10-28-2019, 04:58 PM
** 28-Oct-2019 World View: Algorithms for the Singularity
The thing is that the Singularity is not drastically different at all.
Here's an example. The first major chess-playing program was written
in the 1960s, using the "minimax algorithm." The way that algorithm
works is "if I play A then you'll play B and I'll play C or if I
play A and you play X then I'll play Y." So the algorithm creates
a tree of possible moves and responses, and uses that to select the
best move. It was a pretty weak chess player.
So when IBM's Deep Blue beat Garry Kasparov to become world chess
champion in 1996, it was using the EXACT SAME minimax algorithm. The
only difference is that computers were much faster in 1996. In the
1960s, the move tree was only about 3 levels deep, but in 1996, on an
IBM supercomputer, it was more like 15 levels deep.
So there's nothing mystical about a computer chess champion. In fact,
this was already recognized in the 1960s that one day the 1960s
computer program would be world champion on a sufficiently fast computers.
The Singularity is exactly the same. A computer making decisions to
win a war or invent a new widget can do so with the same minimax
algorithm, except that the tree will have to be much wider. This
is well understood, and the only question is how long it will be before
computers are fast enough. I estimate 2030.
This is the article that I wrote in 2005 on the Singularity, and
reposted in 2015:
** Artificial Intelligence and the Singularity by 2030
** http://www.generationaldynamics.com/pg/w...151228.htm
In that article, I described in detail the algorithm that will be used
to make computers smarter than humans. That algorithm could never
have been implement in 2005 because computers were too slow. But by
2030, computers should be fast enough to implement it.
In recent years, the minimax algorithm has been buttressed by
additional algorithms. Pattern matching algorithms will provide a
kind of "intuition," where an AI robot can made decisions by comparing
the current situation to a large database of similar situations in the
past. Deep Learning algorithms are used to create the large database
to be used in pattern matching.
You mentioned quantum computing in a previous post. Quantum computing
is perfect for this application because a quantum computer should be
able to perform millions of pattern matching tests in a few
microseconds.
Once again, there's nothing mysterious, magical, or spiritual about
this. There's no need for God to breathe on an AI computer to give it
a soul. The algorithms are well understood today, and all that's
needed is a fast enough computer.
(10-28-2019, 01:25 PM)David Horn Wrote: > Give this a moment's thought. Why should something so drastically
> different be imaginable in any real degree? The best thought
> exercise I ever heard, (note: it's not mine to claim) made the
> point perfectly. Get in a Wayback Machine, and travel to
> Runnymede at the time of the signing of the Magna Carta -- surely
> an example of advanced thinking for its time. Once there, find
> someone to harangue about the future, and expound to him or her
> (most likely, him) about the Internet. For that matter, just get
> them to understand life in the 21st century. No one in that time
> would have had a usable benchmark for anything we take for
> granted. Air conditioning and automobiles would be magic, but the
> internet -- simply beyond comprehension.
> For us, the Singularity is our future "Internet".
The thing is that the Singularity is not drastically different at all.
Here's an example. The first major chess-playing program was written
in the 1960s, using the "minimax algorithm." The way that algorithm
works is "if I play A then you'll play B and I'll play C or if I
play A and you play X then I'll play Y." So the algorithm creates
a tree of possible moves and responses, and uses that to select the
best move. It was a pretty weak chess player.
So when IBM's Deep Blue beat Garry Kasparov to become world chess
champion in 1996, it was using the EXACT SAME minimax algorithm. The
only difference is that computers were much faster in 1996. In the
1960s, the move tree was only about 3 levels deep, but in 1996, on an
IBM supercomputer, it was more like 15 levels deep.
So there's nothing mystical about a computer chess champion. In fact,
this was already recognized in the 1960s that one day the 1960s
computer program would be world champion on a sufficiently fast computers.
The Singularity is exactly the same. A computer making decisions to
win a war or invent a new widget can do so with the same minimax
algorithm, except that the tree will have to be much wider. This
is well understood, and the only question is how long it will be before
computers are fast enough. I estimate 2030.
This is the article that I wrote in 2005 on the Singularity, and
reposted in 2015:
** Artificial Intelligence and the Singularity by 2030
** http://www.generationaldynamics.com/pg/w...151228.htm
In that article, I described in detail the algorithm that will be used
to make computers smarter than humans. That algorithm could never
have been implement in 2005 because computers were too slow. But by
2030, computers should be fast enough to implement it.
In recent years, the minimax algorithm has been buttressed by
additional algorithms. Pattern matching algorithms will provide a
kind of "intuition," where an AI robot can made decisions by comparing
the current situation to a large database of similar situations in the
past. Deep Learning algorithms are used to create the large database
to be used in pattern matching.
You mentioned quantum computing in a previous post. Quantum computing
is perfect for this application because a quantum computer should be
able to perform millions of pattern matching tests in a few
microseconds.
Once again, there's nothing mysterious, magical, or spiritual about
this. There's no need for God to breathe on an AI computer to give it
a soul. The algorithms are well understood today, and all that's
needed is a fast enough computer.
