Artificial Intelligence is neither good nor bad, it is what you make of it.
Faced with the increasing use of these very advanced computer programs, companies must both seize them to remain competitive and control them, to avoid bias.
To meet this double challenge, the Impact AI collective is publishing a white paper “Trustworthy AI”.
“
A luxury hotel group wanted to set up an AI-based facial recognition system so that its customers were immediately recognized.
It was a good intention.
But it was also necessary to immediately ask questions and identify the problems that could arise from them
”, illustrates Agnes Van de Walle, President of Impact AI and in charge of partnerships at Microsoft France.
Finally, the hotel group did not complete its project because it "
had governance capable of asking the right questions
".
The establishment of a governance dedicated to the use of AI is one of Impact IA's recommendations.
It is even a "
prerequisite for the implementation of artificial intelligence
".
This applies to all businesses, regardless of their size.
“
The use of AI should not be managed by IT departments.
It is a decision that involves the whole company, regardless of its size.
It's operational,
”adds Marcin Detyniecki, in charge of the IA working group responsible for impact AI and responsible for research at AXA
.
Concrete measures
The authors of the Impact IA report are not fooled, however.
They believe that "there
are already more than 200 charters and texts aimed at providing an ethical framework for artificial intelligence, so that a real risk of ethical-washing has appeared
".
Ethical-washing, as we speak of green-washing in the environmental field.
We must therefore be able to go beyond declarations of intent and ensure the implementation of concrete measures, in particular to avoid biases linked to the use of AI.
Behind these major principles lie concrete measures.
For example, at the design stage, consider whether or not the data that will be used by AI is private and how to guarantee its protection.
It can also be decided that certain data will only leave an enclosure in an anonymized, aggregated or encrypted form ... It is also recommended to set up a cell to avoid detecting and correcting potential errors.
Human intervention and its understanding of the functioning and the decision-making world of AI remains essential to avoid bias.
Just as it is increasingly essential for companies to switch to AI to stay in the race.
In the world of insurance, banking ... AIs are already essential, particularly in risk assessments.
It's all a question of balance.
The authors of the report also insist on the importance of the implementation of these ethical practices in order to respond to the concerns of end customers.