One of the problems with AI

You might know that I’m the coordinator of a master’s degree, in Italy. Ennio Flaiano, an Italian screenwriter, playwright, novelist, journalist, author of numerous enlightening aphorisms, said that:

In Italy the shortest line between two points is the arabesque

So, I’ve been buried with (mostly) absurd issues for a large portion of my time in the past five years (and counting), but I’ve also had the chance to learn a lot, from many different perspectives. A situation I recently had to consider is about a deaf student that recently enrolled. In my University there is an office devoted to having an initial evaluation of the condition of students with any kind of disability, so that they have a privacy preserving document describing the measures to be taken during classes and examinations; then, of course, they monitor the situation and are available for managing issues that might arise.

Well, it is in this situation that I was struck with one of the good usages of potentially controversial AI technologies. Our classrooms are provided with a podium connected to a microphone system, cameras, and screen recording technologies; they were installed before the COVID outbreak, and of course they seemed a curious gimmick initially, then it was sadly even too clear that they would have been extremely important. We also have a cloud storage provider that luckily provides a (presumably AI based) automated captioning system. I tested it with an ongoing orientation video, in which the speakers were using Italian, and I was quite happy with the results, the biggest problems I faced were with a disappointing overall UX of the system (I had to learn my way through the menus, then I had to realize that the captions were not immediately available after they were produced, I had to manually allow users to access them). Although, of course, this does not represent a complete solution to issues raised by this form of disability, this service surely represents a good step in the direction of having more inclusive teaching materials and supports.

The technologies underlying this kind of service are largely shared with recent text to speech software that can generate natural sounding voices, also very specific voices (e.g. mine), provided one has access to a sufficiently representative sample of that voice (for instance, see this kind of limited online service). Similar considerations apply for videos as well.

It is a bit creepy to consider that there is an abundant dataset of audio recordings of my voice, mostly in Italian but also in English, as well as some videos, around the cloud… Let us just say that:

• these days, if you really want to be sure about my opinion on any kind of topic, please ask me directly;
• as I was trying to say some time ago in a previous post, we should hope in some form of more thorough regulation not just for AI related services, but for the Internet as well. Of course, I’ve got my pink glasses on… so, always consider the previous bullet!

Basically, this is one of the problems with AI and its applications these days: there might be perfectly legitimate uses of AI for the social good, but at the same time it can also represent a serious threat, in a society that relies so much on digital contents that could be simply forged, created with just small samples of genuine audio/video recordings, with tools of growing general availability, to make anyone say or do whatever we please. Please add your favorite AI generated photo, like Pope Francis with a white puffer coat or Donald Trump’s false arrest.