Human–AI Alignment

Human–AI alignment is the work of making AI systems act in ways people actually want: pursuing the goals we intend, respecting the limits we care about, and making trade-offs the way thoughtful humans would.

It sounds simple until you try to write the values down. People disagree, change their minds with context, and hold principles they would never trade, even when the numbers say they should. An aligned system has to cope with all of that, which is why alignment is as much a question about humans as it is about machines.

How real people decide

What value alignment means

Value alignment asks whether an AI system’s behaviour matches human values, not just its stated instructions. A system can follow a request to the letter and still do something its users would reject, because the request left out what everyone assumed went without saying.

The hard part is that those unstated assumptions are exactly what people find difficult to explain. We notice them mostly when they are broken.

Why human values are hard to specify

Context: the same act can be right in one setting and wrong in another. Telling a lie to protect someone is judged very differently from lying for gain.

Disagreement: people from different backgrounds weigh fairness, loyalty, autonomy and harm differently, and none of those weightings is simply a mistake.

Things people will not trade: most people treat some values as off the table, such as dignity or consent, even when trading them would produce a better total outcome.

Intent and responsibility: humans care about why something happened and who chose it, not only about the result.

How AI systems learn values today

Most current systems learn from human feedback: people compare outputs and say which is better, and the model is trained towards the preferred answers. Others are guided by written principles or rules that the system is asked to follow.

Both approaches depend on the quality of the human judgement behind them. If the feedback comes from a small, narrow group, or from rushed ratings, the system inherits those blind spots.

Where Resonant fits

Resonant asks real people to make short, forced-choice decisions about hard trade-offs: fairness against loyalty, mercy against accountability, privacy against protection. Answers are anonymous and accounts go through regular checks to show a person is behind them.

Each decision page shows how the answers split. Together they form an open, growing record of how people actually weigh competing values, which alignment work can use to test whether a system’s choices look like human ones.

Common questions

What is the difference between AI alignment and AI safety?

AI safety covers every way an AI system could cause harm. Alignment is the part concerned with making a system’s goals and choices match what people actually value.

Can AI be aligned with everyone at once?

Not perfectly, because people disagree. Good alignment work maps where people agree, where they differ, and which values most people refuse to trade.

Why use forced-choice dilemmas?

Asking people to choose between two real costs shows what they prioritise, which general opinion questions tend to hide.

All human decisions · The Trolley Problem · Human–AI Alignment · Articles · Resonant