Personal Statement on AI Risk
docs.google.com
2 threads
At the cost of drawing a potentially controversial comparison. At this rate of progression I expect we will have as much success "engineering" AI as we would engineering our own kids.
Outputs in humans are typically a weighted sum of all the learnings throughout a lifetime of inputs, applied to a specific problem. And equally difficult to predict without seeing the resulting behavior to a similar problem in the past.
Perhaps we just need to be looking at what we're actually teaching AI instead.
The time will come where AI will be learning by and for itself. I just hope we'll been able to teach it some positive core values before that time comes.