Turing Machines. Roger Penrose. Artificial Intelligence.
Colossus. Enigma. Chess. Machine Intelligence.
I’ve been trolling YouTube and fell into some videos about
Alan Turing and the enigma machine. I like math and science videos. That led me
to some history, plus some mathematics. And then AI, since Alan Turing created
the Turing Test. That’s where you talk to ‘someone’ through a computer screen.
You can ask them anything you want and say anything you want and consider their
response. If, after a certain amount of time, you can’t tell whether you are
connected to an actual human being or to a computer simulation creating
appropriate sounding answers to your dialog, then you have to admit that that
computer is demonstrating intelligence.
Maybe even consciousness.
This always bothered me. Simulated is not Real. Simulated is
make believe. Simulated is generated according to rules without consequences.
The simulation machine does not suffer or die if it gets the simulation wrong.
It just gets reset and runs another simulation.
Ay, there’s the rub.
Take chess. It’s a game of mathematics. No, really. Pieces
can move on a board a number of times. Two spaces forward. As many spaces
diagonal as you can. Two forward and one left or right. Landing on another
piece removes them from the board. Now there’re consequences. Putting the other
player’s king in inescapable jeopardy wins the game. Ultimate consequences. All
math. Math with a touch of death.
Let’s say we begin. You’re white. What are your choices?
Well, you can move one of your pawns either one or two spaces, or you can move
one of your knights forward two and one to the left or right one. That’s a
total of 20 possible moves. Sixteen for the pawns. Four for the knights. Those
are you legal, mathematical, moves.
There are 20 ways to start a game of chess. For each one of
those, there are a number of legal, mathematical moves allowed to your partner.
Twenty, just like you. So for each of your opening moves, your partner has 20
possible countermoves. This makes 400 possible first two moves in a game of
chess.
Now. Imagine you are walking down a path. After a while you
come to a divergence in the path. It may split into two or three or any number
of new paths, in this case 20, and you have to choose one. Let’s say that each
path has a sign in front of it. It has three numbers on it. The number of times
you will die by taking this path. The number of times you will live by taking
that path. And the number of times that, well, nothing, really. It’ll just end
in a draw.
Now you can probably see where I am going here. Why not just
calculate every possible chess move into the huge Chess State Space Almanac and
follow each path down to its ending? Each move has a number of lines coming out
of it to potential next moves by either white or black. Every move is a node on
that total chess tree. And every node has many possible moves beneath it. Beneath
each node are many paths for black and white to play. Another set of moves for
each player. Each path ends in a number of possible wins for black, white, and
a number of stalemates. Let’s assume that every chess game is limited to 200
moves. This will eliminate the games that end in two kings chasing each other
around the board for ever. If no one wins in 200 moves, it’s a stalemate. That
makes the total possible number of games of chess finite, though formidably
huge. Now let’s figure out how many games are left.
Assuming these rules and the standard moves for each piece,
each node in the state space chess tree can be assigned three numbers
representing all of the results of all of the possible games beneath it. Wins
for white. Wins for black. Stalemates. If I am white, I just have to look at
the node that I am currently on, which is my opponent’s last move, and then at
all of my possible next moves. There may be dozens. There may be few. I have to
choose one. But they will all have numbers assigned to them telling me how many
times white wins if I choose this path, how many times black wins if I choose
this path, and how many stalemates. I just have to choose the one with the
greatest number of wins for white beneath it.
And then black uses the same strategy. Just look at the Big
Book of All Chess Games, identify where you are at the current move, and choose
the path ahead with the greatest number of wins for black.
Is this intelligence? Is this statistics? Is this
impossible? Well, yes. It is impossible to create a catalog of every possible,
legal chess games that can be played in 200 moves. The numbers would confound
Carl Sagan and take longer to compute than the life of the universe. But that’s
not the point.
Still. This illustrates a point and raises a question. The
point being that intelligence can’t just be about calculating all possible results
and then choosing the one with the most desired outcome. The one with the
highest number of wins. That would be computationally impossible. The question
is; Is that what our brains are doing? Are they just creating a matrix of
possible outcomes and then choosing the best? Are they calculating into the
future and choosing the path that has the greatest number of successes for our
survival strategy? Maybe only going down a fixed number of nodes and making the
best guess of it? How is that even possible without knowing what will happen
all the way to the end? How can you calculate ‘a fixed number of nodes’ what’s
going to happen through all of eventuality?
I don’t think that is what intelligence is. That may be what
brute force is. What computational efficiency is. Machines may be programmed to
simulate intelligence and be shown how to work towards a goal. But what
criteria will they adopt to achieve that goal? Simply grinding down numbers
until they get the highest reward and then choosing that path? Sounds alien.
I think there is one thing missing in the whole AI debate.
Machines can’t feel pain. Machines can’t feel consequences. Machines don’t live
in time and don’t know they will die. All these things are a visceral part of
our brains and our consciousness. A machine is not going to say; If I do this
thing, it may come back to haunt me. Its calculation is going to be simply; What
gives it the greatest benefit in the next transaction? No thinking ahead. No
regrets. No bargaining for one’s life. No; How will this affect me in the
future? My life? My children? My society? Basically, any machine intelligence
is not going to be human. Not alive at all, actually. Just calculations without
thought. Just brains without a conscience.
Simple math. If I do this I get three points. If I do that I
will get four. No brainer.
Not subject to any rules that threaten its existence, nor
frightened by any demon that can harm it, the machine will be a dumb thing
unaware of itself and of what it is doing. Only following rules and
mathematics. Not a bad thing. Depending on what goals we give it. But hardly
self-aware. Or aware of anything else, really. But still. Very powerful in an unconscious
way. Capable of madness even if it is not capable of awareness. Cruel in its anonymity.
Playing by the numbers.
Let’s be careful where we are going with this, shall we?
No comments:
Post a Comment