We knew about soul mates, now researchers are developing digital twins, with artificial intelligence that would clone our personality.
Once again, when it comes to artificial intelligence, the fundamental question is what data is used to train the algorithmic model. This is what will contribute to the performance of the results obtained. Researchers from Stanford University in the United States with teams from Alphabet's AI entity, Google DeepMind, published a study at the end of 2024 aimed at manufacturing replicas of the personalities of completely ordinary individuals.
It all starts with writing a very personal questionnaire. The subject answers questions about notable memories, key moments of his childhood, the stages of his professional career, his opinions on social subjects… In two hours, the aim is to collect as many elements as possible that led to to shape the person she is today. The scientists questioned a thousand people using a multitude of different criteria: age, level of training, regional origin, political affiliation, etc.
The experts are clear: nothing beats face-to-face interviews to obtain more value-added information than when presenting a digitized form. This very human step seems decisive in the quality of the data thus collected.
The tests carried out then show an 85% concordance between the answers given by the clones and their human original, to a battery of examinations: logic games, surveys on social facts, personality tests… Obviously, this mechanism is still imperfect for elaborate conceptual or philosophical points of view. Here, experimental profiling remains quite basic. And this would require very significant technical resources, if we wanted to deploy their use beyond this scientific exercise.
Nevertheless, this illustrates the ability to approach the mechanisms of human reasoning. With obviously risks regarding deviant uses to usurp identities, if we combined this technology with deepfakes which imitate the image or gestures in video.
Since October 2022, with ChatGPT we have known generative AI which allows us to produce images, texts, sounds or videos. Here we are talking about agentic AI. These “agents” will not carry out just one task, such as writing a paragraph, but carry out a sequence of actions.
Like for example when you ask him to write an e-mail, after having made the report of the meeting, and checked in the agendas the availability of people for the next session. This involves breaking down a complex action into a series of simple sequences. With applications that are sure to multiply in 2025!
