Ethics
In the context of our collaborative work with the INA and the French National Assembly on the restoration of Robert Badinter’s 1981 speech based on GAN technology, real issues arise regarding the use of AI-generated images. We are working on qualified databases that respect the scientific data from the speech. The training images are all of optimal quality, taken from a corpus validated by the INA and scrupulously respecting the data of Robert Badinter in this period of his life.
We also carried out high-definition photographic campaigns in the hemicycle of the National Assembly, with the aim of reducing any bias and avoiding any interpretation of the artificial intelligence.
We try to get as close as possible to the truth.
Eco-responsibility
AI training is a major environmental issue. To this end, we optimise our algorithms to be as efficient as possible and implement structured working methods to avoid unnecessary training, in order to significantly reduce our carbon footprint.
We also use solutions to reduce the consumption of computing servers, notably by preferring local approaches. Indeed, we have invested in our own computers in order to carry out these drives locally in France, without having to travel thousands of kilometres to reach Google or Amazon servers. Thus, we optimise their energy consumption, notably by maximising the size of the batches and optimising the depth of the network.
To help us in this step, we are incubated by NVidia via their “Inception” programme, which gives us access to their latest tools and expertise to design the greenest possible computers.
Finally, in our energy sourcing, we favour supplies from renewable energies.