Worth Reading: “An Understanding of AI's Limitations Is Starting to Sink In”

We share pieces from outside Currux Vision when they're worth an engineer's ten minutes. This one, by Tim Cross for The Economist, is a clear-eyed look at where machine learning genuinely delivers — and where the industry's own hype has outrun it.

PwC has forecast that artificial intelligence will add $16 trillion to the global economy by 2030, comparable to China's entire economic output today. Google's Sundar Pichai has called AI developments “more profound than fire or electricity.” The piece argues that, against those claims, skepticism is mounting.

Machine learning has genuinely improved at pattern recognition — it powers search engines, voice assistants and facial recognition — but many of the more grandiose promises remain unfulfilled. Self-driving cars, the article notes, remain perpetually “on the cusp of being safe enough,” despite decades of development.

The core limitation, as Cross frames it, is that today's AI systems are “powerful pattern-recognition tools” lacking genuine reasoning, common sense, and the ability to generalize. They excel at bounded tasks and fail on unexpected inputs — which is exactly the property any AI system deployed on public roads has to be honest about.