In 2024, I just ended my master thesis on how to debunkt job offers biases thanks to an AI protoype (on sex bias).
My conclusion was that using AI for this task asks a lot of computing resources, leading to a list of biased words… That we already know since 2011. This is list is only two-pages long, and it would be much more efficient to train hiring professionals to write unbiased job offers.
AI basics
AI is a technology based upon statistics. An AI tool will provide the most probable answer to your question.
AI impacts
Biases
AI amplifies human biases. It tells a lot on how much the feeding data was biased. Biases are small on high quality data, checked by humans, like juridical content (laws, judgements) and medical content. Biases are hudge on unqualified content, like the web (Big data).
Unlike what is said, it has been proven that AI does not need a big amount of data to deliver proper content (Mustafa ALABADLA et al.: Systematic Review of Using Machine Learning in Imputing Missing Values, in : IEEE Access 10 (2022), Conference Name : IEEE Access, p. 44483-44502).
It has also be proven that best results are met when we ask humans to pick amongst AI propositions (source).
Human living conditions
AI asks for a hudge amount of energy and of cooling. In 2023, AI was using 25% of Microsoft energy needs. In 2024, Microsoft and Google got licensed to build up nuclear plants. In 2025, California hudge fires could not be ended because fire fighters had not enough water… because of AI. (source)
AI maturity
When to use it
Enhance human productivity: AI is a great tool. Use it wisely.
When not to use it
Replace humans (artists, employees…): AI does not buy your work. When you end up trade, you’ll starve from it too.