The term "Slightly Evil",  coined by internet guru Venkatesh Rao, refers to actions and choices that are advantageous from a game-theoretical perspective, are not inherentlyevil (that is, not unambiguously evil in the relevant contexts), but are outside the norm as judged from a naive social perspective.

Many of the AI apps that will be emerging in the coming days, weeks, and months and prove unexpectedly successful are likely to fall into this category, at least once all opportunities for straightforward adaptation of legacy services to a chat format will be exhausted.

The Moralising Big Media (MBM - for example the sort of mass publication dailies hailing from New York, Washington, and Manchester) will approach these with a characteristically low threshold and fail to distinguish between the "slightly evil" and the unambiguously, inherently, dangerously evil. And that leaves you and me, dear reader, to exercise our critical thinking muscles, if we are to make sense of this exciting new world.

Let's try a few (obvious) examples:

Communiclone

AI agent capable of communicating with others using text, audio, and video in a manner that is in both form and content indistinguishable to real flesh-and-bones you.

MBM Panic: Initially intended for simple tasks like reserving a table at the restaurant or answering unsolicited phone calls, it is now used by terrible people like a self-centred woman who lets it communicate with her insufferable, aging mother-in-law and an absent husband and father using it to be a pretend workaholic to his family while spending his evenings down at the pub.

Not inherently evil because: the technology is good enough so that the other side never notices; Billions of people around the world will be leading a happier and more productive life now that they don't have to talk to people they never really wanted to talk to.

Mechanical Wingman

Takes the soul-sucking drudgery out of online dating by automating all digital interactions end-to-end from the moment you sign up and until you are ready to go on a real life date, including picking and retouching your best photos and strategically swiping in the optimal direction.

MBM Panic: Algorithms take the romance out of dating. Innocent women and men exploited to go on dates with people they never even talked to.

Not inherently evil because: online dating really is soul-sucking drudgery. All anyone ever wanted is to go on actual real life dates.

Legalezer

Automated lawyer / hacker reads through millions of publicly available legal documents to uncover exploitable weaknesses and loopholes. Then exploits them. Again. And again. Poorly written satisfaction guarantee policies. Incomplete insurance schemes. Sloppy air-miles programmes. Rental contracts. All are nothing but opportunities to make a quick buck.

MBM Panic: Rule of Law subverted by robots.

Not inherently evil because: legal hacking is by definition legal; These contracts were written in the first place by slightly evil lawyers taking advantage of their legal skill and not expecting to be outcompeted by superintelligence. Time for them to go get a real job.

Machiavelli as a Service

Armed with centuries of research into politics, negotiation, and scheming, this bot is your personal slightly evil life and career coach. It reads your email, listens in on every meeting you have, knows you better than you know yourself, and, not being an actual human, has no problem confronting you with some uncomfortable truths you need to hear for your own benefit.

MBM Panic: Generation Z abandons binge drinking, unprotected sex, and dead-end careers in the arts in favour of office politics.

Not inherently evil because: political power and influence shouldn't be exclusive to naturals and sociopaths. Time to level the playing field.

The Mandatorian

Browser extension completes all your mandatory corporate trainings on your behalf while you play Tetris and sends you a one-pager summary for a quick review when it's done.

MBM Panic: Decades of DEI training flushed down the drain in AI reversal.

Not inherently evil because: 45m of unskippable, 1.0X speed videos, just to tell us that it's not OK to be racist? You gotta be kidding.

New to LessWrong?

New Comment
2 comments, sorted by Click to highlight new comments since: Today at 3:33 PM

The legalizer is the only thing here that isn’t inherently evil. The others may not be end-the-world kind of evil, but still evil.

To expound on the first two: they’re morally wrong because they’re lying. Your explanation of why they’re ok seems like some very short-sighted utilitarian thinking.

It's hard to define what is inherently evil. Your definition (IIUC) includes all lying. I don't think I can agree to that. Everyone lies all the time, and most of these acts or lying couldn't be classified as evil.