Quote:
Originally Posted by Rimlog
A lot of people here have the wrong idea about the limitations of software. (I have started several software companies, have written chess and poker programs, and have developed "learning" hedge fund algorithms that outperform human experts). A few points:
Stars may be detecting many bots, but I am confident that the best ones go undetected. Bot-detection depends on the software playing in a machine-like way, which means unnatural levels of repetitiveness (of speed of action, strategic lines, lack of mistakes etc). A smart bot developer can program in all of these kinds of fallibility and randomness, putting in enough variation to fool the detection algorithms, but not so much that it falls out of profitability.
Some people are assuming that bots will always be GTO-based. That is a solid starting point, but just as the most successful players understand GTO but move away from it to exploit specific player leaks when they spot them, so a well-constructed bot can incorporate this capability.
Multi-player games are an increased challenge, but much more so for humans than for computers. The successful bots in these complex situations will not be ones based on clever multi-player GTO theory, but much dumber ones that learn through genetic approaches. With the ability to try random strategies and let the best ones "evolve", that sort of software will easily crush anything that tries to implement a human-designed algorithm. They will use evolution to derive base strategies, and then on-the-fly learning approaches to detect and exploit player habits and table dynamics.
I have no doubt at all that on-line poker will eventually be killed by these programs, and I am suspicious that some of the steady increase in toughness of games in the last couple of years is already coming from this factor.
Although I agree with most of this I again claim three things: (Although only number 1 really matters.)
1. There are poker games that are more fun, allow looser play and more multi way pots and enough volatility such that even "perfect" players or bots won't really ruin the game or intimidate the recreational player.
2. Bots will never be programmed to take full advantage of bad players because it is a needless risk given they are certain to have an edge without doing this. Thus great players will outperform them in situations where there are bad players in the mix.
3. Bots that adjust away from GTO based on how players play will probably increase their winnings UNLESS the opponent is an expert who KNOWS that this is what he is up against (while the computer doesn't know he knows). Do you see why?