Generative AI chatbots like ChatGPT and Google Gemini are exhibiting AI cognitive decline over time, resembling human brain aging, according to a new study by Tel Aviv University researchers.
A group of data scientists and neurologists at Tel Aviv University decided to investigate how aging affects AI cognitive intelligence, finding that chatbots display symptoms similar to cognitive impairment and degradation in humans. While AI cognitive learning is relatively new, researchers are beginning to find out how these systems operate in the long term.
The study called “Age against the machine—susceptibility of large language models to cognitive impairment: cross sectional analysis”, conducted by neurologists Roy Dayan and Benjamin Uliel, tested several chatbots using examples of cognitive technologies of AI used to assess human brain function, such as short-term memory, attention, and problem-solving. They are traditionally used to diagnose conditions such as Parkinson’s disease and Alzheimer’s.
Chatbots Struggle with Cognitive Tests
Scientists observed “cognitive impairment that appears similar to neurodegenerative mechanisms in the human brain.”
The Montréal Cognitive Assessment (MoCA), a 30-point cognitive function test was among the evaluations, with GPT-4o achieving the highest score of 26/30 – seemingly strong, indicating mild AI and cognitive intelligence impairment.
Gemini 1, Google’s old chatbot model, scored only 16 score – indicating severe cognitive impairment in humans, similar to dementia, finding the performance mimicked symptoms seen in dementia patients, with the worst-performing bots performing extremely poorly.
What’s more disturbing is that some chatbots failed to complete the tests, even after explicit instructions on how to tackle them, which one its own casts doubt on whether they can improve their cognitive AI and learn for themselves over time.
A Disturbing Future for Cognitive AI
More disturbing about this report, however, is that these chatbots are not new. Gemini 1, for example, was released back in December 2023. The drastic decline in AI involves cognitive behaviors performance which suggests that AI programs have the ability to build up severe issues over time, even after only a very short period of time.
The study fuels growing concerns about the future of AI in cognitive psychology. Researchers fear that AI might have crossed a “red line” where it’s gaining power while becoming increasingly unstable.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.