Robots Have 'Feelings' Too: Google Deepmind Can Get Aggressive at Will Depending on Stress
Google's DeepMind artificial intelligence has made a major advancement in helping scientists understant the inner workings of AI systems. A recent experiment observed that the Google DeepMind is fully capable of expressing aggression in tight situations in order to win.
Using DeepMind and a fruit-gathering game, Science Alert said the research by Google instructed the AIs to gather as much "apples" in the field as they could. During the experiment, the two AIs showed signs of aggression towards each other when the "apples" were starting to run out.
Results showed that the AIs attacked each other despite having no extra reward for it -- except for knocking out the other to allow the other AI to collect more apples.
Gizmodo said the signs of "sabotage" and "aggression," especially for complex DeepMind AI models, is different from the behavior of smaller models, which opted for a more "peaceful" gathering session.
A team member, Joel Liebo, told Wired that this might be because a more complex AI is able to learn from its environment. This "human" quality also paved way for patterns such as greed and aggression.
Another game played during the experiment is called Wolfpack, this time involving three AIs. Two of them are wolves, and the other one is prey. The game actively encourages cooperation; if both wolves are near the prey when it was captured, they both receive a reward.
Both models are still very simple "simulations" for AI, but it clearly shows that AIs are capable of being peaceful and aggressive, depending on the environment they are in.
This is not the first time that Google DeepMind showed human-like characteristics. Last year, DeepMind beat some of the world's best Go players.
Despite these recent leaps in technology, Stephen Hawking recently warned about the complications of robots. Hawking theorized that the rise of AIs may lead to a revolution.