AI research until 2010

 Until 2010 there was only a little progress visible for Artificial intelligence. Even if the subject was researched over decades, the amount of practical applications was low. The most advanced AI project until 2010 was the Deep Blue chess computer who won in 1996 against the strongest human player, and Honda Asimo robot from 2000 who was able to walk on even ground. But, both projects were really expensive, and the demonstrated form of intelligence wasn't useful in the real world.

Its a bit surprising, but especially in the years from 1990 until 2010 AI has made no progress. In parallel the Internet has developed rapidly, and computers have become powerful and useful machines. Before the year 2010, AI was a dead end, that means, no success was available and the researchers didn't even know why they have failed.

Until 2010, computer chess was the most promising direction in the AI community because it provided an accepted indicator for the thinking process. If a computer was able to beat a human player in this board game, this would be equal to an early form of machine intelligence. The dominant problem until 2010 was the low amount of CPU resources. Computer hardware was only able to calculate a certain a number of movements per second which is equal to the search horizon. This limited search horizon prevented the advent of powerful AI programs. The mentioned Deep Blue chess engine was only able to win against Kasparaov, because it was running on a dedicated supercomputer. So many experts were in doubt, if deep blue was a good example for Artificial Intelligence.

The interesting situation is, that not only a single university or a single discipline in AI was struggling with realizing the dream of robots and thinking machines, but the entire AI community until 2010 was unsure how to overcome the obstacles. Even very advanced neural network algorithms in combination with reduced problem complexity, for example to control a simple maze robot, were not successful at all. There was a serious doubt, if AI can be realized in the near future.

Despite the huge of effort put into different AI disciplines like computer vision, natural language processing, planning and reasoning, it was hard to find practical demonstration of AI. Minimalistic research problems like the traveling salesman problem, or the block sorting problems were described in the literature without any modification over decades. In other words, the AI landscape was froozen and the absence of progress was common.

Until 2010 the theoretical description of np hard problem was available, but it was unclear how to solve these kind of problems. Existing algorithms were not powerful enough to plan the trajectories of robots and even more recent approaches like probabilistic algorithms were not powerful enough. In summary, AI research until 2010 was in sad condition. The gap between practical algorithms and the description of robots in the science fiction literature was huge. A common assumption until 2010 was, that in reality it will take 200 years and more even robotics can be realized which means, that the AI subject will become a dead end forever.