If one assumes that the biggest bottleneck to AGI is large insight from geniuses,
Don’t be too sure about that. You might want to take a look at Little Bets: How Breakthrough Ideas Emerge from Small Discoveries. My own experience when dealing with difficult problems has been that an incremental approach can be much more productive than sitting at my desk thinking real hard about the parts I don’t know how to do. I just go ahead and do the parts that I can do, and let the harder problems percolate in my mind. By the time I’ve got the manageable parts done, I’ve learned enough about the problem, turned vague abstractions into concrete realizations, and reduced the uncertainty enough that the hard parts often fall right into place. Consider a probability distribution over variables x1, …, xn, with the variables representing solutions to parts of a problem. P(xn) may be quite diffuse and spread out, but P(xn | x1, x2, …, xk) may be much more concentrated.
I don’t expect the problem of FAI to fall easily, but any incremental advance can help set the stage for the definitive advances.
Good point! I also appreciate how a lot of your ideas are referenced to books. Assuming your right (I think I would have to know much more about AGI in order to evaluate the claim with any certainty), the next obvious question is, what’s limiting progress? Are there not enough people making small discoveries (good reason to go into the field)? Are they not sharing their discoveries (don’t know how you’d fix this one)? Is there not enough funding for them to do their work (good reason to stay out unless you plan on outcompeting everybody because you feel that’s what you can do best to help)?
Don’t be too sure about that. You might want to take a look at Little Bets: How Breakthrough Ideas Emerge from Small Discoveries. My own experience when dealing with difficult problems has been that an incremental approach can be much more productive than sitting at my desk thinking real hard about the parts I don’t know how to do. I just go ahead and do the parts that I can do, and let the harder problems percolate in my mind. By the time I’ve got the manageable parts done, I’ve learned enough about the problem, turned vague abstractions into concrete realizations, and reduced the uncertainty enough that the hard parts often fall right into place. Consider a probability distribution over variables x1, …, xn, with the variables representing solutions to parts of a problem. P(xn) may be quite diffuse and spread out, but P(xn | x1, x2, …, xk) may be much more concentrated.
I don’t expect the problem of FAI to fall easily, but any incremental advance can help set the stage for the definitive advances.
Good point! I also appreciate how a lot of your ideas are referenced to books. Assuming your right (I think I would have to know much more about AGI in order to evaluate the claim with any certainty), the next obvious question is, what’s limiting progress? Are there not enough people making small discoveries (good reason to go into the field)? Are they not sharing their discoveries (don’t know how you’d fix this one)? Is there not enough funding for them to do their work (good reason to stay out unless you plan on outcompeting everybody because you feel that’s what you can do best to help)?