In a twist that could give AI developers a dose of their own medicine, a recent study has found that older AI chatbots show signs of cognitive decline.
This revelation, published in the BMJ on December 20, 2024, treads the fine line between human-like artificial intelligence and the inevitable limitations of aging technology.
Picture this: a world where AI is not just your go-to for diagnosing that nagging cough but a trusted ally in unraveling complex medical mysteries.
Yet, as researchers discovered, this technological utopia might have a few more roadblocks than anticipated.
As the AI landscape evolves, some of the marvels of modern tech are showing their age—literally.
The study put several well-known large language models (LLMs)—including the likes of OpenAI’s ChatGPT, Anthropic’s Sonnet, and Alphabet’s Gemini—through their paces with the Montreal Cognitive Assessment (MoCA) test.
This test, a staple in the toolkit of neurologists for detecting cognitive impairment in humans, includes tasks like drawing specific times on a clock, subtracting numbers serially, and recalling spoken words.
While fresh-faced ChatGPT 4 danced through the test with a respectable score of 26 out of 30, Gemini 1.0 stumbled in with a mere 16.
The findings suggest that older models of AI, much like their human counterparts, aren’t immune to the passage of time, showing diminished capabilities in areas such as visual/spatial skills and executive tasks.
The implications are both profound and, dare we say, a little humorous.
Imagine a future where AIs, once seen as infallible, might require their own version of cognitive therapy.
Perhaps human neurologists could find a niche market in treating these digital patients, diagnosing cases of “silicon dementia” with the same care and precision they apply to human brains.
But the study’s authors caution against drawing direct parallels between AI’s quirks and human cognitive function, emphasizing that these results are observational.
The inherent differences in how AI and the human brain operate mean that while these findings are intriguing, they don’t quite equate to a digital version of Alzheimer’s.
Nonetheless, this study serves as a reminder of the limitations of AI, especially in clinical settings.
The cautionary tale here is clear: while AI continues to make strides, particularly in tasks demanding visual abstraction and executive function, there’s still a need for human oversight.
After all, as much as we might dream of a robot doctor in every home, it seems we might have to hold onto our human practitioners a little longer.
In the end, this research offers more than just a glimpse into the future of AI; it prompts a reevaluation of our expectations.
As we grapple with the notion of digital cognition, we’re reminded that even the most advanced technology isn’t immune to growing old.
And perhaps, in this shared frailty, we find a little more humanity in our machines.
The enterprise software giant is successfully decoupling growth from human headcount by monetizing the infrastructure required to manage autonomous agents.
The property group’s latest sustainability note issuance marks a strategic pivot toward integrating environmental performance with essential housing services.
A single malware infection on a senior officer’s computer exposed the sensitive interrogation records of a Syrian National Army unit, revealing systemic security failures.
