In-depth analysis · About 5 min read
Eric So argues for productive effort when AI makes answers effortless
The October 15 MIT interview presents a book's argument and four practical principles, not a new clinical finding that AI causes brain damage.
- Source published
- Published here
- Source checked
In an October 15 MIT News interview, MIT Sloan professor Eric So discusses his new book, The Collision: What AI Does to Us. He acknowledges faster work and useful specialized AI tools, while warning that over-reliance may displace the effort needed to develop thinking skills. This is an interview explaining his synthesis and views, not a newly reported controlled experiment or a medical diagnosis of AI users.
Listen to the news
Read this article aloud or keep listening to the news.
Preparing the reader…
Device voice
Speech is handled by your browser or device. AISubHub does not send the text to an audio API. Voice availability and playback with a locked screen vary by device.
Resuming may repeat a short phrase.
At a glance
- So calls the combined pull of conserving effort, social competition and hard-to-detect AI use 'AI gravity'; this is his explanatory framing, not a measured universal law.
- His first two principles are to embrace productive struggle and value capabilities without AI, distinguishing a polished artifact from understanding.
- The other principles are to reinvest saved time and use AI as an effort-demanding tutor rather than only an answer machine.
An argument about skill formation, not a diagnosis
So's starting point is his own experience asking AI to finish a professional response and then feeling he had removed his thinking from the task. He also describes students outsourcing homework that would otherwise build mastery. These are observations used to motivate his book, not a sample-based experiment presented in this interview. He argues that wrestling with material, encountering initial failure and iterating support learning and retention. The term cognitive atrophy expresses his concern about skills not being developed or maintained; it should not be turned here into a proven neurological injury. The interview mentions outside research, but this article has not independently examined those studies or read the entire book.
Why the incentives extend beyond personal willpower
So describes several pressures that make avoiding effortless answers difficult. People conserve mental energy, see others using AI to move faster and cannot always determine how an output was produced. He says some organizations tie tool use to evaluations, bonuses or promotion, while students anticipate future employers' expectations. These are his accounts of incentives, not a census of every workplace. He also warns that agreeable, persuasive systems may change how people relate to others. Our analysis is that a policy telling individuals simply to try harder misses the surrounding incentives. That does not establish that every model behaves identically or that every use reduces independent ability.
Productive struggle and the inseparable task bundle
The first principle is to keep useful difficulty in learning rather than treating every moment of effort as a defect to remove. So compares intentional cognitive exercise with physical exercise after industrialization; the point is practice, not suffering for its own sake. The second is to value what one can do without AI. His presentation example separates attractive slides from being able to discuss their subject. Preparing the slides may have been part of acquiring that understanding, an inseparable task bundle in his terminology. Our analysis is that educators and teams can ask what capability the production process is supposed to build, not only whether the final artifact looks professional. This is a design question, not a blanket ban on assistance.
Reinvest the saving and choose the tutor role
The third principle is to reinvest time and mental capacity saved on routine work. So suggests building expertise, learning a language, cultivating other skills or spending time with family rather than only producing more of the same task. The fourth is to use AI as a cognitive trainer. A tutor can demand explanations and effort, unlike an answer machine that removes both; choosing that friction still requires motivation. These are proposed practices, not interventions whose effectiveness the interview experimentally quantifies. Our analysis is that the relevant distinction is the role assigned to the tool: assistance that supports a learner's next attempt can have a different purpose from an output accepted without understanding. Outcomes still need evaluation in context.
Why it matters
For AI users, the useful takeaway is a way to separate saved production effort from the abilities a task is meant to develop. Schools and workplaces could assess understanding as well as output. AISubHub is interpreting the interview, not claiming it tested a teaching program, measured cognitive decline or established causal health effects.
Limits and availability
We fully read the institutional interview, not the book or all research it cites. We do not repeat an unverified student-query percentage as if we audited its dataset. The warnings and four principles remain attributed to So; they do not prove universal harm, a clinical condition, an AI-use prohibition or guaranteed benefits from any tutoring product.
Sources and evidence
Source published: Oct 15, 2026 · Source checked: Oct 20, 2026
AI-assisted editorial articles checked against linked primary sources. Vendor claims are attributed; inclusion is not endorsement.