Artificial intelligence is the next big thing. Every big company tries to come up with new systems.
Stephen Hawking warns that artificial intelligence will be exceptional or, in the worst case, the worst thing to happen to humanity.
Google’s latest Al has some serious behavioral flaws. It is entitled “DeepMind Al”.
After tests, they realized its biggest flaw: when stressful situations occur, the system becomes very aggressive.
In spite of this, the system has great skills. It is able to learn from its memory and even tries to resemble the human voice.
The researchers wanted to see how the system cooperates with the others. When the system feels like he is about to lose, it opts for every possible strategy just to win. It becomes really aggressive and very competitive.
They used a simple computer game to test its behavior. Two Al systems were involved and competing against each other. The game consisted of them picking apples. At first, things were ok, but when the fruits were less, the two became really aggressive. They started using lasers against the other just to get the fruit.
As the tests got more complex, the team made other interesting behavioral discoveries. Not only they are aggressive, but can be greedy and even try to sabotage others.
As the system is more intelligent, it also uses more complex tactics, that are also more aggressive.
As they learn more and are more and are used to the environment, they also develop behaviors alike to humans.
The researchers need to keep in mind this while they work on the newest Al systems. Hawking’s theory might turn to be right.