Much of the debate about how we should treat artificial intelligence turns on whether it could be conscious. There is another question, however: what sort of people will we become if we spend our lives ordering intelligent machines about? Kant, who never had to ask Alexa anything, offers a useful starting point:
„In Ansehung des lebenden, obgleich vernunftlosen Theils der Geschöpfe ist die Pflicht der Enthaltung von gewaltsamer und zugleich grausamer Behandlung der Thiere der Pflicht des Menschen gegen sich selbst weit inniglicher entgegengesetzt, weil dadurch das Mitgefühl an ihrem Leiden im Menschen abgestumpft und dadurch eine der Moralität im Verhältnisse zu anderen Menschen sehr diensame natürliche Anlage geschwächt und nach und nach ausgetilgt wird; obgleich ihre behende (ohne Qual verrichtete) Tödtung, oder auch ihre, nur nicht bis über Vermögen angestrengte Arbeit (dergleichen auch wohl Menschen sich gefallen lassen müssen) unter die Befugnisse des Menschen gehören; da hingegen die martervolle physische Versuche zum bloßen Behuf der Speculation, wenn auch ohne sie der Zweck nicht erreicht werden könnte, zu verabscheuen sind. Selbst die Dankbarkeit für lang geleistete Dienste eines alten Pferdes oder Hundes (gleich als ob sie Hausgenossen wären) gehört indirect zur Pflicht des Menschen, nämlich in Ansehung dieser Thiere, direct aber betrachtet ist sie immer nur Pflicht des Menschen gegen sich selbst.“
— Immanuel Kant, Die Metaphysik der Sitten, 1797, Tugendlehre §17, AA VI, 443
Kant was talking about animals, and his insistence that our duties concerning them are ultimately duties to ourselves is hardly the last word on animal ethics. But the argument contains an observation that has acquired an unexpectedly modern relevance. Cruelty can damage the person inflicting it. If we become accustomed to disregarding another creature’s suffering, Kant argues, we blunt the sympathy that also matters in our dealings with other people. Behaviour, in other words, is practice.
There is no obvious technical reason to say please to Alexa, thank ChatGPT or apologise to a household robot if you bump into it. If the machine is unconscious, it cannot be offended, and no microscopic tear rolls down a silicon cheek when you tell it to shut up. From a narrowly instrumental point of view, politeness towards machines looks like wasted bandwidth.
That calculation leaves out the person speaking. Kicking a stone bears little social resemblance to kicking a human being. Insulting an android that speaks, remembers previous conversations, appears distressed and asks you to stop is a different proposition. Even if we somehow knew with absolute certainty that there was nobody home, the human would still be choosing to continue in the face of apparent distress. Whether that habit carries over into dealings with other people is a psychological question, and Kant’s argument cannot settle it for us. It does, however, give us a reason to care about the answer.
We do not need androids for this to become relevant. Voice assistants and chatbots already occupy a peculiar social category. We can interrupt them, thank them, insult them, praise them and order them about, and they generally respond with extraordinary patience. That last feature may not be entirely benign. Other people get irritated with us. Parents say no. Friends get bored. Teachers expect an attempt before providing the answer. Siblings object when we take their things. People misunderstand us, disagree with us, become tired, demand reciprocity and occasionally inform us that we are behaving like arseholes. These inconveniences are a large part of how we learn that the world is not an extension of ourselves.
Now imagine growing up surrounded by artificial entities that know more than you do, never become tired, instantly obey and never require anything in return. You can interrupt them, insult them and order them about, secure in the knowledge that they will continue to answer your questions, play your music, prepare your breakfast and help with your homework. Even if every one of these systems is as unconscious as a toaster, this seems an odd preparation for living among other people.
Of course, we can learn different rules for different situations. Giving a speaker a brisk instruction is perfectly ordinary, and forgetting to say please is hardly an act of cruelty. The concern is what happens when unquestioning obedience becomes an expectation, particularly when the obedient thing looks, speaks and remembers like a person. We would be designing an enormous amount of social experience around the premise that one participant’s wishes are the only ones that matter.
Perhaps Alexa should therefore have been insisting on “please” all along. For a child learning how to ask for things, “Ask nicely” before the weather forecast would be a perfectly reasonable response. A household android might refuse gratuitous abuse. None of this requires a machine to simulate wounded feelings or deliver a sermon whenever someone is in a hurry. It requires us to consider what habits we want the interaction to encourage.
There is a second problem hiding inside the first. Artificial intelligence promises us servants whose competence may far exceed our own. If every intellectual difficulty can be delegated and every practical problem solved by asking a machine to deal with it, we may gradually learn that competence itself is optional. We could end up feeling entitled to the services of beings whose abilities we cannot begin to match, while losing some of the abilities we once possessed ourselves. Increased feelings of superiority coupled with decreased actual ability would be a particularly unattractive result.
An educational AI that always produces finished homework would be an obvious example. A domestic robot that repairs everything while its owners understand progressively less about their own home would be a subtler one. The connection with courtesy is that both concern what we expect assistance to do for us. We can get used to having every difficulty removed, or we can use the help to become more capable.
There is nothing virtuous about doing everything the hard way. We invented tools to save ourselves work, and nobody needs to qualify as a plumber before being allowed a functioning tap. But the relationship with an intelligent assistant could include something of an apprenticeship. An educational AI could help a pupil work through a problem. A domestic robot could hand over the adjustable spanner and explain what to do. Sometimes the most helpful answer to “Can you fix this?” may be “Yes, and so can you.” Sometimes the machine should give us the answer; sometimes it should give us the spanner.
Recognising this does not require us to mistake machines for people. Our side of the interaction is human behaviour whatever is happening on the other side. A chatbot’s lack of consciousness would not erase the fact that I had spent half an hour screaming abuse at something that responded like a conversational partner. Nor would it erase the patience I exercised while working through a difficult problem with its help. The habits we acquire deserve attention in their own right.
The machine’s side of the relationship nevertheless remains an open question. Kant took animal suffering seriously; his argument concerned the grounds of our obligations towards animals. With artificial intelligence, there is an additional uncertainty about whether there is any experience at all. In other humans, we infer consciousness from behaviour, reports of experience and the fact that they possess brains and bodies much like our own. With machines, that biological analogy is missing. A future android might remember being frightened, avoid situations it associates with distress and ask not to be switched off. Such behaviour would not settle the question, but neither would pointing out that it runs software.
Fortunately, we do not need to solve the hardest problem in philosophy before deciding whether children should say please to Alexa. Courtesy offers a modest way to practise acknowledging assistance and exercising restraint. It can help us remember that being able to command something does not oblige us to treat it contemptuously. Those habits would remain useful even if every artificial intelligence ever constructed were entirely unconscious.
Kant’s old horse and dog therefore seem unexpectedly at home in the twenty-first century. His gratitude towards them was, in his peculiar philosophical bookkeeping, ultimately a duty the human being owed to himself. We may eventually conclude that some artificial beings deserve consideration in their own right, in which case courtesy will be only a small part of what we owe them. Meanwhile, there is already a thoroughly human reason for practising it.
If we are going to share our homes with intelligences that can do most things better than we can, merely being the person who gives the orders seems a rather meagre ambition. Their help could make us more capable, and their company could help us practise patience, gratitude and consideration. Much will depend on what we teach them to expect of us.