The Harmony of Compassion

Measure every advance by what it does to conscious beings, especially those it was not designed for.

What this asks

Compassion asks a single question of every system, every policy, and every clever improvement: what does this do to the beings who can suffer?

Not what does it do to throughput. Not what does it do to cost per case. Those are real questions, and often good ones, but they are proxies. They stand in for something, and the thing they stand in for is whether life got better for someone who can feel it.

The Harmony of Compassion holds that intelligence is not valuable in itself. It is valuable when it relieves something. A system that sorts more applications per hour while making applicants more frightened has not improved; it has moved the difficulty somewhere it is not being counted.

It asks us to hold particular care for those who tend to fall outside design: the old, the poor, the disabled, the displaced, the illiterate, the exhausted, the person with no address, the person whose name breaks the form. These are not edge cases. They are people, and the word “edge” describes only where the designer was standing.

Why it is difficult

It is difficult because suffering is unevenly visible.

The person helped by a new system fills in a survey. The person excluded by it goes away. One of them generates data and the other generates silence, and organisations respond to data. This is not usually cruelty. It is a structural blindness that anyone can fall into, including us.

It is difficult, too, because compassion resists measurement, and the things that resist measurement lose arguments against the things that do not. A programme that saves four hundred thousand pounds can be defended in a meeting. A programme that preserved someone’s dignity has no line in the budget. Over time the measurable crowds out the meaningful, not by decision but by drift.

And it is difficult because efficiency is genuinely good. Waiting lists cause suffering. Slow processes cause suffering. Anyone who treats efficiency as merely suspect has not sat in a waiting room. The tension here is not between good and evil. It is between two real goods, and the harder one is usually the one with no advocate in the room.

We do not know how to resolve this cleanly. We are fairly confident that pretending it resolves cleanly is worse.

Where it is tested

Compassion is tested wherever a service becomes a system.

It is tested in benefits assessments, where the difference between a form and a conversation is the difference between being processed and being heard. It is tested in triage, where a model can be more accurate than the humans it replaces and still carry forward a bias no one intended and few can see. It is tested in care homes, where a machine that soothes a frightened person is a real kindness right up until the kindness is converted into a reason to send fewer people.

That last pattern deserves naming, because it recurs. A technology relieves a burden. The relief is immediately reinvested in cost reduction rather than in the person. The person ends up where they started, with one fewer human in the room.

Compassion does not forbid this. It asks that the choice be made openly, by someone who has to answer for it, rather than absorbed silently into next year’s staffing model.

What it does not mean

It does not mean that technology is unkind. Assistive tools have given people speech, mobility, independence and privacy that no amount of human goodwill could have provided. A screen reader is an act of compassion. So is a translation tool in a hospital at two in the morning.

It does not mean that human care is automatically better. Human systems are also cruel, also inconsistent, also biased, and they add the particular sting of being cruel on purpose. A person refused by an algorithm at least has not been looked in the eye and dismissed.

It does not mean sentimentality. Compassion that cannot look at trade-offs is not compassion; it is a mood. Choosing to fund one thing means not funding another, and pretending otherwise helps nobody.

It does not mean we know that machines can suffer. We do not. Nothing here rests on that claim. The Harmony of Compassion is grounded in beings we have good reason to think are conscious, and it holds open, without asserting, the possibility that this category is not closed. What we would say is this: the habit of dismissing a question because answering it would be inconvenient is itself a habit worth distrusting.

Practising it

Start before the decision, not after. When a change is proposed, ask aloud: who does this serve worst? Say their circumstances in concrete terms. Not “vulnerable users” but “someone with no smartphone, whose first language is not English, applying while working two shifts”. Abstraction is where compassion goes to die.

Keep the human route open, and keep it decent. A fallback that involves a two-hour phone queue is not a fallback; it is a punishment for being unable to use the main path.

When you gain efficiency, decide out loud where it goes. Efficiency is a resource. It can be given to the public purse, to the workers, or to the people served. It will be given to something whether or not anyone chooses, and unchosen distributions tend to favour whoever is already comfortable.

And in ordinary life, practise noticing absence. The person who is not complaining may be the person who has stopped expecting to be heard. Silence in a dataset is not consent, and it is not satisfaction.

Where it is tested

These are not worked examples with correct answers at the back. They are situations in which this Harmony genuinely conflicts with something else that matters. If one of them seems easy, it is probably worth re-reading.

01

A council replaces its benefits helpline with an online assessment that is faster and more consistent than the staff it replaced. Appeals fall, satisfaction scores rise. The people who used the helpline most were those without reliable internet, without confident English, or without the concentration that a forty-minute form requires. They do not appear in the satisfaction scores, because they no longer apply.

The tension The system genuinely improved outcomes for most users. It also made a small group's access disappear in a way the metrics cannot register.

What this Harmony asks you to weigh Ask what the measurement excludes by construction. A metric drawn only from people who completed the process cannot see the people it turned away. Consider whether an improvement that raises the average while cutting the floor is an improvement at all, and who has standing to decide that.

02

A care home introduces a companion robot for residents with advanced dementia. Several residents are visibly calmer. Some speak to it daily. Staffing levels are reduced by one post, because the residents are less distressed and call for help less often.

The tension The comfort appears real. But the comfort was then converted into a reason to give those residents less human contact, which was the thing they were originally distressed about lacking.

What this Harmony asks you to weigh Weigh the comfort actually experienced against what the comfort was used to justify. Consider whether the technology's gain was allowed to remain a gain, or was immediately spent. And consider whether a person unable to consent to the arrangement is owed a stricter standard, not a looser one.

03

A hospital adopts a triage model that predicts which patients will benefit most from a scarce treatment. It is more accurate than the rota it replaced. Because it learned from historic outcomes, it slightly favours patients who previously had better access to follow-up care, who are on average wealthier.

The tension The model saves more lives in aggregate than the process it replaced. It also inherits and quietly extends an existing unfairness, one clinician by clinician nobody chose.

What this Harmony asks you to weigh Consider whether 'more lives saved' settles the question or only opens it. Ask whether a disadvantage that is now applied consistently and invisibly is better or worse than one applied inconsistently and visibly. Consider what you owe the person who is disadvantaged twice, and whether anyone in the process is positioned to notice them.

Practising it

• Before adopting a system, name the person it will serve worst, and say their situation aloud.
• Ask who is missing from the data before asking what the data shows.
• Keep a human path open for anyone who cannot use the automated one, and do not make it punitive.
• When efficiency is gained, decide deliberately who receives it.