[ 17/7/20 ]
At 43:40 Dan starts talking about applicability domain analysis.
That is all well and good if one is in a cooperative environment and one has the time to do that analysis.
If one is in a competitive environment, then an agent that is well trained and does not do a domain applicability analysis will always out perform one that does (all other things being roughly equal).
Thus this form of safety analysis is viable in cooperative domains, but fails in competitive ones.
Current economic and political dogma is that competition is good.
Competition can be “good” in the general case if (and only if) the fundamental systemic base is cooperative.
This is not the case with current societal systems.
Introducing these systems into today’s competitive social systems is instantiating existential level risk – no ifs buts or otherwise.
These systems can be safe only in cooperative systems. That means ensuring that every person on the planet has what they consider reasonable levels of security, freedom and empowerment.
That isn’t all that technically difficult to deliver, but it does break the current economic system of markets and money.
This is the defining issue of our age.
We either solve it, or we go extinct.
There really isn’t any half measure – we are in fact that technologically inventive.
Many more people need to see and appreciate this simple fact.