Going Beyond Asimov
It is important to look at what Asimov and his works are not, in order to fully understand what they are. Isaac Asimov was not a mathematician, an engineer, a or computer scientist. He wasn’t even a futurist, at least not in the sense that he did so to support himself. He was a chemist who eventually became a writer of fiction and scientific instruction for the popular market.
Asimov was not capable of providing rigorous, formalized definitions for the concepts he addressed; neither could he demonstrate how to take the natural-language formulations of the Three Laws and operationalize them. Asimov did not claim he could force artificial minds to obey his Laws; he didn’t even claim that it was necessarily possible to do so, at least in the fail-safe manner of his fictional worlds. For all we know, there are mathematical principles that are to the Three Laws what Arrow’s Theorem is to idealized voting systems. Maybe the Laws can never be enforced. Asimov never claimed they could be, one way or another.
So what’s the harm in Yudkowsky following in Asimov’s footsteps? There’s no particular reason to think the ideas were stolen from Asimov. It would be easy for someone thinking along the same lines to produce the same sorts of conclusions and hypotheses. We don’t attribute mobile phones to Star Trek, despite its fictional communicators helping to inspire engineers to make miniturized and portable phones. Star Trek had no technical details or even coherent speculation, just fictional technology in a fictional universe. Someone (in reality, a whole lot of someones) had to go beyond SF speculation and make real accomplishments in material science, electronics, and engineering.
So clearly Eliezer has gone beyond Asimov in the same way engineers went beyond Star Trek. He demonstrated how a definition of ‘Friendliness’ could be made explicit and rigorous in mathematical language, then he did so. His mathematical analysis demonstrated that it was possible to specify constraints for complex associational engines. He doesn’t yet know how to make them, or write computer code to accomplish his goals, but he showed that his goals were coherent and attainable.
He did do those things, right? There are accomplishments of Eliezer’s that go beyond talking about some ideas in everyday English. Equations, analysis in the fields of information science and abstract mathematics, rigorous arguments that have been examined and verified by others. They do exist. They must.
Because it would be absurd for Eliezer to claim to have made contributions to artificial intelligence, even purely abstract and theoretical ones, without being able to put forward anything more than what Asimov already did.
Because it would be absurd for us to take such claims seriously if we hadn’t already been shown such work.