In the next 50 years, will a computer be able to write a literary masterpiece ?
Inria / Photo H. Raguet
Jérôme Euzenat heads the Moex team, which follows on from the Exmo team, at the Inria Grenoble Rhône-Alpes Research Centre. He told us about the subjects studied by the new team. With Inria celebrating its 50th anniversary this year, he also told us what he thinks the world of computing will look like in 2067.
What is the focus of the research carried out by the Moex team ?
Jérôme Euzenat :
Our work focuses on knowledge representation, which is a sub-discipline of artificial intelligence. We human beings have knowledge that we need to be able to represent. For the last 15 years or so, with the Exmo team, we worked on the semantic web. Many libraries and statistical institutes publish public data that conforms to different ontologies. Such ontologies represent knowledge, as applied to this data, and expressed in formal language (or logic). However, there can be huge differences between all these representations, which makes communication problematic. We try to find things that match between them, so that communication can be restored. For example, we look for matches between the ontology used by a librarian and that used by a publisher.
With Moex, we have decided to tackle the problem the other way around. We will start by trying to communicate and, just as humans do, when a problem understanding the other arises, we will modify how we perceive and represent things, and thus how they will be perceived by the other person. Researchers have already succeeded in getting robots to develop a language through interacting with each other. mOeX aims to apply the same principle, not to language this time, but to knowledge representation, which is, by its very definition, hidden, and try to understand its characteristics.
What kind of applications do you think your research will have ?
Our research is very theoretical. However, as an example, imagine a French company that has products made abroad but wants the labels in French. Agents often order goods from suppliers that need to be labelled to suit the customers they are aiming to sell the products to. We could develop software agents which communicate to apply the desired labelling. They’ll be able to determine if something is incorrect and modify their knowledge accordingly if there is an error. This kind of adaptive approach should also make it possible to evolve in line with customers' changing tastes or with modifications to the supplier's services.
Looking ahead fifty years, how do you think artificial intelligence will have evolved ?
Let’s start by looking back. Thirty years ago, I started working on artificial intelligence with a completely crazy idea in mind. I wanted to teach a computer the entire works of Baudelaire and get it to finish the Petits poèmes en prose. At the time, it was totally inconceivable that this could be achieved in the medium term. Now, well, with just a bit of effort, that's totally feasible.
In fifty years' time, I wouldn't be surprised to see amazing works of literature being written by computers...
Today on the other hand, the techniques required are still quite limited, especially as regards creativity. We can get a computer to learn to write a certain type of book, but we cannot give it a particularly broad cultural knowledge. We don't know how to teach a robot to respond in all different kinds of situation. That will be a great challenge in the coming years: how to learn based on just a small amount of information. And how to learn on a broader basis, rather than one task at a time.