Q: Isn’t the definition of “exceptional” a moving target? How do you draw the line?

A: Yes. There are always many moving parts in both tech, org, & society. The boundary should move at a pace that controls risk while increasing value. With many moving parts, the best way to control risk is to test with restricted demos while using statistical & many eyeball analysis to evaluate the coverage of the demos that have been run. Then the rollout is another moving part, so the pace of that also needs control.

For Robotics & VR, there is a lot of value in pushing the increasing detail, comprehensiveness, and agility of simulations, and digestable claims about their properties. For what they cover accurately, this gives a huge boost to the speed of design & testing. But it doesn’t provide a real way to test how current or complete the models are and whether all relevant details in familiar situations have actually been captured. Also, when many people work on the same VR, it shapes their perception so they are subtly influenced to accept the same inaccuracies. It doesn’t have the same independence of error chance as if N people had each made their own reality model.

For the human-human interaction for example, one could design a workflow where 2 parties send their respective AI agents to work out some agreement between time t1 and t2 and then they meet live to review what was agreed & not agreed to mediate. How important is the time & style of their human interaction there? It still matters. There may be exceptional cases. Is there pressure to achieve efficiency from making the review short? Maybe. But maybe the office has more cocktail hours instead & they incorporate more structure to get conversation coverage. Point of this ramble is just to stimulate thinking about the tradeoffs & ways of partially automating to gain efficiency.