Planet of Bots

Note: I won’t promise this is the last thing I’ll write about AI. But it might be. I think this sums up more than anything I’ve written how I feel, what I think, where we’re going, and why.

---

Photo by Mohamed Nohassi on Unsplash

Several years ago, I worked in the marketing department of a local mortgage company. Part of my job was helping lead new hires through the onboarding process, which included a section on information security. One of the modules concerned “social engineering.” Social engineering in this context means behavior meant to trick or manipulate another into making a mistake. The kind of social engineering we were trained to look out for consisted of a person calling the company and trying to obtain confidential financial information, often by posing as a client, real estate agent, or banker.

The training videos depicted a person calling the office, offering little to no information about themselves, and instead asking a few seemingly banal questions, like “Who is on vacation this week,” or “Is there someone I could talk to in sales.” The point of these questions was to obtain the name of somebody, anybody, who actually did work at the firm. Soon after getting a name, the person would hang up, and call back later, and this time use the name he had obtained to get more information from an unsuspecting employee. “Julie told me to call,” or, “I know Julie’s on vacation, I just need some info.” If you’ve ever seen Spielberg’s Catch Me If You Can, you get the gist. Some fast, vague talking to a naive worker can get a determined fraudster the access he needs.

The only defense against this trick was to stop treating conversation in a human way. Instead, we had to treat each unexpected interaction as a potential attack. The appropriate response to “Can I talk to someone in sales” was, “May I inquire as to the purpose of this call?” If someone claimed to forget the name of the person they were trying to reach, we had to verify their identity as much as possible before trying to help. If someone reached out and wasn’t able to quickly provide such verification, we had to assume the call was a trick, even if other factors indicated it probably wasn’t. An entire apparatus, consisting of numbers, data, signatures, the like, had to replace a normal interaction for the sake of security.

One of the effects of this training was to make the phone itself a suspect technology. Guarding against social engineering meant talking in a way that was stiff, impersonal, and accusatory. When the phone rang, nobody heard a potentially positive business call or a friend reaching out. We heard footsteps, nefariously creeping closer to our collection of social security numbers and bank statements.

Talking like a policeman or detective isn’t particularly fun. It’s a stressful, antihuman way of responding to the presence of another human being. The psychological burden of having to think, feel, and talk like a security system is real. Many of my coworkers felt it far more keenly than I did. Being forced to behave like a non-person, and treating someone else like a non-person, even for a few moments a day, is not very good.

One of the curious dilemmas of the AI age is that AI bots sound like us because we sound like AI bots. The most intense form of this dilemma is happening right now in magazines, blogs, newspapers, and book publishers. In desperation, many organizations that want to prioritize human writing are turning to software that purports to detect AI generated text. Some software does this better than others. But none do it perfectly, and I’m not just talking about software’s mistaking AI generated text for human-generated. The reverse situation is happening too, wherein text that is genuinely human is being flagged as AI-generated. Why? Because the human-generated text sounds like AI, and the software’s only job is to use advanced algorithms to detect which clauses, phrases, and sentences sound like something an LLM would churn out.

But what does an LLM churn out? The answer is, of course, human writing. All LLMs are trained on real human writing. The reason the way an LLM talks to you sounds human is because humans fed the LLM trillions upon trillions upon trillions of human sounds, and then programmed it to learn the ingredients of those and reproduce them in various ways.

AI companies have made the purposeful, conscious decision to use human language to blur the distinction between humans and machines. For the foreseeable future, it will not be self-evident to anyone what a human sounds like versus what a machine sounds like. Right now, we are hoping that machines themselves can help us spot the differences.

But my point is this. The last few years have been the climax, rather than the beginning, of the robotification of human language and identity. LLMs are bringing what we’ve been doing in the last few decades to a logical, profitable, and psychologically acidic conclusion. Nearly every technological innovation of the last twenty years has resulted in we human beings acting more and more like the bots that we are now replacing ourselves with. We have arrived at the AI era perfectly primed to accept the conditions of life on a planet of bots, because, phenomenologically at least, we’ve been living that way for a while.

---

Just about everyone I know says they dread the ringing of their phone these days. For millions, spam and junk calls have become far more frequent than any other kind of call. Americans get almost 3 billion spam calls every month, a reasonable explanation for why a lot of people admit they don’t even bother picking up the phone for an unknown number. Amazingly, there appears to be no great fix to this. Certain apps will charge you a monthly fee to screen calls, giving you more info and blocking high-probability junk callers from even getting your device to ring. None of these prophylactics are a sure thing. Most of us just resign ourselves to the three inevitable forces: Death, taxes, and spam.

Customer service, also, has become radically robotizized. Every single industry wants you to use their app, their chat service, or even their social media “help” account, rather than try to talk to a person. A truly human face or voice can, we know, decide to help more urgently and more thoroughly if they encounter a desperate enough person. Bots are unmoved. No hungry children, no sick passenger, no bereaved family can leave one impression on a bot. The bot’s currency of sincerity is data: your order number, your confirmation code, your two and three and ten-step authentication. An acquaintance of mine, adrift in the aftermath of a canceled flight and lost luggage, was even swindled by a fake customer service account. Again, the goal seems not to help customers or solve problems, but to reduce human on human encounters as much as possible.

The one scenario where bots appear to be mildly convenient is when the situation was easy anyway. If you’ve forgot something, for example, that was sent to you in a half dozen emails, there’s a decent chance a bot can simply remind you (more likely, they’ll ask you to confirm your identity a few times, before sending you a verification email, before sending you the email you actually need). Bots excel at doing the minimum possible.

This is a big reason why bots are bad writers. LLM-generated prose is, by virtue the process that creates it, frozen verbal leftovers. An LLM is perfectly capable of imitating a corporate professional’s staid compound sentences, especially the most basic, unsurprising constructions (“This is not that; it’s actually this.”). And because the vast majority of human have no desire for anything better, LLMs appear to be good writers. But they’re not. We are bad writers. The machines we’ve created can do only the bare minimum passably well because the same is true of us.

A little while ago there was a study done on what sources the major LLMs were using when a user prompts them with a relationship question, like “My girlfriend does this, why does she do it,” or, “I thinking I want to leave my husband. Should I?” The study determined that the #1 source of text that LLMs use to predict an appropriate respond to a relationship prompt is Reddit. Reddit is, of course, a notorious graveyard for life insight, for the simple reason that the kind of people most likely to publish regular content on Reddit also tend to be the kind of people who are bad at life in general and at relationships in particular.

Any human with a lick of experience and rationality can determine this with just a few minutes spent on Reddit. It doesn’t matter how much of Reddit you feed to a bot; it will never put this together. You see the cycle this creates? A person unable to navigate basic relationship stuff asks AI what she should do. The AI responds by reproducing slight variations of Reddit’s below-ground lifestyle content. The person follows this advice because she’s only slightly better than AI at discernment. Why? Because long before the relationship, through a long, complex pattern of choices, environment, and values that likely started before puberty, she has learned not to see any real difference between people and machines. She has become a bot.

---

Photo by Almaz Nourzhanov on Unsplash

In 2017, the leftist writer Freddie DeBoer wrote an essay titled “Planet of Cops.” In it, he lamented liberal political culture for the way it was encouraging people to try to catch and cancel each other. “The woke world is a world of snitches, informants, rats,” he wrote. “Go to any space concerned with social justice and what will you find? Endless surveillance. Everybody is to be judged. Everyone is under suspicion. Everything you say is to be scoured, picked over, analyzed for any possible offense.” DeBoer concludes:

See, the panopticon says we all get watched all the time, but there’s still a division between the guards and the prisoners. There’s still people who do the watching separate from the watched. And that’s not real life. No, in real life we’re all guards and prisoners at the same time. We are all informants on each other. Contemporary political culture is an autoimmune disorder. Do you enjoy living like this? Are you not exhausted?

DeBoer pointed out that left politics was unable to achieve robust change in things like policing and criminal justice, partly because its culture was operating like the systems it claimed to be unjust. The “planet of cops” wasn’t just a bad place to live. It was a sclerotic place to live. Almost nothing can be changed for the better if the people trying to change it are constantly at war inside their own tents.

Something very similar is going on with our bots. I suspect the most important reason it is so hard to get a groundswell of consensus against the dominance of AI is that most of us have been living for years as if the singularity were already here. What I mean is that for the nearly the past two decades, our work, our relationships, our self-help, and much of our school and art has been robotified.

Consider your own life. Think about the top five people that you text most often. If you found yourself needing to prove beyond a reasonable doubt that these people were real people and not bots, without leaving your home, could you do it? Could you show photos of shared meals, trips, and experiences? Could you go find that book or memento they gave you years ago? How much hard evidence could you really provide, short of actually getting in your car, that the people in your phone exist as more than a name and a thread?

Could most people do this? I don’t know. I think there are a lot of people for whom the idea of an AI friend or therapist makes intuitive emotional sense precisely because for all practical purposes, those are the only kinds of friends and therapists they have anyway. Insisting that people and AI can never have a meaningful relationship just doesn’t mean all that much if “relationship” has largely come to mean sending texts or DMs on a screen. We’ve been living like this for years. The tools of “connection” have become vehicles for faceless content exchanges at best, and relentless spam more often.

And this is to say nothing of the way that we’ve been reconciling ourselves for years to fake, manipulated, brain dead internet content. The expansion of social media sites from “networks” to “platforms” in the late 2000s and early 2010s created an entire economy: the attention economy, where the product is quite literally anything that someone might put in front of their face for a few seconds. If economies produce what they are designed to produce, then the attention economy was designed to produce overwhelming amounts of staged, non-sincere content. The question “why were they filming” became Internet shorthand to ask why a particular video seemed to start right when an outrageous or illegal thing was about to happen—as if somebody might have known what was coming. People pretending to have emotional meltdowns on camera (with the shot being perfectly calibrated to capture both the tears and their “good side”) is common.

You know who else pretends to have emotions? You know who else uses intimate language to sound real but is really just reading from a script? Bots.

The question most people want to ask about AI is, “Is this OK? How might it help me, and how could it hurt me?” We think of this technology exclusively in terms of how I might experience it in the moments I want to experience it. The question we should be asking is, “Do I and the people I love want to live with more bots?” That’s really what’s being offered. We’re not being offered the chance to have a little button in the corner of our workstation that can speed things up or make us smarter when we need a boost. We’re being offered the chance to live with lots and lots and lots of bots. Bots are AI. AI is bots. The only reason to accept this offer is if you think the way we’ve been living for the past two decades—the spam, the phishing, the tricks, the automated voicemail, the mandatory chat, the invisible friends and voiceless counselors—is a good thing.

---

Here’s what I think:

I’ve written two books, hundreds of essays, and done dozens of talks and interviews about being human in a digital age. In my opinion, in terms of creating actual change, none of this matters, and none of it will matter. At scale, there are only two things that matter: What people are used to. How they live out “normal” (not how they define normal). If talking to and learning from and interacting with real human people is your lived out normal, then you will have a problem with a planet of bots. Even if you’re not ideologically committed, even if you’re not ready to oppose a data center or get Chromebooks out of elementary schools, you will, if that’s your normal, constantly feel as the planet of bots is off. You will feel constantly lied to by it. You will feel constantly ignored by it. You will feel alienated within it. And even if the guys with see-through houses that sell for $100 million manage to convince you are the last person on earth who feels this way, you will still feel that way. You may not say it. You’ll feel it.

It’s a miscalculation to worry about the way AI might disrupt industries or the economy. That kind of thing has been happening for hundreds of years. It might be brutal, but it will balance out eventually. But our lives are a different story. The real worry is not that AI will take our jobs, but that it will make the idea of possessing a job incomprehensible. AI writing our books? Bad. AI reading our books? Much worse. The trajectory we’ve been on for a long time is not merely one where we get machines to make the stuff we love. It’s that we love the machines themselves more than anything they make.

There is no “solution” to the planet of bots, for the same reason there’s no “solution” to divorce, or bad neighborhoods, or addiction. The only thing you can solve is a problem, but these things are not “problems” in that sense. They are lived realities. They are generational, societal, existential cycles that get lived out again and again because what people take as being normal is what they take as being normal for them. These things will persist in the lives of all those for whom they remain normal. And they will change when, and only when, one, or two, or a billion people decide, “This isn’t normal, and it’s not going to be normal for me.” When one person breaks a generational cycle, they are often celebrated and congratulated. When an entire culture breaks a generational cycle, war usually starts.

You know I’m not calling for war. I’m just calling for honesty. We live in the planet of bots. And many of us have already begun the journey of accepting it as normal. We cannot feel ourselves drowning because we’ve been thinking we were fish the whole time.