Sep

16

Cyborg Chess, from Nigel Davies

September 16, 2015 |

'Cyborg Chess' or 'Advanced Chess' is an area that might be of interest to specs in that humans are allowed to use computers during the decision making process. There is evidence that strong human players can add considerable value to pure computer play when the process is managed in the right way, for example Arno Nickel defeated Hydra in a correspondence chess match in which he used a regular PC against the the most powerful supercomputer in the World at that time. This event wasn't publicized as much as it might have been, but you can read more about Nickel and Cyborg Chess here:

Arno Nickel

Advanced Chess

I've experimented with 'Cyborg Chess' in correspondence tournaments in which computers are allowed. The results haven't been great, probably because I don't use deep calculation setting on the engine, but the experience has been educational. A major issue is in understanding where it is that I can add value as there's a temptation to either overrule what the engine recommends or be led by it indiscriminately. Probably a series of protocols would be a good idea but where does one start? Here's a provisional list:

1. Write down your list of candidate moves, in order, and then compare them with the top choices of the engine.

2.Consider whether this is the kind of position in which engines are likely to do better than you (ie highly calculative tactical ones).

3. Give greater weight to particular candidates based on point 2.

4. Check your top candidate(s) more carefully, perhaps using deeper engine settings, until a particular confidence level is arrived at.

It seems reasonable that different people might give a different weighting to their own choices versus those of the computer, but in either case it does seem that better decisions might be arrived at. In fact Nickel's achievement sort of proves that, and even if computers get so powerful that the more or less 'solve' chess the synthesis of man and machine should still have value in less finite fields.

Victor Niederhoffer writes: 

"Cyborg Chess" by Nigel brings up the effectiveness of human versus
robot trading in markets. Certainly costs must be considered as well as
effectiveness the way it is in all the studies of robotic versus human
surgery. Apparently robotic beats laporofic.
There should be areas where the robots have to be turned off for the
evening where the humans could develop an advantage. It seems the robots
are forcing early capitulations in many markets which is presumably an
effect of their programs.

anonymous writes: 

To list just two of scores of regular robot shutdowns that one knows of:

1. On Sunday nights in the professional electronic FX markets (using HotSpot as an example), one only has access to prices from 5PM NYC time unless you get on the phone and call a counterparty direct in New Zealand or early Sydney.

This 'dead zone' is almost completely without 'silicon based entity' interference and often sees a reasonable range that goes unrecorded. A stint in that dead zone is a prized achievement for FX traders learning how markets 'really' trade. Much like time on the floor of an exchange, it is an experience that is dying off.

2. Each night at 5 PM NYC time the professional electronic FX market goes dark for a few minutes as the value date changes.

After reopening, the market making algorithms kick in first with relatively wide spreads that narrow quickly when the Carbon based life forms start to interact. The HFT 'order facilitation' ( Ha!) kicks in next.

What is of increasing concern is that the lunatics are running the asylum. Meaning that the firm's running the robots are deciding when and why markets open and close rather than some supervisory body. I guess this is more a question of nature versus nurture.

Arguably, there is some marginal information that is helpful, in an accretive sense, to the buy or sell decision–from the opening procedures of robot dominated markets.

The first order possibilities for testing might involve: number of transactions per unit time, rate of change of spread contraction, the epps effect et.al. All for relatively short periods as the robo-market opens.

At a practical level, and without investing what I know to be substantial funds to study this issue, I believe it still comes down to basic conditionality, expectations based on that conditionality and finally path dependency.

Additionally, the predictive nature or otherwise of the situations introduced into the price generation process by exchanges, that I have previously posted on - must be tested and incorporated.

Jim Sogi writes: 

By their nature, cyborgs must look for fixed patterns. They have limited adaptability. Sudden bugs, unexpected changes, changes in cycles, and divergences will always surprise them. They can't anticipate. Their advantage is that they are as fast as their circuits, and comm allow. The unknown is how they perform in a complex system with other cyborgs and humans. As Nigel points out, a human can add value and beat a pure cyborg. Human foresight and understanding of human nature can add value.

Hernan Avella writes:

Machines keep improving, some moving away from brute force approaches…

"Deep Learning Machine Teaches Itself Chess in 72 Hours, Plays at International Master Level":

"Lai has created an artificial intelligence machine called Giraffe that has taught itself to play chess by evaluating positions much more like humans and in an entirely different way to conventional chess engines.

Straight out of the box, the new machine plays at the same level as the best conventional chess engines, many of which have been fine-tuned over many years. On a human level, it is equivalent to FIDE International Master status, placing it within the top 2.2 percent of tournament chess players"

Andrew Goodwin writes:

I still have my ticket stub from the match that Kasparov lost to Deep Blue in 1997 in NYC. Maurice Ashley was using the Fritz engine to evaluate the moves of the champion and the supercomputer in real time for the theater audience, as I recall.

Instead of making the next move optimization target the best calculable move, the supercomputer could make goal seeking calculations that lead the match to the most time consuming calculable end game for human competitors. It won that match with clock time to spare. That's the advantage.

The Chair's idea of a downtime for computer engines sounds sound for human comparisons.

Jim Sogi writes: 

I would challenge anyone to quantify what exactly is the difference between a cyborg traded market and a human traded market. Sure it feels different, but how exactly? How do the numbers trade. Are there less big blocks? Are there fewer round sizes? Are there fewer takers on breakouts, i.e stop buy orders? Where are the numbers on the table?

Hernan Avella writes: 

Difference? Generally speaking, most of the time, when bots are the market makers there is less friction, reduced bid-ask spread, more ability to get the trade done with less price disruption. Winners: longer term traders willing to pay the bid ask spread or less to get into or out of a position. Losers: human market makers who want to earn the bid-ask spread. They can no longer compete.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search