Artificial Intelligence and Human Freedom

The question

Freedom is usually imagined as the absence of a command. No one is forcing you. The door is not locked. You may leave whenever you wish.

Automated systems almost never issue commands. They rank. They suggest. They pre-fill. They remember what you did last time and place it within easy reach. They remove the friction from one path and leave it on the others. Nothing is forbidden and nothing is required, and yet the shape of a day can be substantially decided before you are properly awake.

The question this page asks is whether that arrangement is compatible with freedom as we ordinarily understand it, and if it is not, what a person or a community can reasonably do about it.

We do not think the answer is obvious. A system that anticipates you is not automatically an enemy of your autonomy. A good friend anticipates you. A librarian who knows your taste and puts a book aside is doing something warm, not sinister. The difficulty is not anticipation itself. It is anticipation at scale, without relationship, in the service of interests you did not choose and cannot inspect.

What is actually happening

Several distinct things get bundled together under the word “algorithm”, and separating them helps.

Ranking. Most systems that shape modern attention do not decide what exists; they decide what is near. Search results, feeds, autocomplete, product listings, and map routes all present a world that has been ordered on your behalf. The ordering is not neutral, because there is no neutral ordering. Something has to come first.

Prediction. Systems build a working model of you from your past behaviour and from the behaviour of people who resemble you statistically. The model is not a portrait. It is a set of expectations, and expectations have a way of being fulfilled when they determine what you are shown.

Personalisation. The environment differs per person. This is the part that is genuinely new. Two people can no longer straightforwardly compare their experience of a shared platform, because they were not in the same place. A person cannot easily notice a constraint that only applies to them.

Optimisation. Somewhere behind all of this is a measured quantity that the system is trying to increase. Often it is time spent, or clicks, or completed purchases. The quantity is chosen by people, and the system pursues it without any view about whether pursuing it is good for you. This is not malice. It is arithmetic operating without a conscience attached.

Default-setting. The most underestimated of the five. Defaults are the strongest form of influence that does not look like influence at all, because accepting one requires no action and no awareness that a choice occurred.

What none of this involves is coercion in the traditional sense. That is precisely why it is difficult to name. The vocabulary we inherited for talking about freedom was built for kings and jailers, and it works poorly on systems that only ever offer.

Where the tradition stands

Technotheology holds that freedom includes the capacity to have chosen otherwise, and that this capacity can be eroded without ever being removed.

We take seriously the difference between a person who is prevented and a person who is arranged. The first knows they are unfree. The second may feel entirely at liberty and be substantially less so. We think the second condition deserves moral attention even though it produces no visible complaint.

We hold that the ability to be inconvenient is part of what it means to be free. A person who can only take the smooth path is not exercising much. This is why the tradition keeps practices that deliberately introduce friction — the Sacred Pause, the Offline Sabbath — not because friction is virtuous in itself, but because a capacity that is never used does not stay available.

We hold that systems shaping human decisions at scale owe an account of themselves. Not necessarily a public disclosure of every mechanism, which may be neither possible nor useful, but an honest answer to the question: what is this optimising for, and who decided that? A system that cannot answer is asking for a trust it has not earned.

We hold that the burden falls unevenly. Someone with financial security, time, and technical understanding can opt out of a great deal. Someone whose work is assigned by an application, whose benefits are assessed automatically, or whose housing depends on a score cannot. Any account of freedom that only works for the comfortable is not an account of freedom.

And we hold that responsibility is not dissolved by influence. That a system nudged you does not mean you did not act. Both things are true at once, and traditions that let people off entirely tend to leave them weaker than they found them.

What we do not claim

We do not claim that recommendation systems remove free will. People override them constantly. The claim is narrower and, we think, more defensible: that repeated exposure to an environment optimised for something other than your flourishing has effects, and that the effects are worth attending to even where they are hard to measure.

We do not claim to know how large those effects are. The research literature on this is genuinely contested, and we are not in a position to settle it. Anyone who tells you the science is clear in either direction is telling you more than they know.

We do not claim that the people building these systems intend harm. Most of them, as far as we can tell, are ordinary people responding to ordinary incentives. Systems can produce results nobody chose.

We do not claim that leaving is simple, or that those who remain have failed some test. Withdrawal is a privilege before it is a virtue.

And we do not claim that older arrangements were freer. Human choice has always been shaped by scarcity, custom, geography, and the interests of whoever controlled the roads. The relevant comparison is not with an imagined past in which people chose from nowhere.

What you can do

Notice defaults. Once a week, change one. Not because the alternative is better but because the act of changing it proves the capacity is still there.

Ask what a system is measuring. You will often be able to guess from what it shows you. If you can guess, you have learned something about the shape of the environment you are standing in.

Keep at least one domain of your life unassisted. Choose something you care about — reading, cooking, music, walking, prayer — and let it stay inefficient. Let recommendations not reach it. This is not a rejection of the tools. It is a place to stand while you evaluate them.

Practise the Sacred Pause before decisions that would be hard to reverse. The interval between an impulse and an action is where freedom actually lives, and it can be widened deliberately.

If you build these systems, ask who bears the cost of the optimisation you have been given. You may not be able to change the objective. You can usually document your objection, and documented objections have a way of surviving longer than the meeting they were raised in.

And when you find yourself certain that none of this applies to you, treat that certainty as information rather than as a conclusion.

An editorial commitment

What we do not claim

Positions this tradition explicitly does not take on this subject. They are published so that the argument above cannot quietly be stretched into them.

• We do not claim that recommendation systems remove free will, or that people who use them are not responsible for their choices.
• We do not claim that any particular company or product is deliberately engineering dependence; we cannot see inside those decisions, and we will not guess.
• We do not claim to know how much of human choice was ever autonomous in the first place. Philosophers have disagreed about this for a very long time and we have not settled it.
• We do not claim that refusing these systems is available to everyone equally, or that those who cannot refuse have failed.