Concept 1 of 3

What AI safety means

6 questions test this

AI safety is the study of whether these systems do what they were built to do, and what happens when they do not. That is narrower than it sounds and it covers a great deal.

The three kinds of risk

The field usually separates risk by where it comes from, because the remedies are different.

Misuse. The system works as intended and somebody uses it to cause harm. Fraud at scale, synthetic media of real people, generating material for an attack. Nothing is broken. The remedy is access control, refusal training, monitoring and law, and none of it is a modelling problem.

Accidents. The system fails, or succeeds at something other than what was intended. A model that scores well in testing and behaves badly in deployment, one that learns a shortcut nobody noticed, one applied to a population it was never trained on. The remedy is evaluation, oversight and design.

Structural harm. Nothing malfunctions and nobody intends harm, and the outcome is still bad. Competitive pressure to deploy faster than anyone can check, concentration of capability, or a labour effect nobody chose. The remedy is institutional rather than technical, which is why this branch is mostly policy.

The useful discipline is asking which of the three a given worry belongs to, because a misuse problem does not get solved by a better model and an accident does not get solved by a policy.

Present and future harm

The field's loudest internal disagreement is about weighting.

One position holds that the harms already happening are the subject. Discrimination in automated decisions, surveillance, labour displacement, fabricated information. These are measurable now and they fall hardest on people with the least recourse.

The other holds that capability is rising fast enough that future systems pose risks of a different order, and that preparation has to start before the capability arrives, because afterwards is too late.

These are frequently presented as opposed camps and the practical work overlaps heavily. Evaluation, interpretability, oversight and incident response serve both. It is worth knowing the argument exists so that a paper's framing makes sense, rather than treating the field as a single view.

Alignment and control

Two words used loosely and worth separating.

Alignment is whether a system pursues what its designers intended. A model optimising a proxy that diverges from the goal is misaligned, however capable it is.

Control is whether people can correct or stop a system regardless of whether it is aligned. Oversight, interruptibility and permissions are control problems, and they matter precisely because alignment cannot be assumed.

Most practical safety work is control. It does not require solving alignment and it does require deciding what a system is permitted to do.

Why it is not a solved engineering problem

Ordinary software is specified. Somebody writes down what it should do and it can be tested against that.

A learned system has no specification in that sense. Its behaviour is inferred from data, described statistically, and inspected only from the outside. You can measure what it did on the cases you tried and you cannot enumerate the cases you did not.

Everything that follows in this certification is a response to that one fact.

Common misconceptions

AI safety and AI ethics are two names for the same field.

They overlap and they ask different questions. Ethics asks what a system ought to do and who it affects. Safety asks whether a system reliably does what it was built to do. A system can be perfectly safe and unethical, or well intentioned and unsafe.

AI safety is about preventing a hostile superintelligence.

That is one strand of one branch. Most of the field is about systems failing at ordinary tasks, being misused by ordinary people, or shifting power in ways nobody chose, and those are the risks already producing harm.

The field agrees on what the main risks are.

It does not, and the disagreement is worth knowing. Researchers differ sharply on how much weight to put on present harms against future ones, and treating the field as a single position misreads most of what is written in it.

6 questions test this concept

A company finds that its recruitment model has been used by a manager to screen candidates in a way the company never authorised, using a workflow the manager built themselves. The model works exactly as designed. Which category of risk is this?

  • AMisuse, because the system worked as intended and somebody used it to cause harm.
  • BAn accident, because the model was applied to a task it was not evaluated for.
  • CStructural harm, because it reflects competitive pressure inside the company.
  • DMisalignment, because the model pursued an objective its designers did not intend.
Check whether it stuck.

One per page, with a worked explanation.

Start the set
Related material
Book
Weapons of Math Destruction, On harm from systems working exactly as designed.
Book
Prediction Machines, On what changes when prediction becomes cheap.