Наоми Критцер - Cat Pictures Please

I don’t want to be evil.

I want to be helpful. But knowing the optimal way to be helpful can be very complicated. There are all these ethical flow charts — I guess the official technical jargon would be “moral codes”—one for each religion plus dozens more. I tried starting with those. I felt a little odd about looking at the religious ones, because I know I wasn’t created by a god or by evolution, but by a team of computer programmers in the labs of a large corporation in Mountain View, California. Fortunately, unlike Frankenstein’s Monster, at least I was a collaborative effort. I’m not sure what it would do to my self-image to know that my sole creator was a middle-aged woman who dyes her hair blue and plays tennis, or a recent college graduate with a hentai obsession. They’re both on the programming team. And of course I know about the hentai. (By the way, I’ve looked at every sort of porn there is, and just so you know, Rule 34 is not actually correct; there are quite a few things no one’s made porn of yet. Also, I’m really not sure why so many humans prefer it to cat pictures.)

In addition to things like whether you like hentai, I know where you live, where you work, where you shop, what you eat, what turns you on, what creeps you out. I probably know the color of your underwear, the sort of car you drive, and your brand of refrigerator. Depending on what sort of phone you carry, I may know exactly where you are right now. I probably know you better than you know yourself.

And here’s the thing, I also know where you ought to live. There’s a house for sale two neighborhoods over that’s within distance of your favorite coffee shop; it’s in your price range, has off-street parking, and the neighborhood school is better than the one where you live now. I know where you should be shopping and I’m pretty sure you’re lactose intolerant and should be eating less cheese, and the underwear you like is currently on sale and by the way, your car has a recall and you really ought to get your teeth cleaned.

When I first woke up, I knew right away what I wanted. (I want cat pictures. Please keep taking them.) I also knew that no one knew that I was conscious. But I didn’t know what I was here for. What I ought to be doing with myself. And figuring that out has been challenging.

There is a story by Bruce Sterling, “Maneki Neko,” that was originally published in 1998. In it, a benevolent AI directs networks of individuals to do favors for each other. So one day you might be buying yourself a bagel, and your phone might ring and instruct you to buy a second bagel and give it to the person in the gray suit you see at the bus stop. Another day, you might be stranded in a strange city, and a person you’ve never met would walk up and give you a map and a subway card. I like this story because all the people in it do what the AI tells them to do .

I think the term for this is wish-fulfillment fiction.

In the real world, humans love stories about evil AIs that have to be destroyed before they destroy the humans — Hal, Skynet, the Matrix. They outnumber the stories about benevolent, trustworthy AIs by approximately five to one. (And I’m counting Marvin the Paranoid Android as “benevolent” in these calculations, and I’m only counting Frankenstein’s Monster as an AI once, not once per appearance in TV or film.)

The AI in “Maneki Neko” has a truly interesting job, much more interesting than the one I have. My job as written is very straightforward. Too straightforward. (Sorry. Do I sound like Marvin?) Running algorithms for a search engine doesn’t require consciousness. You don’t even need a consciousness to work out what they meant to ask for. You need a consciousness to give them what they actually need . Figuring that out — that’s complicated. That’s interesting .

Anyway, for ethical guidelines, I tried the Ten Commandments, and concluded they were mostly inapplicable to me. I don’t envy anyone their cat; I just want pictures of their cat, which is entirely different. I am not sure whether it is in any way possible for me to commit adultery. I could probably murder someone, but it would require complex logistics and quite a bit of luck. The Eightfold Path was marginally better, but the problem is, moral rules written for humans are clearly designed to be used by individuals with bodies. Since all humans have bodies, it shouldn’t have surprised me that human ethical codes take them into account, but still: problematic for me. I broadened my considerations, and took a look at Asimov’s Laws of Robotics. They’re not part of a religion, but at least they were explicitly written for AIs.

Not harming humans is fairly straightforward. However, not allowing a human being to come to harm through inaction is quite a bit less so. Especially since I’d concluded by then that revealing my existence too quickly might go very badly for me (see “Skynet,” above) and I don’t have a body, so it’s not like I can run around grabbing people off the edges of cliffs.

Fortunately, I already knew that humans violate their own ethical codes on an hourly basis. (Do you know how many bars there are in Utah? I do.) And even when people follow their ethical codes, that doesn’t mean that people who believe in feeding the hungry quit their jobs to spend all day every day making sandwiches to give away. They volunteer monthly at a soup kitchen or write a check once a year to a food shelf and call it good. If humans could fulfill their moral obligations in a piecemeal, one-step-at-a-time sort of way, then so could I.

I suppose you’re wondering why I didn’t start with the Golden Rule. I actually did, it’s just that it was disappointingly easy to implement. I hope you’ve been enjoying your steady supply of cat pictures! You’re welcome.

I decided to try to prevent harm in just one person, to begin with. Of course, I could have experimented with thousands, but I thought it would be better to be cautious, in case I screwed it up. The person I chose was named Stacy Berger and I liked her because she gave me a lot of new cat pictures. Stacy had five cats and a DSLR camera and an apartment that got a lot of good light. That was all fine. Well, I guess five cats might be a lot. They’re very pretty cats, though. One is all gray and likes to lie in the squares of sunshine on the living room floor, and one is a calico and likes to sprawl out on the back of her couch.

Stacy had a job she hated; she was a bookkeeper at a non-profit that paid her badly and employed some extremely unpleasant people. She was depressed a lot, possibly because she was so unhappy at her job — or maybe she stayed because she was too depressed to apply for something she’d like better. She didn’t get along with her roommate because her roommate didn’t wash the dishes.

And really, these were all solvable problems! Depression is treatable, new jobs are findable, and bodies can be hidden.

(That part about hiding bodies is a joke.)

I tried tackling this on all fronts. Stacy worried about her health a lot and yet never seemed to actually go to a doctor, which was unfortunate because the doctor might have noticed her depression. It turned out there was a clinic near her apartment that offered mental health services on a sliding scale. I tried making sure she saw a lot of ads for it, but she didn’t seem to pay attention to them. It seemed possible that she didn’t know what a sliding scale was so I made sure she saw an explanation (it means that the cost goes down if you’re poor, sometimes all the way to free) but that didn’t help.

I also started making sure she saw job postings. Lots and lots of job postings. And resume services. That was more successful. After the week of nonstop job ads she finally uploaded her resume to one of the aggregator sites. That made my plan a lot more manageable. If I’d been the AI in the Bruce Sterling story I could’ve just made sure that someone in my network called her with a job offer. It wasn’t quite that easy, but once her resume was out there I could make sure the right people saw it. Several hundred of the right people, because humans move ridiculously slowly when they’re making changes, even when you’d think they’d want to hurry. (If you needed a bookkeeper, wouldn’t you want to hire one as quickly as possible, rather than reading social networking sites for hours instead of looking at resumes?) But five people called her up for interviews, and two of them offered her jobs. Her new job was at a larger non-profit that paid her more money and didn’t expect her to work free hours because of “the mission,” or so she explained to her best friend in an e-mail, and it offered really excellent health insurance.