"Rich Hutnik" wrote in message
...
On May 8, 12:16 am, Ed Murphy wrote:
wrote:
On May 7, 9:32 am, David Richerby
wrote:
wrote:
On Apr 18, 10:18=A0pm, John Bailey
wrote:
In [Deutsch's] paper, the convenience of thinking that all
computing can be reduced to an equivalentTuringmachine is
considered and rejected.http://xyz.lanl.gov/abs/math.HO/9911150"As
a matter of fact, Richard Feynman, in his talk during the First
Conference on the Physics of Computation held at MIT in 1981,
observed that it appears to be impossible to simulate a general
quantum evolution on a classical probabilistic computer in an
efficient way.
That is largely because the only thing Feynman knew about computers
or computations was main frame computers. But since most of today's
computations depend on massively parellel networks, not mainframes,
his observation mostly concern 1950s Burroughs history, not
computers.
No. Combining two classical computers together makes a computer twice
as big that can do (roughly) twice as much work in a given time.
Connecting two quantum computers would make a computer twice as big
that can do (roughly) the square of the amount of work in a given
time.

Networking classical computers together gives you at best linear
growth; adding qubits to your quantum computer gives exponential
growth.

Networking classical computers gives you microcomputers,
laserdisks,
satellites,, HDTV, Holograms, fiber optics, robots, and a
paycheck,
rather than idiots like computer scientists, that's why they were
invented.

You guys are speaking rather orthogonally. In theory (his topic),
quantum computers could do all this and more. In practice (yours),
adding more qubits is (so far) Really Difficult; it will take a
major technological breakthrough to build a quantum computer with
enough qubits that a network of classical computers can't simulate
it at full speed.

The original topic had to do with the future of chess, and where it
might head. This then spawned "Hereclitian-Calvinball" as a question
of whether or not there is a finite or infinite number of potential
chess variants. Maybe Quantum Computing can answer this question.

But isn't the [gigantic] fly in the ointment the fact that parrallelism to
increase brute-force solutions is still a very questionable paradigm?

After all, the comprehension of two 1600 players does not equal that of a
2200 player.

While there may be an increase in event horizon by quantitative analysis
which may achieve some result, the 'fly' is that these are typically
quantitative assessments based on materials won/lost, aso where is the
qualitative one?

[[ IE: unless a conclusive result is achieved by brute-forcing,
[example; mate] then what does any program do when at ply 12 it sees the win
of a pawn, but costing two tempii? Perhaps it will continue for another 12
plies and discover it recovers one tempo, keeps the pawn, but loses the
initiative... ]]

Therefore what is lacking in brute-force approaches is qualitative
evaluation of /specific/ positions.

The 'fly' turns out to be a man-made one - since evaluating a tempo or other
positional factors such as initiative, are factors that the programmer
assesses, not the chess-engine. Furthermore, these assessments must
necessarily be abstracted ones, sui generis, since they are /initial/ data
programmed in the chess engine, created from mean data, as averaged
ennumerated evaluations.

As we know, many 'averages' never occur, since data sets can be heavily
polarised away from any instance of mean value - the averaged condition that
is pre-programmed may in fact, /never/ occur].

Attempts to correct qualitative analysis lie in provision of yet more data
evaluation sets, such as for middle-games, or sub-sets where 2 bishops have
an open/closed position, etc. But the program itself does not generate the
data set!

Thereby, no contextual evaluation of the worth of material/positional
factors takes place by the act of the chess engine's own calculus - and this
is the stalled point in chess computing emulations, [emulation* since the
program is merely acting on received data and not conducting its own
evaluation] and has been so for 10 years. It is so stalled, that AI
researchers gave it up as anything much useful to them.

To remove chess computing from emulation, the program needs to not only play
its own moves, but successively generate its own evaluation criteria.

Phil Innes

*Whether qualitative, or quantitative such as using opening books.

- Rich

On May 12, 10:02 am, "Chess One" wrote:
But isn't the [gigantic] fly in the ointment the fact that parrallelism to
increase brute-force solutions is still a very questionable paradigm?

After all, the comprehension of two 1600 players does not equal that of a
2200 player.

While there may be an increase in event horizon by quantitative analysis
which may achieve some result, the 'fly' is that these are typically
quantitative assessments based on materials won/lost, aso where is the
qualitative one?

I mentioned what I did in hopes that maybe it can get connected to the
original issue. Brute Force only gets you so far. Brute Force isn't
judgment, it is evaluating everything. Maybe somebody will come up
with something else.

I believe Brute Force would fail in a Hericlitian/Calvinball
environment against a player trained to be adaptive at strategy. I
know a former coworker who ended up trashing Zillions at a bunch of
games (maybe the person was blowing smoke here).

- Rich