5 Comments
User's avatar
Fran Macadam's avatar

There’s that old saw, that to err is human, but to screw up at massive scale requires computing power.

Fran Macadam's avatar

It all points to irrational exuberance.

Fran Macadam's avatar

Compounding the problem, making analysis difficult, is making too many changes at once, rather than incremental testing. The challenge becomes much more demanding in time and effort and requires incremental reversal. Upgrades need to be created in parallel and tested before terminating previous working deployments. And fully tested backup and recovery procedures are essential. Who is now making these policy decisions?

Fran Macadam's avatar

I designed computer hardware and operating system code, manufactured systems, maintained enterprise code, designed and deployed networks, operated ISPs as CTO, ran corporate data centers to the Fortune 500 level as IT Director. From my experience, almost all failures happen during upgrades due to faulty planning, lack of testing and inability to revert when failures happen due to the implementation. You almost always need to ask, what changed recently. Before these recent events, I already noticed that as AI had replaced human positions, the problem solving expertise of the fewer human beings replacing those eliminated had been drastically reduced. As well, AI is remarkably devoid of judgement and makes frequent logical errors with many blind spots. You would have thought that all the old errors would have been fully planned for before rolling out at massive scale, but no.