper, “Computing Machinery
and Intelligence," about the possibility of intelligent
machines. The paper was bold, coming when computers were new and
unimpressive by today’s standards. Slow, heavy pieces of hardware sped up
scientific calculations like code-breaking. After much preparation, they could
be fed physical equations and initial conditions and crank out the radius of a nuclear
blast. IBM quickly grasped their potential for replacing humans doing
calculations for businesses, like updating spreadsheets. But viewing computers
as "thinking" took imagination.
Turing’s proposal was
based on a popular entertainment called the "imitation game," In the
original game, a man and a woman are hidden from view. A third person, the
interrogator, relays questions to one of them at a time and, by reading the
answers, attempts to determine which is the man and which the woman. 'The
twist is that the man has to try to deceive the interrogator while the woman
tries to assist him, making replies from either side suspect. Turing replaced
the man and woman with a computer and a human. Titus began what we now call the
Turing test: a computer and a human receive typed questions from a human
judge, and if the judge can't accurately identify which is the computer, the
computer wins. Turing argued that with such an outcome, we have no good reason
to define the machine as unintelligent, regardless of whether it is human or
not. Thus, the question of whether a machine has intelligence replaces the
question of whether it can truly think.
The Turing test is
actually tough; no computer has ever passed it. Turing, of course, didn't know
this long-term result in 1950; however, by replacing pesky philosophical
questions about “consciousness” and "thinking” with a test of observable
output, he encouraged the view of AI as a legitimate science with a
well-defined aim.
In the course of its
short existence, AI has undergone many changes. These can be summarized in six
stages.
First of all, in the
euphoria of AI’s origins and early successes, the researchers had given free
rein to their imagination, indulging in certain reckless pronouncements for
which they were heavily criticized later. For instance, in 1958, American
political scientist and economist Herbert A. Simon, who received the Nobel
Prize in Economic Sciences in 1978 – had declared that, within ten years,
machines would become world chess champions if they were not barred from
international competitions.
By the mid-1960s,
progress seemed to be slow in coming. A 10-year-old child beat a computer at a
chess game in 1965, and a report commissioned by the US Senate in 1966
described the intrinsic limitations of machine translation. AI got bad press
for about a decade.
The work went on
nevertheless, but the research was given a new direction. It focused on the
psychology of memory and the mechanisms of understanding, with attempts to
simulate these on computers – and the role of knowledge in reasoning. This gave
rise to techniques for the semantic
representation of knowledge,
which developed considerably in the mid-1970s and led to expert systems
development, so-called because they use skilled specialists to reproduce their
thought processes. Expert systems raised enormous hopes in the early 1980s with
many applications, including medical diagnosis.
Technical improvements
led to the development of machine learning algorithms, which allowed computers
to accumulate knowledge and automatically reprogram themselves using their own
experiences.
This led to
industrial applications' development (fingerprint identification, speech
recognition, etc.), where AI, computer science, artificial
life, and other
disciplines were combined to produce hybrid systems.
Starting in the late
1990s, AI was coupled with robotics and human-machine interfaces to produce
intelligent agents that suggested the presence of feelings and emotions. This
gave rise, among other things, to the calculation of emotions (affective
computing), which evaluates the reactions of subject feeling emotions and
reproduces them on a machine, and especially to the development of
conversational agents (chatbots).
Since 2010, machines'
power has made it possible to exploit enormous data quantities (big data) with deep learning techniques based on formal neural networks. A
range of very successful applications in several areas, including speech and
image recognition, natural language comprehension, and autonomous cars, leads
to an AI renaissance.
Some of the applications
Many achievements
using AI techniques surpass human capabilities, in 1997, a computer program
defeated the reigning world chess champion. More recently, in 2016, other
computer programs have beaten the world’s best Go [an ancient Chinese board
game] players and some top poker players. Computers are proving or helping to prove
mathematical theorems; knowledge is being automatically constructed from huge
masses of data, in terabytes (1012 bytes), or even petabytes
(1015 bytes), using machine learning techniques.
As a result, machines
can recognize speech and transcribe it, just like typists did in the past.
Computers can accurately identify faces or fingerprints from tens of millions
or understand texts written in natural languages. Using machine learning
techniques, cars drive themselves; machines are better than dermatologists at
diagnosing melanomas using photographs of skin moles taken with mobile phone
cameras; robots are fighting wars instead of humans, and factory production lines
are becoming increasingly automated.
Scientists are also
using AI techniques to determine certain biological macromolecules' function,
especially proteins and genomes, from their constituents' sequences, amino
acids for proteins, bases for genomes. All the sciences are undergoing a major
epistemological rupture with in silico experiments. They are named so
because they are carried out by computers from massive quantities of data,
using powerful processors whose cores are silicon. In this way, they differ
from in vivo experiments performed on living matter, and above all, from
in vitro experiments carried out in glass test tubes.
Today, AI
applications affect almost all activity fields, particularly in banking,
insurance, health, and defense sectors. Several routine tasks are now
automated, transforming many trades and eventually eliminating some.
The point is that AI
requires improvements in historically confounding areas, like the link between
the brain and intelligence. Progress in areas with historical growth (such as
processing power and neural network size) does not guarantee ultra-intelligence
at all.
Betting against a
catastrophic end to the world has been right so far. On the other hand is Elan
Musk, Kurzweil, et al. and they have some impressive street cred. After all,
Kurzweil rightly predicted the successful timeline of the human genome project,
and not many would bet against Musk’s engineering teams achieve.
For updates click homepage here