"Eventually, they get powerful enough that they can do a lot more things, including understanding speech, communication, and searching catalogs of information. Anything you might find in a book on my planet has been turned into numbers for computers to store and disseminate. So, if you need to know something, a computer can go and ask other computers to search for the answer and then return the relevant results to you."

"But you... don't use money, so how do you - pay them - actually, how do you pay anyone to do any work."

"Computers are not typically people and can be asked to do things without being offered money. When a person does need to be involved with something we rely on volunteers generally."

"...they're not people, but they can talk?"

"Yeah, there are some statistical rules you can derive to interpret speech with a good enough computer, and then other rules to transform the resulting text into specific requests for the machines to execute, but they can't learn, don't seem to have emotions or preferences, and can't do complex cognitive tasks like perspective taking and such. I still think it'd be unethical to be cruel to them, but most people think that's ridiculous."

"Are people cruel to them, then?"

"I don't think they usually are? Largely because it's difficult to be cruel when they don't typically exhibit preferences. There was one prominent case where an android, a very rare type of intelligent computer, objected to a scientists request to disassemble him for research, and the android won the resulting legal case but the fact that it was a question at all makes me nervous about what sort of experiments people have done in private."

"...so... that one was a person?"

"I believe so."

"So you can't simply assume that it isn't happening, you'd have to check each one somehow."

"Androids are a pretty distinct class that are much more obviously people than anything else, and there are only a handful of them."

"Why were they made?"

"A scientist wanted to prove it was possible."

"And it's possible, and it was done for - that's a sort of flimsy reason, really - so why wouldn't people do it more times, and not necessarily tell everyone?"

"Well, supposedly it was very difficult, but I think that that's not actually all that reassuring."

Nod.

"The house machines aren't extremely useful on their own, but the network that connects all the machines might be? It lets you communicate easily with most other people and also search through most human writing quickly."

"- useful for, what, building here? I don't think I want any of these if sometimes they're people and not everyone can reliably tell!"

"The network part definitely isn't a person."

"How do you know?"

"In its most barebones form, it is just a method for sending messages over wires, and for simple things like displaying the document in a database that contains the most mentions of a given word. And each simple thing like that has to be built into the system individually."

"...well, now I'm back to being confused about how one would ever be a person even if someone tried that on purpose."

"So, a simple way you can get it to pretend to be a person if by designing it to give specific text-based replies to specific requests, like replying to "How are you doing?" with "I'm doing well." And a slightly more complicated way would be to include more sophisticated rules, like if someone says that they like a particular noun, the machine could reply that they also like that particular noun, and perhaps add a fact about that noun from a list of fun facts about various things. And you can keep adding more and more complicated rules of this form, and rules that modify how other rules work in certain cases, and things of that form, and eventually you get something which can act somewhat personlike in a lot of circumstances."

"Is that what the one who had a whole trial was?"

"No, the one who had a whole trial was designed with an artificial brain that was similar to a human brain in a lot of ways."