A study from MIT (Massachusetts Institute of Technology) has found that the human brain not only works less hard when using an LLM, but its effects continue, negatively affecting mental activity in future work.
The researchers used a limited number of subjects for their experiments (a limitation stated in the paper [PDF]), who were asked to write essays on a variety of subjects. One group of subjects was allowed to use AI (ChatGPT was chosen; researchers considered there was little difference between it and its competitors), the second permitted to use Google Search, and the third group was termed ‘brain only’ – that is, producing work with no technology aids.
Electroencephalography (EEG) was used on all the subjects to monitor brain activity to asses cognitive engagement and load. The researchers found that the groups exhibited different levels of neural connectivity, which reflected different strategies employed by the brain to write up the assignments. The more support the subjects had, the less hard their brains seemed to work. EEG analysis showed that the most active grey matter belonged to the unaided group, with less neural activity in the ‘search engine group’, and least of all among the AI users.
The study also examined what it termed ‘ownership’ – the ability for the authors to quote what they had written afterwards and summarise their work. Levels of ownership fell dramatically with the more help the subjects received from technology. Few students using an LLM were able to reliably quote what they had written. Additionally, the LLM-using group “produced statistically homogeneous essays within each topic, showing significantly less deviation compared to the other groups.”
Unsurprisingly, the visual cortex of those using a search engine or ChatGPT was more active, with those groups “more inclined to focus on the output of the tools they were using,” the paper states.
Longer-term effects
After several rounds of essay-writing, two more groups were formed from the participating subjects, comprising of ‘Brain-to-LLM’ and ‘LLM-to-Brain’, which as the names suggest, were subjects that had previously had no technological aids now able to use an LLM, and LLM users henceforth instructed to complete assignments ‘solo.’
The researchers found that, “LLM-to-Brain participants showed weaker neural connectivity and under-engagement of alpha and beta networks; and the Brain-to-LLM participants demonstrated higher memory recall, and re‑engagement of widespread occipito-parietal and prefrontal nodes. […] This suggests that AI-supported re-engagement invoked high levels of cognitive integration, memory reactivation, and top-down control.”
In short, humans using their brains to tackle a subject can benefit from using an AI after they have already fully-explored their thoughts, experience, knowledge, and feelings without using technology. But those using AI from the outset show reduced brain activity over time, and were less able to perform cognitive tasks when asked to go ChatGPT-free.
The paper states, “As we demonstrated over the course of four months, the LLM group’s participants performed worse than their counterparts in the brain-only group at all levels: neural, linguistic, [and] scoring.”
Limited study
With only a few dozen subjects in the study, the research group were working with a limited sample. The authors admit it will be necessary to use more volunteers that have a more diverse range of backgrounds for more statistically-reliable findings to be uncovered. Yet as AI is used increasingly in schools, colleges, and everyday life, the researchers have highlighted what they term a “pressing matter” of a “likely decrease in learning skills” that come about as a result of using AI as a replacement for humans’ brains.
Conclusions
If the trend of using ChatGPT in place of the very human activities of thinking, considering, and summarising continues, it seems likely that the ability to think effectively will diminish into the longer term. Having an AI add context or additional material later in any process of intellectual consideration produces better results than its use from the outset.
Search engine use fell into the middle ground between unaided thought and being spoon-fed AI-generated materials, according to the paper. However the need by Google, Microsoft, et al. to insert AI-generation into users’ search results (LLM results appearing uppermost on SERPs [search engine results pages]) means that cognitive activity among everyday search users may decline, should they only focus on AI-generated search results.
The research group states that more study is required to understand the long-term effects of AIs on the brain, “before LLMs are recognised as something that is net positive for […] humans.”
(Image source: “Cognitive testing” by Nestlé is licensed under CC BY-NC-ND 2.0.)
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information.
AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
