Artificial stupidity or logimorphism? How misuse of language warps our thinking about ‘artificial intelligence’

‘Artificial intelligence’ (AI), as a blanket term, covers many advanced computing techniques that employ divergent methodologies for a wide array of functions. It is too late to affect general usage of the term without qualification—although it is a misnomer—but perhaps not too late to argue for the preferential use of more specific and realistic terminologies when discussing useful applications in science and medicine. Large language models are statistical tools for predicting text; machine learning algorithms are programs that discern patterns from data; and neural networks are mathematical models that predict outputs from inputs. In all cases, the software is unaware of what it is doing or why. The real intelligence is human, exemplified by the expertize of the engineers who designed any particular system, and by the scepticism, realism and vision of those who interpret and apply its outputs. Users may suspend their incredulity if large language models that are not sentient creatures are programmed to answer questions in the first person, since that encourages anthropomorphism. Perhaps we need a new word—which could be ‘logimorphism’—to emphasize the dangers of interacting uncritically with inanimate software programs while overtly or subconsciously ascribing them human cognitive powers. Implementing high standards when applying new computing tools in clinical research and practice should start with the avoidance of inappropriate language that degrades our thinking.