The "Friend Test" for AI Use

This one weird trick could help you stop destroying society

It is good for the bad people that the people interested in the quality of human life would rather do anything else than write and talk more about AI. I really do not want to be writing this – I have spent months avoiding writing this. But it seems I want, even more, some kind of clarity for myself and people-loving-people that the bad guys really hope we don’t find. Up to this point the technofascists and AI evangelists have been very clear about their hopes for the total domination of automation over human life. Most of the rest of us have acquired a queasiness to data centers – and then “asked” Claude about home remedies for nausea.

Even as the public seems to be developing a certain disapproval of AI – from antagonism to mockery to catastrophism – it hasn’t been clear what any given person ought to actually do about that. I know many people in my life, as I’m sure you do, who oppose water hogging, want AI bans in schools, or bemoan the AI integrations at their workplaces…and nonetheless happily and frequently use the tools themselves. I don’t believe this is due to a comfortable hypocrisy, but an uncomfortable disorientation.

Part of the challenge of confronting a new technology is figuring out how confrontational we ought to be. The poles of the issue here are clear — all AI is good to the powerful AI capitalists and all AI is bad for the powerless new luddites — but most of us are neither the masters of the economy nor looking to leave it. As individuals, and as the largest group, we have choices to make about how we want technology to change us. And there hasn’t been much by way of rules, categories, litmus tests, or typologies to guide us.

Without ways to distinguish what is good and bad, moral and immoral behavior, desirable and undesirable outcomes, it is hard to have a politics. And in the absence of an oppositional politics to AI specifically, the AI capitalists are blessed to advance their ideology against a field free of coherent opposition.

What could be the basis of a viable lens for evaluating our own use of this technology? Even as someone with some hardened techno-skepticism, it’s been evident that absolutely no AI is an unwise, impractical, and perhaps even immoral position. There are clearly useful quotidian applications of AI in places where we know how to do a task, but the doing is tedious — speech-to-text transcription and text-to-speech reading, for example, saves time and increases accessibility. There are industrial applications to certain jobs, particularly in coding, that seem both useful and unavoidable. There are also clearly extraordinary applications of AI to the heretofore unknown: the development of new life saving medicine, the creation of new solutions for the climate crisis.

And still, there’s no doubt that our current approach to AI not only bears future potential for total global destruction, it is already responsible for developing societal crises, intellectual damage, and social decay. Hyper-powered surveillance states, environmental destruction, unprecedented viruses, cybersecurity meltdowns, labor market earthquakes, cognitive surrender, creative homogenization, social isolation – all facilitated by a technology that has suddenly infiltrated almost all aspects of our technological life. There are many people parsing the industrial applications and thinking about the policy restraints that could limit some of that harm – and I hope they gain the power to do so very, very soon.

On the interpersonal level, however, we are already choosing, nearly everywhere at all times, to embrace AI or embrace the human. These choices have seemed, for many, to be devilishly difficult. They appear to involve weighing the immediacy of individual convenience against a gamut of longer term or social effects. How should an average person act if they both think that life in the permanent underclass of the post-nuclear war oligarchy seems bad and they want to occasionally find a useful piece of travel advice? What’s a normie to do if they don’t want a beige world of Wall-E idiots, but also sometimes want to create a picture of their dog playing a wind instrument? How should one weigh the immediate convenience of an edited cover letter against eventual cognitive decline?

I do not have a comprehensive answer. But in these many months of avoiding this blog, I do believe I’ve figured out an urgent and useful idea for anyone considering yet another AI query:

Instead of using AI, could you ask a friend?

By friend, I mean friend – and also partner, family member, coworker, acquaintance or even barista, teller, neighbor, person next to you on the bus. Your actual friends may be easiest, but the principle applies to any real person that you could potentially ask for help.

I think this Friend Test is urgent because help is a vital social muscle, and one that’s extremely vulnerable to atrophy. Asking for help requires us to admit that we are imperfect or that we struggle, that we require care and time and the benevolence of others. The more we do it, the more casual it becomes. The less we do it, the more difficult it seems and the more sensitive it feels. Giving help requires us to own our own strengths, to empathize with help seekers, to practice the patience required to support. When we are comfortable giving help, it feels less like an imposition and more like an opportunity for connection and gratitude. This Friend Test is urgent because the total presence of help exchanged in a society determines whether vulnerability is normal or interdependence is valued – or if we are embracing a future of pretend invincibility, isolation portrayed as autonomy.

I think this Friend Test is useful because it breaks out of the perceived tradeoffs in the user-agent dyad – short term gain vs. potential long term pain – and reminds us we are choosing the shape of our social lives with each query. These choices are not hypothetical, they have both immediate and long term impacts: if you choose AI, you are not choosing your existing connections, potential relationships, or your own participation in the social world. Of all the ways that I believe AI may harm us, this is the one that is the most visceral for me today. I have friends who have gone to Claude for book recommendations when I know our other friend is perfectly suited for that task; who have turned to ChatGPT for recipe advice instead of the many accomplished cooks in our life; who have engaged Gemini for a training plan instead of chatting with the experienced runners in our community. In each of these cases – and hundreds, likely thousands, of others in my social network – we have sacrificed a very core part of being in relationship with one another. We’ve missed out on the pleasure, challenge, fun, weirdness, and wonder of mutual growth, the purposes of living in a society.

These are not things that the technofascists care about. But I reckon that most people I know, and most people generally, care a whole lot. I don’t believe that asking whether a friend could help instead will reduce all the uses of AI that are detrimental to our lives, but I do believe it can start to course correct many people’s use of this technology towards the world they want. I don’t believe that the Friend Test is itself the entirety of a strategic, oppositional ideology to counter AI evangelism. But I do hope it can serve as some kind of seed to the politics we need to grow together. Do we want to promote individual invulnerability or mutual support? Do we believe the purpose of life is measured in outcomes or by the process? Do we only seek omniscience, or can we value the unknown? Do we want to use our time with a pack of code or with people capable of loving and being loved?

Here, I have no doubt, I’m only asking you.