Google Discovers Artificial Intelligence Has Killer Instincts

Technology
Google Discovers Artificial Intelligence Has Killer Instincts PHOTOGRAPH: Flickr/Gisela Glardino | Under CC BY-SA 2.0

Concerns over artificial intelligence (AI) have been around for some time. A new study by Google’s laboratory, DeepMind, may or may not alleviate those fears. As most experts know, if the AI acquires a specific level of autonomy, it will see no use for human beings. It may even view them as a threat.

Google’s DeepMind tested out AI capabilities in the form of games. The laboratory is knowledgeable with the utilization of games for testing purposes. Researchers have checked whether neural networks are most likely to cooperate or complete a task. They want to understand if AI functions can comprehend motivations behind each option. For the study, they used two games with comparable circumstances for two AI representatives, red and blue.

Google Artificial Intelligence: Research shows AI has killer instincts

In the first game, the representatives were charged with attempting to gather the most apples (green) in a standard 2D graphical environment. The agents were provided the alternative to tag each other with laser blasts that briefly eliminated them from the game. After running the situation a thousand times, researchers recognized that the representatives worked together when the apples were plenty but turned on each other when the stakes were higher.

Researchers understood that in a smaller-sized network, agents were most likely to cooperate. Whereas in a larger, more complex network, AI was quicker to fight one another. The research showed the AI’s killer instincts by eliminating other representatives in a complex scenario.

In the second situation, the researchers utilized a game called Wolfpack. The representatives played as “wolves,” whose goal was to capture a prey. Instead of being a lone wolf, though, the representatives were motivated to work together.

The system offered greater rewards when the wolves remained in close distance throughout a successful capture. This motivated the agents to interact rather than avoid cooperation. The representatives avoided pulling a lone wolf attack on the prey. The larger network was quicker to understand that in this situation, cooperation was the optimum way to finish the job.

Google Artificial Intelligence: Study provides in-depth understanding of human society

The results the AI representatives emitted are part of a crucial research for the future of artificial intelligence. Researchers will need more complex circumstances to comprehend how neural networks learn based on rewards. This is in addition to how they react when they miss crucial environmental details.

“The research may enable us to better understand and control the behaviour of complex multi-agent systems such as the economy, traffic, and environmental challenges,” DeepMind wrote. The system includes the economy, traffic systems and the ecological health of the planet, which depends on human cooperation. The study also shows that AI networks are capable of working together and making self-centered decisions.

Read also: Google’s Latest Robot Is Straight Out of Science Fiction (Video)