The promise of true artificial intelligence has long been one of the holy grails of computer science. With the eventual goal being true intelligence capable of self-awareness, our current attempts cannot truthfully fall under that strict heading. AI, as it currently stands, is more a comprehensive understanding of rules and responses, rather than an ability to freely think and reason.
This might seem like a limitation, but in reality, this can prove to be an excellent approximation of how we operate as humans. We are, after all, a form of biological computer. As biological computers, we can often approach challenges with similar understandings of strict rules and possible outcomes, which is what makes our even primitive AI regularly so successful.
It is even becoming increasingly popular to utilize AI and machine learning to overcome issues which might be too complex or time consuming for human thinking. From poker to chess, science, and video games, these forms of AI have become a powerful tool in education and entertainment.
The question we want to look at today is what exactly makes it possible for a human to beat an AI, and when are the odds stacked in our favors? We’ll look at a few popular examples, to determine exactly where this line of challenge should be drawn.
Perhaps the best chance we have against AI comes from games which are themselves based on chance, like casino games. Poker strategy, for example, relies on mathematical understanding and interpersonal skills to gain an edge. To know the best poker hands and odds in Hold’em and Omaha will help your chances, but we need to remember that poker is more than skill.
No matter how good you are at the game, there will always be luck involved. This luck comes down to individual hands, and the way the cards come out over longer games. In the case of true randomization after each hand, it would entirely possible for a player to match or beat an AI player, depending on their luck. Hand odds can be memorized and calculated by people, after all.
If, on the other hand, the decks were not shuffled after each hand, then the AI would have a definite edge. With the ability to instantly memorize each card played, an AI would possess the ability to rule out far more card combinations than a human player reasonable could, giving it a definite edge.
This is somewhat analogous to card-counting in blackjack. With these card-counting methods, human rely on simplification systems, but an AI with perfect memory would not have to use such mental or mechanical shortcuts.
The most famous examples of AI taking on human opponents come from the world of chess. As a game of intense strategy and deep forethought, Chess has long been a sport where humans have shown their intellectual flexibility and capability.
Much of this has been drilled into the best players over decades, with lessons learned going back dozens or hundreds of years throughout chess history. Since the early days of computers, their AI programs had proven capable of beating beginners and even advanced players, but it took until the mid-nineties for a grand-master to be dethroned.
1996 was an exciting year for chess and AI both, as World Champion Garry Kasparov took on the IMB chess supercomputer known as Deep Blue. The first match had some back and forth, but ultimately man would defeat machine as Kasparov won 4-2. The rematch, in 1997, however, did not go in Kasparov’s favor, as he lost 3½ to 2½. So how is it that the machine eventually beat the man?
Chess supercomputers are incredibly dedicated devices. Within their memory, they store information from millions of different games and combinations of moves. From this information that can categorize what the most likely winning strategy will be after each step, judging the outcomes which have followed that move in similar games.
Of course, this is not perfect. Even high-level games of skill such as chess rely on some degree of chance. What if your opponent makes an unexpected gamble like a bluff in poker, for example, or throws apiece to manipulate the opponent? While it is possible to take much of this into account, it also means that there is no such thing as a perfect chess computer. This is because the unpredictability of man is not something which can always be accurately measured.
The most common AI foe which millions face daily lingers within the realm of video games. These play a part in almost every game developed to some degree and run the gambit of seemingly human to inhumanly obtuse.
Unlike something like poker or chess, video games are constantly changing, and with each change comes the need for new systems of AI. Even sequels in a game series usually require artificial intelligence to be completely rewritten, as the game’s underlying code governing rules evolve. Sometimes these AI systems are given immense attention, other times they are barely functional.
Making AI in a video game is a balancing act, as the general idea is to provide a challenge, rather than to create an unbeatable system. This is difficult in that these enemies or opponents also have to be enjoyable to play against. An enemy who can track and hit targets instantly, for example, might be difficult, but it will only drive players away.
There are experimental exceptions to this rule, however. OpenAI Five, a recently developed AI for DOTA 2, took Deep Blue’s example by being to the most recent AI to beat the best in human competitors. With a simulated 45,000 years of play experience, this is perhaps not much of a surprise.
The Human Element
The fundamental difference in what makes humans and AI different comes down to their flexibility and memory. A computer can memorize everything and human memory cannot compare. A human can employ seemingly random strategies with long-term viability whereas an AI’s randomization is too strict to be truly useful, at least without preparation and testing.
Whether or not you can beat an AI means understanding its limitations, and doing what you can do to abuse these limitations. The greater the chance involved the better your random success, but the greater the odds of math not swinging in your favor. Beating AI means picking your battles, and understanding that there is no such thing as a sure bet.