Robots that learn and share new language
Researchers at OpenAI, which is the artificial intelligence lab created by Tesla founder Elon Musk, have taught robots to use language to complete simple tasks. ROBOTS have communicated with each other using a shared language that they made up as they went along in an amazing new experiment.
Experts at OpenAI, an artificial intelligence lab created by Tesla founder Elon Musk based in San Francisco, have taught robots to start their own language so they could help each other complete simple tasks.
In the future, researchers hope that as the language becomes more complex, they will have a “translator bot” that will be able to convert robot chat into English.
The team created a world where robots learnt a language that enabled them to complete simple tasks such as moving themselves from A to B.
Experts created a two-dimensional world to navigate and the robots — or agents — had to find their way around.
The robots relied on a technique called reinforcement learning which involved them using different sounds.
They taught agents to create language by dropping them into a set of simple worlds, giving them the ability to communicate, and then giving them goals that can be best achieved by communicating with other agent.
“If they achieve a goal, then they get rewarded,” said the team.
“With reinforcement learning, they developed a shared language to help them achieve their goals”, they said.
For example, a speaker would start to form an association between the word “house” and the image of a house.
As they build up their vocabulary, they can start using sentences that allow them to convey ideas to one another.
As the language becomes more complex, experts hope they will be able to translate what they are saying into English.
The team gave the robots two types of actions, one environment action such as looking at something; and communication actions, such as broadcasting a word to one of the other agents.
Robots were able to modify their messages to improve the effectiveness of communication.
For example, if an agent realises it could have performed something better had it had supplementary information, he will communicate this information to the next agent so its instructions are as clear as possible.
The team found that agents were able to evolve languages to fit the situation and able to accomplish tasks by communicating with one another.
“We hope that this research into growing a language will let us develop machines that have their own language tied to their own lived experience,” the researchers wrote.
“We think that if we slowly increase the complexity of their environment, and the range of actions the agents themselves are allowed to take, it’s possible they’ll create an expressive language which contains concepts beyond the basic verbs and nouns.”