Page 22 - CyberSecurityEssentialsEnglish

What Is Artificial Intelligence?

How can we understand artificial intelligence before we use it?

A practical introduction to AI capabilities and limitations, while keeping goals, decisions,
and responsibility in human hands.

Story: Move 37, the Move That Looked Wrong and Changed the Game
In March 2016, Korean Go champion Lee Sedol faced AlphaGo, a system developed by
Google DeepMind. Go is an ancient board game in which two players place stones on a grid
in order to surround territory. It is one of the hardest strategy games for computers, both
because the number of possible move sequences is enormous and because judging which
side is ahead in a given position is difficult to reduce to a simple rule. In the second game of
the match, AlphaGo placed a stone in an unexpected position on move 37.

The move looked so unusual that commentators initially thought the system had made a
mistake. Its value became clear later, however, and it helped AlphaGo win the game. The
system’s own policy network had estimated the probability that a human expert would play
that move at roughly one in ten thousand [1].
The story may invite a hasty conclusion: if the system saw what the game’s champions did

not, had it begun to think like a human? The answer is no. AlphaGo excelled at a specific task
with explicit rules and a measurable objective. It could not write a management report,
understand a human situation, review a contract, or run an organization. Exceptional
performance in a narrow domain is not the same as general human intelligence. This leads

to the chapter’s central question: why does AI excel at some tasks yet fail at others that
appear simpler?

The story does not mean that the machine acquired consciousness or human intuition. It
shows that the word “intelligence” covers different capabilities, and that any system should

be evaluated in relation to its task, data, objective, and operating environment. AI can
accelerate work, detect patterns, and suggest alternatives, but it does not relieve people of
the responsibility to define the goal, examine the output, and remain accountable for the

decision.
Source: Google DeepMind, AlphaGo [1].

“Excelling at one task does not amount to general intelligence; it means the system is
powerful within the boundaries of that task.”

21