First Design for Double Checking Mind

Use neural networks to solve an expression and output the result, trusting its correct.

Separately, double check the output, evaluating whether the final equation balances. Then output a "woops" message if it was wrong.

The next step is for that to trigger a reanalysis, but with a different state of mind, so that it acts more carefully. (That mental model of mind state will be very important when learning new skills to track how reliable the processors are now, and this how much double checking is required. But its too complex for now).

More:

• That we produced a result at all, given a request.
• That we aren't in an infinite thought cycle.

What other double checks can we add now?

(Added 2019-11-16. Labels: work-in-progress)