Teaching machines to communicate effectively with humans is said to be the Next Big Thing in tech, but while Internet of Things devices are rapidly gaining in popularity, truly effective dialogue between man and machine remains somewhat elusive — as several public gaffes involving chatbots have demonstrated, and anyone with a thick accent who has failed to get OK Google to play a song successfully can testify.
Learning to communicate: https://t.co/1oNJ90GWhI
— OpenAI (@OpenAI) March 16, 2017
However, experts at artificial intelligence research group OpenAI believe they can significantly improve a machine's ability to converse with a human being by creating a means by which machines can converse with one another. The team conducted an experiment that challenged software bots to complete a series of tasks, such as moving to a specific location, in a simple, two-dimensional virtual world — and used a technique called "reinforcement learning" to present the challenges as cooperative rather than competitive to stimulate collaboration between devices.
"We [taught] AI 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 agents. Our hypothesis was true language understanding would come from agents that learned words in combination with how they affect the world, rather than spotting patterns in a huge corpus of text," the team wrote in a blog post.
The robots proceeded to learn to collaborate and communicate via trial and error, remembering symbols, words and signals that helped them achieve goals, then storing them in a private recurrent neural network.
In short, the agents began to invent a simple, grounded and compositional language — for example, a speaker forming an association between the word "tree" and images or experiences of trees.
"If one agent realizes it could have performed a task better if a second agent had sent different information, the first agent can tell the second exactly how to modify its messages to make them as useful as possible. In other words, agents ask the question: 'how should I modify my communication output to get the most communal reward in the future?' " the post continues.
The team tweaked the experiment so that there was a slight penalty for every "utterance" by every bot, and added incentives to get tasks performed quicker — in turn, the language evolved, with robots learning to work together by composing sentences comprised of multiple words.
The researchers hope as AI language continues to develop and grow increasingly complex, they can construct a translator bot capable of interpreting these communications for humans — and grow a universal machine language understandable by humans. The researchers add that it's possible AI machines could create an expressive language, which contains concepts beyond the basic verbs and nouns that evolved in their experiment.
Scientific and technological luminaries, including Bill Gates, Stephen Hawking and Steve Wozniak, have suggested AI could be highly destructive for the world, and mankind with it. As AI has the potential to become more intelligent than humans, humans have no surefire way of predicting how it may behave in future.
Humans currently reign supreme on the Earth due to their superior intelligence — if they cease to be the most intelligent force on the planet, their continuing control cannot be assured.
AI developing means of communication independent of human input could allow such devices to create a private system of dialogue humans cannot understand — and, if they are focused on collaborating to fulfil tasks more effectively, they may see human beings as a nuisance to be circumvented or eliminated.
