- Too much of a good thing, even ChatGPT, can be bad, according to research from OpenAI and MIT.
- Relying on it for common uses like advice, explanations, or ideas can foster dependency.
- The research asks whether it’ll lead people to a “loss of agency and confidence in their decisions.”
The former UK Prime Minister Benjamin Disraeli — who lived and died in the 19th century and left a legacy in politics and literature —couldn’t have predicted how AI would reshape the world. However, he may have grasped its implications better than some people today.
“Moderation is the center wherein all philosophies, both human and divine, meet,” he’s believed to have once said.
That advice might serve some ChatGPT users well given a new study that MIT Media Lab published in partnership with OpenAI on Friday. The researchers studied nearly 1,000 people on how they used ChatGPT over four weeks and found that some people overuse the technology — which could have repercussions on their sense of self.
Users who often turned to the bot for non-personal conversations, including seeking advice or suggestions, conceptual explanations, and assistance with idea generation and brainstorming — which is a common use case — had a higher likelihood of becoming emotionally dependent on it.
Those over-users were also more likely to engage in “problematic use,” which the researchers defined as “addictive behaviors and compulsive use that ultimately results in negative consequences for both physical and psychosocial well-being.”
These users were already primed to see ChatGPT as a friend, have a high level of trust in it and feel that the chatbot would be affected by and worried about their emotions.
The authors said future research could investigate whether this leads to a kind of “cognitive dependence where users increasingly rely on AI systems for decision-making and problem-solving rather than direct emotional support perhaps leading to loss of agency and confidence in their decisions.”
A common refrain among AI researchers is that the technology will augment, not replace, human workers.
“Automation does not equal autonomy,” Avijit Ghosh, an applied policy researcher at Hugging Face, previously told BI. “Repetitive tasks are being automated, but the idea of a fully independent AI workforce? That’s still speculative at best.”
The ideal is a meeting between humans and artificial intelligence that will produce something greater than the sum of its parts. But the race to adopt AI has posed a conundrum. Workers are worried they’ll fall behind their peers if they don’t use the technology enough. If they use it too much, though, researchers are worried they’ll risk losing their sense of self.
Whether it’s divine or simply the next technological frontier, moderation is still a virtue — even in the AI age.
