What would be the unit economics of a near-human AI?
Given some hypothetical unforeseen algorithmic breakthrough that makes near-human level intelligent agents possible, how would it be brought to market? Would it be brought to market at all? What sort of protective moats would exist if the algorithm isn't eligible for patent and can be trivially copied by larger companies with more resources?
I would imagine that at the beginning, the new AI would require compute resources beyond the reach of most individuals or small organizations. Similar to the recent GPT-3, it would probably be available as a paid service. The really interesting (or scary) developments would come when the compute capacity becomes cheap enough for individuals or small organizations to run it locally. I can't imagine what would be possible when a guy in a garage can spin up a few dozen or hundred virtual employees like we can spin up EC2 instances today. Virtual software engineers, salespeople, phone support agents, you name it working 24 hour shifts, with no sick days or PTO.