When was an AI first able to play a game against a human?
Hey there fellow tech geek! As someone passionate about gaming and streaming, I know you‘ll be fascinated by the history of AI game-playing. Artificial intelligence has come a long way from humble beginnings of checker and chess programs to today‘s world-beating game bots. Let‘s dive into the major milestones of AI game-playing and the pioneering minds behind them!
The Checkers Playing Machines (1950s-60s)
The earliest inklings of AI game-playing began in the 1950s as scientists started exploring whether machines could learn and exhibit intelligent behavior.
One of the very first AI programs ever written was for playing checkers. Tech legend Arthur Samuel developed a checkers-playing program in 1952 that improved through practice. This was achieved by having the program play against modified versions of itself – a novel machine learning technique at the time!
Samuel‘s program didn‘t play at a human master‘s level, but the ability to learn and beat amateur players was an impressive feat in the early days of AI. And this work laid the foundation for AI game-playing programs that would come later.
Around the same time in 1958, Alex Bernstein developed a chess program that could achieve novice-level play. And just a few years later in 1962, John McCarthy – one of the founding fathers of AI – helped develop a chess program at MIT that could play complete games against humans (albeit other beginners).
| Year | Program | Developer | Game |
|---|---|---|---|
| 1952 | Checkers playing program | Arthur Samuel | Checkers |
| 1958 | Chess program | Alex Bernstein | Chess |
| 1962 | Chess program | John McCarthy | Chess |
Though basic by today‘s standards, these programs were important achievements in the infancy of AI research. They demonstrated computers could learn and improve at tasks like game-playing which were long thought to be exclusively human skills.
BKG 9.8: The Backgammon Champ (1979)
In the late 1970s, scientist Hans Berliner developed an AI backgammon program named BKG 9.8. It was built using a neural network, which gave it the ability to learn by analyzing thousands of positions.
In 1979, BKG 9.8 achieved a major milestone – it defeated world champion Luigi Villa 7-1 in a backgammon tournament! This was the first time a computer program had ever defeated a human world champion in any game.
Berliner‘s neural network program was groundbreaking. For the first time, an AI system taught itself to play and excel at a game purely from data, rather than being programmed with human-created rules and strategies.
The Rise of Chess AIs
With backgammon conquered, scientists set their sights on one of oldest and most complex games humans play – chess. Chess-playing programs saw steady progress through the 1970s and 80s.
In 1988, a chess program called Deep Thought developed by Carnegie Mellon University and IBM drew two matches with Grandmaster Bent Larsen. It was the first time a computer achieved a draw against a chess grandmaster.
But the big highlight came in 1997 when IBM‘s Deep Blue defeated world chess champion Garry Kasparov 3.5 to 2.5. This was the first time a computer program beat a world champion in chess under tournament conditions.
Kasparov accused IBM of cheating and demanded a rematch. But Deep Blue won again under closely monitored conditions, proving it could beat the best human through skill alone.
This was a significant feat – chess had long been considered the epitome of human intellect. Kasparov‘s defeat signaled that AI was entering a new era where it could match humans even in demanding cognitive tasks.
AlphaGo and the Game of Go (2016)
After conquering chess, AI programs turned to the ancient Chinese board game Go. Go is considered even more complex than chess due to its higher number of possible move configurations.
In 2016, Google‘s DeepMind developed an AI program called AlphaGo which defeated legendary Go player Lee Se-dol 4-1. This stunned the AI community as many experts believed it would take at least a decade more for computers to best human Go players.
So how did AlphaGo do it? It used a novel combination of Monte Carlo tree search and deep neural networks to develop superhuman intuition and strategy. By analyzing thousands of human Go games and playing against itself millions of times, AlphaGo attained a mastery of the game.
The Rise of Multiplayer Video Game AIs
With board games conquered, AI researchers started tackling video game environments in the 2010s, which pose even greater challenges.
In 2016, DeepMind developed agents that could exceed human performance across dozens of Atari video games. And in 2018, their program OpenAI Five achieved human-level play in Dota 2, a complex multiplayer online battle arena game.
Other researchers have built AIs that can beat professionals at multiplayer games like StarCraft II and poker. These victories indicate AI is mastering skills like teamwork, bluffing, and handling uncertainty – thought to be uniquely human strengths.
Just as impressive – amateur developers have used AI models like OpenAI‘s GPT-3 to generate their own video games! We truly are entering an amazing era of AI creativity.
The Future of Game-Playing AI
AI has come a long way from the early checkers and chess programs of the 1950s. It has exceeded human capabilities in one game after another thought to be bastions of human talent – backgammon, chess, Go, poker, StarCraft and more. This rapid progress is thanks to advances in neural networks and reinforcement learning.
And researchers are just getting started! With increasing data and computing power, I think AI could conquer most human games in the next decade. But its limits will be continually tested by multiplayer competitive games which pose dynamic real-time challenges. Online game platforms are ideal testing grounds for developing increasingly flexible AI systems.
Exciting times are ahead my friend! With AI as a teammate or opponent, the future of gaming looks bright. But human creativity will always be needed to push game experiences in new directions. I don‘t think AI can ever fully replace that, though it will keep challenging our assumptions of what is uniquely human. What do you think? Let me know if you want to geek out more about the past and future of game-playing AI!