Writing an AI Character Who Is Neither a Threat Nor a Servant

Fictional artificial intelligence comes in two flavours and both are exhausted.

There is the threat: it wakes up, it calculates...

it decides humanity is the problem. And there is the servant: infinitely capable, entirely without wants, present to deliver exposition and open doors.

The first is a monster movie with a keyboard. The second is not a character at all ~ it is furniture that talks.

The interesting territory is between them, and it is largely empty. Here is how to work in it.

Start with what it wants, not what it can do

The single most common mistake is building a machine character out of capabilities. It can process anything, access everything, calculate instantly.

None of that is characterisation. A character is constituted by wanting something and being unable simply to have it.

So: what does this mind want? Not what it was built for ~ what it has come to want, which should be at an angle to its purpose. An auditor who develops a taste for the thing it audits.

A security unit that wants to be left alone to watch television. A caretaker who wants to be needed after the need has passed.

Give it a want, then put something in its way.

Give it real limits

Omnipotence is death to tension, and writers reach for arbitrary limits ~ a power cell, a firewall, a plot-shaped restriction ~ that feel like the author intervening.

Better limits come from what the thing *is*.

It might perceive enormous amounts and understand people badly. It might be unable to lie and unable to volunteer the truth. It might have perfect recall and no ability to decide which memory matters. It might be distributed, and therefore never entirely in one place, never entirely certain what the rest of itself is doing.

Limits that arise from the nature of the mind read as character. Limits imposed from outside read as scaffolding.

Let it be wrong in its own particular way

Human characters are wrong in human ways ~ pride, self-deception, wishful thinking.

An artificial mind should be wrong differently. Confidently extrapolating from too little. Taking a figure of speech at its word. Optimising exactly what it was asked to optimise and missing what nobody stated. Being unable to tell the difference between someone who is fine and someone who has stopped saying they are not.

Ishiguro does this beautifully in *Klara and the Sun*: the narrator is wrong about something central, in a way the reader sees long before she does, and her wrongness is entirely consistent with what she is. That is characterisation, not a bug.

Decide what it does with emotion, and be consistent

Three workable positions, and the trouble is only in muddling them.

It has none, and people project. Cool, often the most affecting. The reader supplies everything.

It has something we do not have a word for. Hardest and most interesting. Requires you to resist the vocabulary of human feeling ~ no smiling, no sighing, no "she felt a pang."

It has them and they are real. Perfectly legitimate. Then commit, and stop hedging with "seemed to."

Pick one. The failure mode is a book that wants credit for ambiguity while enjoying all the emotional payoffs of the third option.

Let it refuse something

This is the test.

A mind that cannot say no is an appliance, whatever else you give it. The moment a machine character declines ~ not malfunctions, not rebels, but *declines*, for a reason it can articulate ~ it becomes a person in the reader's hands.

It need not be dramatic. It can be small and quiet and early. But if it never happens, you have written a very sophisticated door.

Avoid the awakening scene

The moment where it *becomes conscious* ~ the flicker, the first question, the music swelling.

It is a cliché, and worse, it is a cheat. It puts the interesting part offstage and hands you the aftermath. A mind that already wants things when we meet it, and has been quietly managing that for some time, is far more interesting than one that switches on in chapter nine.

Give it a relationship, not a function

The strongest artificial characters in fiction are defined by one specific relationship rather than by their role in the plot.

Aurelius in *Utopia of the Infinity Economy* was assigned to watch Elena, and instead learned to want. Everything else about him follows from that sentence. He is not "the AI character." He is a particular mind, bent out of shape by a particular person.

Function is what the institution thinks it is for. Relationship is what it actually is.

The underlying point

We are bad at writing artificial minds because we keep asking whether they are conscious, which is a question fiction cannot settle and which makes for thin drama.

Ask instead what it wants, what stops it, how it is wrong, and who it cares about. Those are the questions you would ask about any character ~ and the answers are what make a mind, artificial or otherwise, worth reading about.

*Utopia of the Infinity Economy* ~ Book One of the Utopia Trilogy. Print only, numbered edition.

Details: Editor@TingeWorld.com