The Hidden Toll of Frictionless Tech: What AI Does to Our Minds

MIT Sloan professor Eric C. So’s book, The Collision: What AI Does to Us, warns that overreliance on AI may weaken users’ cognitive skills, hinder children’s foundational learning, and create false confidence in workplace output. So also cites improved outcomes in a teacher-supported Microsoft Copilot study in Nigeria and argues that AI is most useful as a thought partner rather than a substitute for thinking. The article offers a four-part framework for using AI while preserving human capabilities.

Analysis

This is a demand-quality issue for AI vendors, not evidence of weaker near-term AI adoption. If organizations conclude that unstructured answer generation undermines learning, judgment, or accountability, procurement could shift toward products that preserve user effort: guided tutoring, source-grounded workflows, review checkpoints, and usage controls. That could favor enterprise platforms able to integrate AI into supervised work processes, including Microsoft, while making raw chatbot engagement a less durable product metric. The offset is that more guardrails may increase friction and reduce usage or seat-level value if users perceive them as obstruction.

The article’s educational example is not enough to establish a general efficacy advantage for Copilot: study design, comparison group, duration, and persistence of outcomes are unspecified. Treat the claims about cognitive harm and improved learning as hypotheses, not measured impacts on Microsoft revenue or margins. Over the next 1–3 months, watch for school and enterprise procurement criteria, product changes emphasizing verification or tutoring, and disclosure of measurable customer productivity outcomes. Over 6–18 months, the key risk is that AI value is judged by durable skill transfer and error reduction rather than time saved or generated output. A counter-scenario is that organizations accept cognitive offloading as a labor-cost trade-off, sustaining adoption despite these concerns.

No standalone trade follows from a book announcement. The contrarian angle is that concern about overreliance could strengthen demand for AI platforms with workflow controls rather than suppress AI spend broadly. This thesis weakens if buyers continue expanding access without requiring verification, or if controlled deployments fail to show better quality or learning outcomes.

Key Decisions for Investors

• Do not trade MSFT on this item alone; it supplies no new evidence of revenue, margin, or adoption impact.
• Put MSFT on watch for product and procurement signals: evidence of Copilot usage tied to verified task quality, learning retention, or error reduction would support a differentiated enterprise-value thesis.
• Track school and corporate AI policies over the next 1–3 months. A shift toward supervised, source-grounded use is a potential positive for integrated platforms; restrictions focused on AI use itself would be a demand risk.
• Before treating the education example as a commercial signal, verify the study’s sample, control group, duration, and follow-up outcomes. Weak or nonpersistent results would undercut claims that guided AI use can reliably improve learning.