Can AI develop and turn out to be more complicated when put in a competitive circumstance, same to how life on Earth developed via natural selection and competition? That is a question the scientists at OpenAI have been attempting to answer via its tests, comprising its latest one that placed AI agents against one another in almost 500 Million rounds of hide-and-seek. They discovered that the AI bots or agents were capable of conjuring up various different plans as they played, designing new ones to counter methods the opposite team came up with.
At first, the seekers and the hiders simply ran through the environment. But after 25 Million rounds, the hiders discovered how to employs boxes to barricade themselves within rooms and block exits. They also learned how to collaborate with one another, passing boxes to each other to swiftly ban the exits. Then the seekers learned how to search the hiders within those forts after 75 Million games by shifting ramps against walls and employing them to get over barriers. Almost after 85 Million rounds, though, the hiders learned to take the ramp with them within the fort before banning the exits, so the seekers have no way to employ it.
On a related note, earlier Microsoft declared that it is spending $1 Billion in the Elon Musk-established firm OpenAI. The 2 firms will operate together to convey AI and supercomputing techs to Microsoft Azure. And OpenAI will operate its services only in Microsoft’s cloud. Eventually, the associates expect to develop AGI (artificial general intelligence), a tech that some claim will exceed or match human intellect.
Earlier, OpenAI has made the headlines for designing AI that can overtake human players. But it is also behind tests like improving the dexterity of robotic hands and writing believable fake news.