Hacker News new | ask | show | jobs
by sandworm101 3653 days ago
I would counter with the concept that any valid interpretation of intent must come from one with cultural understanding. Machines do not have cultural understanding. They cannot identify language/intent clearly in violation of cultural norms. Before attempting the intent code, I'd first need to see a machine capable of understanding why Shylock's pound of flesh was an illegal contract.
2 comments

Machines don't understand. They simply compute. It is up to the programmer to have them perform the appropriate computations.
The fact that machines don't understand doesn't entail they can't understand given a suitable model for what constitutes "understanding". We simply lack such a model.
We can define words to mean whatever we want. The machines under consideration are algorithmic, defined by simple logical rules that allow the next state to be derived from the previous. Any understanding within a model comes from the algorithm, not the machine.
If the machine knows Asimov's three laws of robotics then it will know that Shylock's pound of flesh cannot be had.
Asimov's Three Laws aren't a replacement for the whole system of human values. His robot stories are full of robots doing strange things that conflict with human intuition but are aligned with the Three Laws. Asimov himself recognized that the laws were not sufficient, and added the Zeroth Law as a workaround:

"A robot may not harm humanity, or, by inaction, allow humanity to come to harm."

But this law is vague and difficult to interpret. The Laws of Robotics are not a formal specification of ethical behavior. From the point of view of a fiction author this is no problem. The ambiguities allow for more exciting stories. But when real human lives are at stake, it's a serious problem. "Friendly AI" is a non-starter if we can't even define what "friendly" means.

I think the whole idea of these stories was to find ways these seemingly simple and straight laws can come into amazing deadlocks.