The Wall Street Journal, in Christopher Mims's piece "What AI Can't—or Shouldn't—Do for You," argues that the notion that AI outperforms humans at nearly everything is false, and warns against the rush by companies to apply AI everywhere.
Essay · On AI's Practical Limits
When AI Does More Harm Than Good
The new conventional wisdom—that AI can do almost anything better than people—isn't true. Where the job calls for empathy, authenticity or transparency , handing it to a machine can backfire.
Empathy
Customer care where tone and trust matter
Authenticity
Marketing that must feel genuinely human
Transparency
Where regulators demand accountability
Where humans should stay in the loop
AI assists
Summarize an angry customer's message
Draft a first reply
Gather and organize context
→
Human decides
The final tone of the reply
Legal, medical & confidential advice
High-value quotes & sales follow-ups
Personnel evaluations
A historical parallel
Project Plowshares once tried to use nuclear explosions for civil engineering. The lesson: what is technically possible and what should be done are two different things.
Broadly welcomed
Ask not what AI can do, but what it should not be allowed to do—a framing practitioners find useful.
Caveats on the premise
Tool-using systems can exceed a standalone LLM's compute limits—and few claim AI beats humans at everything .
The real question
As generative AI moves into deployment, the practical challenge isn't capability—it's designing clear criteria for where to keep a human in the loop and where not to delegate.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…