Many people see artificial intelligence as a valuable source of diverse information, appreciating its ability to simplify complex tasks.
However, one of the most debated characteristics of AI is its propensity to "hallucinate," a phenomenon that experts refer to as "artificial intelligence hallucinations."
This phenomenon results in the invention of information that appears credible and realistic, but which, in reality, are only unverified or completely erroneous hypotheses.
Some researchers, however, believe that these hallucinations could prove beneficial. For example, they could foster innovation and contribute to the discovery of new drugs and treatments, according to the site Psychology Today.
The site quotes John Nosta, an American expert in technology, science, medicine and innovation, who argues that the "hallucinations of artificial intelligence" can sometimes be an asset rather than a flaw. He explains that recent research provides strong evidence in favor of this idea, demonstrating how AI's uncertain assumptions can stimulate advances in fields where creativity and imagination are essential.
Hallucinations as a driver of innovation
In the field of new drug development, it is essential to identify therapeutic compounds and chemical structures, before testing them to assess their effectiveness against certain diseases.
Major advances in this field often come when researchers explore unconventional and unexpected hypotheses or ideas. This is precisely where “artificial intelligence hallucinations” can be useful.
A recent study found that hallucinations generated by various advanced language models (LLMs) designed to process, understand, and produce text can yield “novel chemical molecular structures” that transcend the limits of known data. For example, when these models are asked to come up with ideas for potential drug molecules, they sometimes suggest compounds that seem unlikely or disconnected from traditional chemistry. Yet, after careful review, some of these suggestions, initially considered hallucinations, turn out to have promising properties for developing new treatments.
Ethical issues
Despite the potential benefits of AI hallucinations, they also raise significant ethical and practical concerns.
By nature, these hallucinations are unpredictable and require rigorous validation to ensure their usefulness and safety, especially in sensitive fields such as medicine. Thus, human supervision remains essential to distinguish useful hallucinations, capable of stimulating creativity, from those that could cause harm. This is where the main challenge lies.
By our partner Nidaa al Watan (with Reuters)
