Overpromises
and Unrealistic pledges of AI
If we could
find a clue, the date people had started to think about a machine that can
think and get decisions alone, that history runs up to the ancient Greeks era.
They had myths about robots. But it was formally got into the picture in the
first half of the 20th century. Actually, AI wasn’t announced until 1956, at a
conference at Dartmouth College. This concept has been developed over longer
than half a decade and there were several hikes and drops in its propagation.
In this
letter I will discuss the winters of AI which was faced in past, the situation in the present and future prediction of winters. Although concepts such as
“knowledge-based systems” or “intelligent agents” were intervened in guaranteed
funding to the AI projects, some criticisms caused to AI caught into winters in
the past. AI had to face two winters throughout the journey. which occurred in
1974 – 1980 and 1987 – 1993 period. Even though the field collapsed without
funding and sponsorship from the government and other fund releasing parties,
researchers carried on the improvement despite the criticisms.
In 1973, the
UK Science Research Council published the Lighthill Report, which criticized
the utter failure of AI to reach its “grandiose objectives” and noted that “in
no part of the field have the discoveries made so far produced the major impact
that was then promised.” At the same time, Richard Karp proved 21 difficult
problems in computer science to be NP-complete, which led to the famous
unsolved P vs NP problem with a 1 million US dollars reward prize. This pointed
the problem of “combinatorial explosion”, where the computing time required to
solve the problem increased exponentially as a function of the input size. This
meant that it was unable to scale up any of the AI solutions into useful
real-life applications with the available resources.
In Moravec’s
Paradox, limitation stated that “It is relatively easy to make computers to
show results on intelligence tests, playing checkers or calculating pi to a
billion digits, but difficult to give them the skills of a one-year-old child’s
when it comes to perception and mobility. For an example, simple mental
abilities like recognizing faces, lifting little things, walking around were
hard to do. But could solve the hardest engineering problems. As a matter of fact, hard
problems are easy and easy problems are hard. Because of those reasons the progress
of the field showed a significant drop in 1974.
In the next
decade AI field boomed up by few millions of dollars from 1980 to billion
dollars investment by 1988. XCON, LISP machines, and Symbolics became famous as
that systems simulated the decision-making ability of the human experts to solve
problems such as diagnosing infectious diseases or identifying chemical
compounds. During that time Apple and IBM came up with desktop computers with
high speed and power which overtook the expensive LISP machines.
However, those
systems were very expensive to maintain since they were difficult to update,
could not learn, and brittle. Therefore, users were reluctant to spend money on those
specialized machines. This led to the collapse of the market for AI in 1987.
Currently, there
are tangible achievements gained with deep learning and this has a huge
difference from what happened in the past. Today we have many applications
working in real environments. But today also everybody expects deep learning to
be able to solve every problem. Indeed, this is not true, and a “little” winter
is coming in the future. Nowadays major investments in AI are for making self-driven
cars, conversational assistants in mobile phones, home appliances and games, etc.
The requirements have not still been proven as becoming absolutely indispensable
to the world. Because of that, they could still end up with a big collapse.
That means there
is still a significant risk that we can have another AI winter even with the
scale and the level of current financial and developer commitments. It could
happen if there is a decline in market fortunes of big tech leaders such as
Google, Microsoft, Amazon, and Apple, and unable to justify why to spend
billions of dollars every year on thousands of skilled AI researchers and
engineers if those efforts do not contribute directly to the bottom line of the
business. In that respect, the current AI wave is clearly tied to the market’s turnover.
That may change suddenly If the key players of the market are not feeling profitable
by spending funds on AI.
Even though at
the moment there are no signs of upcoming winter, there is a possibility to end
up the story with kind of skimpy again in the future.
James Lighthill |
[Tharidu Dilshan]