The Human-Machine Dialogue. Artificial Intelligence/Human Intelligence: Manipulation and Evaluation

After a first dossier on “the Brain and Learning” (Futuribles 428), followed by a second on the plasticity of the brain (issue 431), Futuribles is opening a third strand in the “Brain” series, this time on human-machine interactions and the impact of screens on the development of young people. As a specialist in human-machine interaction, Laurence Devillers presents the issues inherent in the development of “conversational agents” and other robots endowed with artificial intelligence that are increasingly found interacting with individuals in various contexts. After reminding us how these (self-)learning systems operate, she stresses how vigilant we must be about the possible manipulation of individuals by these types of interface (particularly through “nudge” techniques). She also shows how emotions are used in human-machine interactions (emotional triggers, humour etc.) and outlines the tools available today to evaluate artificial intelligence — and even to compare it to human intelligence (particularly the Turing test and its limitations). Given the rapid advances in machine learning, Devillers calls for the development of new tests for assessing machine capabilities, aimed in particular at monitoring their ability to manipulate individuals. Though technical progress is exponential, responsibility for the way its application is regulated in society and in the real world still falls — at least for the moment — upon citizens: it is up to individuals, as of now, to determine the ethical, regulatory and other frameworks within which such human-machine interfaces should be embedded.