Luc Steels is Professor C-3PO, he’s the human-robot language expert. His work at the VUB and Sony Computer Science Lab Paris is all about language and how to get robots to talk to each other and to talk to us, their human makers and users. Previous understandings of artificial intelligence have depended on a version of the old programmers maxim ‘garbage in, garbage out’ meaning if you program poorly you’ll get useless results. For robots you could say they’re limited by the sophistication of their creators and can only communicate using a form of language designed to imitate human speech. That’s why robots have vaguely human sounding voices, human type listening systems and movable limbs (for gestures and body language cues).
Steels reckons the way to get machines to communicate is to think about how language evolves. This means doing lab experiments and simulating how we build grammar rules and syntax from words and most importantly from context. We talk differently whether at work or at home, whether speaking to a ticket collector or to kids. Language also fundamentally changes over time, new words (such as crowdsourcing or tweet) appear and of course grammar changes too, many people have observed how writing is becoming more conversational for example. Trying to design robots that can adapt to the constantly shifting sands of personal communication is still a major challenge with some extremely complex problems to solve.
One of the hardest nuts to crack is that our own mental models of language, communication and speech depend strongly on whom we’re interacting with and how we learned them in the first place. Reproducing the circumstances of interpersonal communication is just too difficult to do with machines in a computer science lab – even with the social skills of geeks at full throttle!
The name of the game is self organisation. Our natural ability to learn language from our parents or teachers is a subconscious form of self organisation. Thank goodness it is, if we had to spend our time working out how to extract grammar rules from speaking, we’d be struck dumb within hours. Professor Steels has worked out a way however for robots to do the same thing, mostly by using the idea of putting machines in real world learning situations with one machine in dialogue with another. Both robots look at a board of symbols and over time one teaches the other what the symbols mean and eventually they come up with a language.
The significance of Luc Steels work for our deep future could be more natural, independent communication between machines and people. More useful robots then would be able to hold a conversation, be language agnostic, carry out more complex tasks and ultimately evolve to equal status with human intelligence. Open the pod bay doors Hal.
Luc Steels will be at TEDx Brussels November 22
John Fass Art Director TEDx Brussels