Julien Girard-Satabin @rhapsodos@scholar.social

And don't let me started on how we consider the data in the overall "trustworthy AI things": it's quite overlooked, really. We just collect data from the wild that was constituted using microlabour platforms that create a new proletariat, we scrape data from the web without considering any consent framework. How to integrate the data (or a representation of the environment) into our research, and how to do it well (eg without killing the planet and enforcing marginalized groups exclusion)?

Another topic is the explainability and accountability of artificial intelligence program. how to ensure stakeholders can query a program / the program's owner and require accountability for how they were categorized by the program?

But it only addresses a small part of the topic.

For instance, regarding the privacy of a data-based program. Those big monsters tend to do a lot of memorization (and they were trained in what a consider an unrespectful way, but I digress) . Is there a way to quantify how much they memorized? Could we "patch" those programs, or design them in a better way so that you don't end up finding personal data while querying such program?

Basically you have the whole "formal safety" stuff I made my PhD on. Since data-based programs are so huge, it's very complicated to have sound guarantees that they behave according to their specification. But it also comes from the fact that it tries to work on very abstract, intricated specifications. For instance, "do I need to pet my cat again today, or go out to buy them treat?" require to define the very high-level concepts of "cats", "treats", "outside", and combine them together.

Sorry for the lack of activity by the way, I had my PhD defense in november and had quite some trouble to adapt to my new working environment. But now it's becoming better, so I'm back!

I'm now a trustworthy artificial intelligence researcher and, huh, part of the work is to figure out exactly what "trustworthy" means?

My parents (and some of my colleagues) are teasing me, basically "I deserved it by taking the train and not the plane".

Yeah, sorry for trying to not kill us even more with unnecessary carbon emissions I guess.

There's more of it though. I decided to go to the conference venue (Wien) in train, rather than plane. From Paris to Wien, it's about 11h of train. Aaaand I missed a connection in München, so I will probably not be at my hotel before 23h :')

On my way to the IJCAI-ECAI conference, presenting at the AISafety Workshop (although I will totally try to attend to the whole conference).

First onsite conference since three years, I must say I'm both happy and stressed.

Just sent my manuscript to my comitee, AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA

Hello,

For all women in cybersecurity, here is a workshop that might be of interest: https://casa.rub.de/en/wisc-workshop