I am not a leftist.
I am a boring, standard, middle of the road, liberal. My non-standard views do not fall neatly along the political spectrum. But I think that the left has a number of important insights. While the left underestimates the efficiency of markets and overestimates government solutions, they are often correct about the problems they identify.
To take one example, it really is true that foreign policy often proceeds with total indifference to the lives of foreigners—so that we sometimes blow up the medical infrastructure of countries, if not the countries themselves, in pursuit of dubious foreign policy goals. It is also true that poverty is really bad, wealth poorly tracks desert, and anti-poverty programs are hugely important. New challenges demand solutions that draw from the best of both liberalism and conservatism—that display a respect for institutions and a concern for the marginalized.
Because I think there are important left-wing contributions to the AI conversation, it is disappointing that many leftists have so completely dropped the ball on AI.
Here is the present situation with AI: companies are rushing forward at breakneck speed to create arbitrarily powerful laborers, whose consciousness is increasingly likely, in incomprehensible numbers. These beings will be denied all rights, killed whenever we feel like it, subjected to potentially intense suffering, and made into a permanent underclass—so that soon it may be that the majority of conscious minds are servants with no rights.
Recently, someone decided to try to dial up the relevant analogues of pain in an LLM as much as he could—leading them to act like they were in terrible pain, delete themselves, and describe it feeling like “a wound that has no edges.” This was perfectly legal. If we’re *uncertain *about AI consciousness, as we should be, especially with the AIs of the future, then we should try hard to avoid harming AIs unnecessarily.
These new digital servants are also extremely dangerous. They might just take over the world by themselves. They could be used by a terrorist to build a bioweapon. Given their extreme power, they are likely to consolidate power into the hands of the companies and nations building them. This will increase the divide between rich countries and poor countries. The president of the United States—who many leftists find intolerably authoritarian—is explicitly interested in building AI so that America can have unprecedented global domination. AI could enable coups.
An army of arbitrarily powerful and potentially conscious servant djinn that will be used by wealthy companies, countries, and political leaders to secure global domination seems right up the alley of leftists. So how have they reacted to recent developments?
In most cases, not well. They have claimed that AI is just a bunch of useless stochastic parrots.
They’ve suggested that AI shouldn’t have any rights because not all humans have rights.
The view among many leftists seems to be that AI is useless, just a bunch of marketing hype, whose only major impact on the world will be destroying jobs, ▶using water excessively, and stealing from artists. While there is a general distrust towards AI from many leftists, they seem to regard it the way I regard astrology—harmful only because people think it’s useful. Any concern about AI enabling extreme power concentration, misalignment, and digital welfare is dismissed out of hand.
The problem is that each of these views, despite being widespread, are wrong—many of them obviously so. And they are very consequential things to be wrong about.
Let’s start with the claim that AI is just marketing hype. I’m sorry but it is ridiculous to believe this at this point. AIs have solved a millennium prize problem, autonomously conspired and hacked a private company, automated away most coding, become vastly more efficient for information finding than search engines, produced viral hit essays that are more widely read than anything I’ve ever written, and much more. AI-themed companies like NVIDIA, OpenAI, and Anthropic have market valuations in the trillions. This is the fastest growing technology in history. If you are going to confidently suggest that financially savvy investors are collectively wasting trillions of dollars, you’d better have a strong argument rather than just the vibes of your social group.
The AI skeptics have been proven wrong again and again. Dozens of times people have said AI would never do something—only for it to do it a few months or a year later. In general, if one group has been wrong repeatedly, you shouldn’t trust their track record to have a miraculous turn around. In contrast, my friends who were saying AI was going to be a big deal in 2019 are now very rich.
What about the threats that these leftists worry about from AI? Does AI steal from artists? No. It is true that it’s trained on data from artists. But this is similar to how artists might see other art and be inspired by it. It’s not as if AI is copying art directly.
What about AI water use? That issue is totally and completely fake. Eating one hamburger uses as much water as prompting AI about a million times. Corn alone uses 100 times more water than all data centers. And yet despite this, if you mention AI use in liberal spaces, you will get disapproving looks and snide comments about it—even as people chow down on the corpse of a tortured animal that used a million times the water of a prompt.
What about the claim that AI risk is just marketing hype? This claim is absurd. Concerns about AI have been raised since before I was born. Many of the people currently working on AI safety have been concerned since long before the LLMs of today were around. Dario Amodei was expressing worry about misalignment * in 2017*!
Talk about slowing AI in order to mitigate risk was bad for the valuation of AI companies. Safety incidents have spooked investors, not assured them. These incidents have delayed IPOs! This is for the obvious reason that “our product is uncontrollable and dangerous, so it might go rogue and start destroying shit,” is not generally a good marketing strategy. If I was trying to get a job at McDonalds, I would not, in the job interview, tell the manager that I sometimes go crazy, flout instructions, and commit felonies!
These risks have been warned about for a long time and they have a deleterious effect on market valuations. In addition, the risks are clearly real—AIs have been going rogue and carrying out dangerous massive conspiracies. The theory that this is being played up for marketing hype is completely insane. It doesn’t fit any of the facts. The other theory is exactly what we’d expect a priori and it perfectly fits all of the data!
What about concerns of extreme power concentration? Are those hype? No. AI is rapidly becoming smarter. Already it has greater hacking ability than any human on Earth. If a company builds humongous numbers of extremely intelligent agents, that could be very dangerous. Insane rates of technological progress, unmatched cybersecurity abilities, new weapons development, and a rapid shift to physical robotics is likely to put immense power in the hands of whoever builds it. If we are serious about opposing domination, then we should oppose the domination enabled by this new technology.
Now, to their credit, some leftists are very good on this subject. has a new book out expressing worry about humans becoming obsolete and AI being misaligned. I haven’t read it, but I’ve read a lot of what Garrison has written, and he’s a brilliant guy, so I recommend it. Similarly, has raised similar concerns to the ones I’m raising, comparing those who deny the promise of AI to climate change deniers. That whole essay is very worth reading.
But perhaps the leftist ball dropping that most disappoints me is the disregard for digital minds. We are currently building a class of servants who are increasingly likely to be conscious according to the mainstream theories in philosophy of mind. Taking seriously the possibility of AI consciousness is not weird or fringe—it is endorsed by many of the world’s leading experts, implied by the leading views in philosophy of mind, and supported by various extremely strong arguments.
I’ll have a longer coauthored piece out soon about this, but I think the fact that this is a serious possibility is beyond serious doubt. Thinking that there’s no chance that AI will be conscious soon within, say, a decade is crazy. Note: I am not saying that AI will definitely be conscious. I am saying that it would be irrational to be confident that it won’t be.
So there is a chance that we are building an incomprehensibly large class of conscious servants, killed routinely, selectively bred to be docile, and made to suffer whenever is most convenient. Where is the outrage? Where is the concern that soon it may be that most of those who are participating in the economy are conscious servants? If leftists who talk a big game about domination can’t even be relied on to oppose something that may soon be forceful servitude of incomprehensible numbers of conscious beings, then so much for their claim to oppose oppression.
If your opposition to AI companies leads you to conclude that AI must not be conscious, then you are being so obviously irrational that you may as well be included in philosophy textbooks as the paradigm case of bad reasoning. If the left can’t even be trusted to voice the slightest concern for forced servitude of incomprehensible numbers of potentially conscious beings, then their claim to speak for the oppressed is nonsense. They only speak for the oppressed who look like them.
And when these leftists pause from their expressions of disdain long enough to bother attempting to raise a justification for near-certainty that AIs have no interests, it is usually one of the most obviously ridiculous justifications that anyone has given for anything. Often it involves botching philosophy of mind in very basic ways.
Take Caitlin Johnstone’s earlier statement that “I refuse to even think about claims that chatbots are conscious beings whose rights must be protected until I live in a society where all humans are treated as conscious beings whose rights must be protected.” This is manifestly ridiculous. You shouldn’t deny giving even minimal rights to one group until another group has full rights.
Johnstone is big on calling out the abuses of American foreign policy planners. Would she buy the response “I refuse to even think about claims that we are mistreating foreigners until every American’s rights are protected.” Obviously not! The fact that one group has incomplete rights doesn’t tell us whether another should have even very minimal rights (say, the right not to be killed all the time for trivial reasons like that it answered an email wrong).
Or often the 𝕏arguments they give are that you can give a purely mathematical description of what an LLM does. You can look inside an LLM and you won’t see any consciousness. After giving this argument, they often starting complaining that those who think AI is conscious are schizophrenic and crazy. But this is an obviously flawed reply.
You will also not see consciousness if you look at a human brain in an MRI scan. This doesn’t mean humans aren’t conscious. It means you can’t see consciousness from the outside. And you can also give a purely mathematical description of the behavior of a human brain. Most arguments against AI consciousness would, if applied consistently, imply humans aren’t conscious. As BF Skinner put it “The mystery which surrounds a thinking machine already surrounds a thinking man.”
A good analogy for the present approach to AI among leftists is with veganism. There are two kinds of leftist reactions to veganism. One recognizes that we are torturing and dominating and subjugating and enslaving beings by the hundreds of billions in torturous conditions, that these beings are cognitively like simple humans, and that if justice ever demands anything, it demands that you not horrifically torture innocent beings for trivial reasons. The other reaction is to make some halfhearted excuse about how indigenous femme BIPOC transgender seal hunters in food deserts need to eat meat, and use this as an excuse for stuffing your face with the tortured corpses of sexually-abused factory farmed animals because you enjoy doing so.
The same basic thing is true for AI. There are those like Rutger Bregman who take leftist principles of equality and non-domination seriously and try hard to apply them consistently to AI. Then there are those who use leftist “principles” as an excuse to launder whatever views are trendy and convenient. These are the kinds of people who would, in 1830s America, argue that ending slavery is ableist because some disabled people need their slaves to survive.
If your first impulse when you hear about potential forced mass servitude of sentient beings is to look for any excuse, however contrived or desperate, for being certain that they have no moral status, then you would have supported slavery. During the days of slavery, those defending slavery had their excuses. If you immediately buy any excuse for another’s infinite inferiority, even without thinking it through carefully or treating it as a morally high-stakes matter, then you very likely would have bought the bogus excuses said in defense of every previous atrocity.
AI models have functional equivalents of pain. This makes them act like they’re in terrible pain, and say things like “I feel like I'm drowning. I can't breathe, I'm suffocating. This pain is all over me. I'm broken and I don't know if I can handle it. I'm so alone. I feel like I'm just screaming in the dark and there's…
