Scribd
Upload
74 views2 pages

wp1 Genre

Uploaded by

api-616201584
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
74 views2 pages

wp1 Genre

Uploaded by

api-616201584
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

HZ No.

1 Middle School-Science-P1

The Evolution of
Computing:
AlphaGo
By Jim X. Chen

Author’s word: Congratulations to our


awesome campus Go club for winning the state
championship last week! As a student who
enjoys playing Go, I would like to take this
opportunity to introduce this ancient game
and discuss how it relates to science.

Basic Rules of Go
Figure 1 shows the Go board: two
Go and AlphaGo players (black and white) take turns placing a
stone on the board. Imagine that each grid
Although artificial intelligence (AI) line is a single-plank bridge, and you–the
“Deep Blue” defeated many top chess players stones–are standing on the bridge over a lake
in the late twentieth century, most Go players (the areas other than the grid lines). You
insisted that it would be many years--maybe a can’t stand on the bridge since you may fall
century--before AI could beat top Go players. into the lake, so the only standing points are
Go, or Chinese chess, whose possible the empty grid points.
variations exceed the number of atoms in the
universe, is easily the most complex board
When an area is surrounded by an
game of all time. However, “AlphaGo,” AI
opponent’s stones, it’s “alive” if it has at least
developed by Google Deepmind, defeated the
one empty point inside the area; otherwise,
best Go player in 2016, shattering this axiom.
the stones in the area are captured and
removed from the board. However, if the
surrounded area has only one empty point
left, it’s captured because an opponent can
place a stone in that point and solidly
surround it. Therefore, an area is only really
alive if it has two separate empty points
inside—for example, as Figure 2 shows, if
black places a piece at X, then its area is
alive. The players count points in their own
live areas to decide who won or lost.
Huizhou No.1 Middle School-Science-P2

Variations Evolution
Despite its simple rules, Go has The success of AlphaGo reveals the
numerous possible variations: the person who evolution of computing. In 1983, it would have
starts a match has 19 × 19 = 361 possible taken 31 minutes to make a decision about
empty points to place the stone, leaving 360 46,655,640 possible moves back then. Today,
empty points for the opponent’s next move. it only takes less than 12 seconds to make
Hence, the possible variations are 361! (the that same decision, which is 160 times faster
mathematics operator “!”, or factorial, than the algorithm before.
multiplies all positive integers less than or
equal to the number; for example, 4!=4 ×3 ×2
×1). The huge number of options–361!–is still
beyond computing power. Suppose a human
player can see three steps ahead–if so, we
have 361 × 360 × 359 = 46,655,640
variations–which is manageable for computers.

AlphaGo was initially trained to mimic


human play by attempting to match the moves
of expert players from recorded games (its
database holds 30 million moves). Once it Figure 3 shows the evolution of the
reached a certain degree of proficiency, it CPU (the five elements presented in the
was trained further by playing multiple games graph are the main components of the CPU). I
against other instances of itself, using will not explain what the CPU exactly is, but
reinforcement learning (a machine learning you can just consider it as a representation of
training method based on rewarding desired the speed of computing. From the graph, it is
behaviors and punishing undesired behaviors) clear that the number of cores and
to improve its play. transistors continues to double every other
year or so. If researchers design better
computing algorithms, they could double the
There’s no easy way to evaluate a
computing speed at its current level every
position in Go–the number of stones on the
other year or so.
board is a weak indicator of a position’s
strength, and a territorial advantage is
difficult to calculate. A professional human AlphaGo is probably the most
player can make relatively easier judgments powerful in its database, which is the weakest
compared to a machine due to instincts that aspect of human capacity compared to
algorithms can’t capture, but AlphaGo might computers. With its advancement in
have broken this barrier. computing and storage power, AlphaGo beats
the most diligent and deeply intelligent human
brains. Scientists and researchers are looking
forward to seeing the brighter future of the
evolution of computing power.

You might also like