Last year Scientists at the Google’s DeepMind AI conducted a study to analyze how their AI would react in social situations. The purpose of this study was to observe that whether the AI would willingly be co-operative or highly aggressive in a competitive situation.
For this study The Google team developed a simple 2D game namely ‘Fruit gathering’. A computer game that asks the two AI characters (known as agents) to compete against each other to collect as much virtual apples as they could. The agents were also equipped with laser beams to knock out their opponents temporarily.
Over 40 million rounds of this game were conducted and it was found that the agents were peaceful as long as there were enough apples, but when the apples began to diminish, the agents got highly aggressive and started using laser beams at each other to knock down the other one out and gather all the apples.
The DeepMind team reported : “We let the agents play this game many thousands of times and let them learn how to behave rationally using deep multi-agent reinforcement learning. Rather naturally, when there are enough apples in the environment, the agents learn to peacefully coexist and collect as many apples as they can. However, as the number of apples is reduced, the agents learn that it may be better for them to tag the other agent to give themselves time on their own to collect the scarce apples.”
This showed that the human like behavior emerge as a product of the environment and learning. It was also noticed that more complex and intelligent AI agents were less co-operative and got aggressive sooner and easily.
The team then developed another game namely ‘Wolfpack’. In this game there were three AI characters, two of them played as wolves chased the third AI character – the prey. Here the team focused on co-operation, so if both the wolves were near when it was captured, they both were rewarded.
“The idea is that the prey is dangerous – a lone wolf can overcome it, but is at risk of losing the carcass to scavengers. However, when the two wolves capture the prey together, they can better protect the carcass from scavengers, and hence receive a higher reward,” the team reported in the paper.
While these are just computer games it is learned that the AI agents showed aggression and selfishness in particular environment but they can also be co-operative for their self interest.
Joel Weibo reports at Wired :”This model also shows that some aspects of human-like behaviour emerge as a product of the environment and learning.”
He further said : “Say you want to know what the impact on traffic patterns would be if you installed a traffic light at a specific intersection. You could try out the experiment in the model first and get a reasonable idea of how an agent would adapt.”
This type of research may help us in the future in real world application to better control of the economy, traffic and environmental challenges. For this we need to create agents that co-operate with each other and we also need to induce helpful nature into our machines.
On other hand Stephen Hawking and Elon musk, the founder of Tesla warned us about AI by saying, “The development of full artificial intelligence could spell the end of the human race.”
So far Google has set up the ethics and safety to avoid Skynet kind of situation. Google also said that there will be a “Kill switch” button for such emergencies. However these AI systems are still far off compared to human brain so robots ending human race is not a threat.