AI addiction is like "drug addiction"! MIT's latest research found that long-term reliance on large models can lead to decreased learning ability, brain damage, and a 47% reduction in neural connections. The claim that AI improves efficiency may be a misunderstanding!
ChatGPT is "draining" your brain!
MIT has completed the first brain scan study of ChatGPT users, and the results are jaw-dropping.
In response to the impact of using ChatGPT in daily writing, this time the AI version of the "devil's deal" was revealed with data:
Relying on AI writing is equivalent to using long-term thinking ability in exchange for short-term efficiency.
It can be called "selling the soul"!
Researchers from the MIT Media Lab spent three months studying the cognitive cost of LLM in depth, revealing a pressing issue:
Learning ability may decline due to the use of LLM.
Although the use of LLM brought obvious efficiency advantages in the early stage, the 4-month experiment found that the LLM group lagged behind the control group that "wrote only with the brain" in terms of neural activity, language quality and scoring.
In short, the past claims about AI improving productivity may be all wrong!
▲ Participants in the experiment wore Enobio EEG headsets and AttentivU concentration monitoring headsets, and used BioSignal Recorder software to record data
Deal with the devil
AI "sucks the brain dry"
Using ChatGPT, it only takes a few minutes to complete daily writing, which is convenient and fast, but at what cost?
83.3% of ChatGP users cannot cite what they wrote, even if it is just a paper completed a few minutes ago.
Think about it:
You finish writing, save, and then forget it-because from beginning to end, it is ChatGPT that thinks, not you.
▲Percentage of participants in each group who could not recall any citations from Session 1 papers
Brain scans reveal the damage of using AI: the number of neural connections in the brain plummeted from 79 to just 42.
That’s a 47% reduction.
If a computer loses half its processing power, everyone will think it’s broken. The same thing is happening to the brains of ChatGPT users!
Unassisted writing (pure brain group) showed stronger neural connection strength in all measured frequency bands, with particularly significant increases in theta and high alpha bands.
After reading the papers written by students using AI, teachers don’t know which articles used AI, but they can feel that something is wrong:
- No soul.
- Empty.
- The language is close to perfect, but there is no real insight.
Even if the human brain cannot explicitly say that cognitive debt exists, it can still detect its impact.
The scary thing is: when researchers asked ChatGPT users to write without AI assistance, they performed worse than those who had never used AI.
This is not just dependence, but cognitive atrophy. Like a muscle that has forgotten how to work.
The MIT team conducted electroencephalogram (EEG) brain scans on 54 participants for four months.
They tracked alpha waves (creative processing), beta waves (active thinking), and neural connection patterns, and found brain damage caused by overuse of AI.
This is not an opinion, but a measurable experiment.
****
It turns out that AI does not make people more efficient, but makes people lazy to think and cognitively bankrupt!
This time, researchers also discovered the AI productivity paradox:
- There is no doubt that ChatGPT makes people complete tasks 60% faster.
- But it also reduces the "Germane Cognitive Load" required for real learning - up to 32%.
- This is a long-term brain capacity in exchange for short-term efficiency.
- Companies that celebrate AI's improved work efficiency are inadvertently creating teams with weaker cognitive abilities.
- Employees become dependent on tools they cannot leave, while their ability to think independently decreases.
Many recent studies have highlighted the same problem.
Earlier this year, Microsoft conducted a similar study, and the relevant conclusions have been widely reported in the media:
MIT researchers call this phenomenon "cognitive debt" - the brain version of technical debt:
Every time you take a shortcut with AI, you are paying "interest" with your future thinking ability. Just like financial debt, this debt will be repaid sooner or later.
Perhaps the most noteworthy finding in the study is that participants in the LLM-to-Brain group showed a clear tendency to narrow their thinking.
But there is good news -
The surprising finding of the fourth phase of the study: "High baseline cognizers" (people with strong thinking ability) actually improved their brain neural connectivity when using AI - AI became their "cognitive enhancer".
But when "long-term dependents" were forced to work without AI, they performed even worse than "people who had never used AI" - their basic cognitive abilities showed a "use it or lose it" type of degradation.
The cost of using AI
It’s all “cognitive debt”
We are at a critical turning point in technological development, and we must seriously and comprehensively understand the cognitive impact of introducing large language models (LLMs) into education and information environments.
These tools do provide unprecedented convenience for learning and obtaining information, but their potential impact on people’s cognitive development, critical thinking, and independent thinking ability deserves our high attention and continued in-depth research.
Research shows that compared with using search engines, LLMs significantly reduce the thinking cost required by participants when answering questions.
But this “effort saving” comes at a cost: participants are less inclined to question or think deeply about the answers provided by LLMs-
These so-called “views” are actually just probabilistic results generated based on training data.
This situation is worrying because the “echo chamber effect” that was originally prevalent in social media is now continuing in AI tools-
The content that users are exposed to is increasingly influenced by algorithmic recommendation mechanisms, and behind these mechanisms are the priorities of the companies and shareholders represented by the models.
In the interview, only a few participants said that they did not follow the "ideas" of LLM to write, but insisted on their own ideas and thinking paths.
From an ethical point of view, participants who wrote only with their brains not only had higher satisfaction, but their EEGs also showed stronger brain connectivity.
In contrast, participants who used LLM to assist in writing also had obvious difficulties in recalling or citing their own writing content, had a lower sense of ownership of their own articles, and spent less time.
Before LLM is truly widely accepted and regarded as a positive tool, it is necessary to conduct long-term follow-up studies to fully evaluate its deep impact on human thinking ability and brain development.
▲ Comparison of alpha band dynamic direct transfer function (dDTF) EEG analysis results (LLM group, search engine Search group, pure brain group), and annotated with significance levels (* indicates moderate significance, ** indicates high significance)
Solution
The solution is not to ban artificial intelligence, but to use it strategically.
The choice is in your hands:
Incur cognitive debt and become AI-dependent. Or improve cognitive capabilities and become an AI multiplier.
