We collected 50,000 newspaper articles published in the last ten years that deal with AI. We then organized a workshop at the Medialab in Science Po with journalists and scientists to explore and understand everything that people have been saying about AI.
We projected these 50 000 articles onto a two-dimensional space, based on their semantic proximity.
This simple exercise makes one aware of the mass of what has been produced on the issue of AI. But above all, it allows us to realize that it is impossible to read all this information, even in a superficial way. The conversation around AI covers hundreds of different issues.
It is necessary to organize it, to summarize it, to structure it. Thanks to recent developments in Natural Language Processing and Deep Learning, it is now possible to group thousands of documents according to their semantic proximity and identify very quickly what makes them similar. In yellow (top), we can see the emergence of a cluster related to health issues.
You can then zoom in, and look at what these articles are about. The more yellow a point is, the more it talks about the central theme of the cluster (i.e. health issue in AI).
And we can also’ fold’ this space in the dimensions that interest us, such as whether the issue is positive (promises) or negative (criticisms) (horiztontal dimension), or involves humans or machines (vertical dimension).
If we take another cluster, such as racism and anti-semitism, we can see that the articles are distributed differently, in a much more negative way.
With Bunka, people can thus grasp at a glance what are the main issues about AI, what people think of these issues, and they can then explore further in one direction, and zoom in on a particular sub-issues. Instead of “searching” about AI, they are “exploring” about AI.