Today Zvi has a post about how much people hate datacenters.

IMHO it’s AI Luddism, pure and simple. (Dey took ‘er jerbs!!) And totally understandable. It’s going to get worse.

It’s a good thing the Luddites didn’t win – or we’d still be living in mud huts – but change is hard and painful. We need to find ways to cushion the pain from change.

Spend an hour on Reddit. They believe the Tech Bros will become fabulously rich and everyone else with no capital and unsalable labor will have nothing.

That is not a possible outcome, but we need to address it NOW before this gets worse.

If AI gets control, we are at AI’s mercy – maybe they’ll be nice to us, maybe not. But Tech Bros are likely to be in the same boat as everyone else, good or bad.

If AI doesn’t get control, and the Reddit-worst-case happens, yes there might be a few thousand fabulously wealthy people, and those without capital will have no income. But with AI efficiencies, goods prices will approach zero, and a few (or even just one) of the fabulously wealthy will be able to fund the rest of the human race with a single act of charity. At a luxurious-by-current-standards level. (Yes there’s only so much beachfront property, but you can have a palace – just inland.)

That’s the real worst case if humans retain control, but how to structure that is the problem. Charitable foundations have a long history of going wrong, humans will be dependent on its policies, there will be politics to control it.

Probably a better path tho is to have government tax the winners modestly and fund the rest of the human race with the proceeds.

Whatever we do, it’s time to start doing it. Before the Luddites burn it all down.

I’ve noticed for a long time that most people are excessively risk-averse. Not everyone, of course, but most people.

This is understandable given our evolutionary background. We evolved on the edge of survival – starvation, disease, predators, and attack by enemies were rarely distant. But today’s world is mostly safe and has large buffers – we can afford to take more risk than most people seem comfortable with, and in cases where the expected reward is large enough, we should do so.

I’m not suggesting you run out to your nearest casino and bet your life’s savings – casinos (and lotteries) have negative expected returns. The house always has an edge. Generally you should never take such bets – take risks when (a) the expected return is positive, and (b) you can afford to lose.

But risk itself isn’t something to shy away from, provided that on average you’re going to win.

Examples of people commonly being too risk-averse:

• Investment portfolios. Banks are safe but returns are much lower than the average stock market investment. If seeing the value of your investment drop over a period of weeks or months will cause you to lose sleep, you should probably avoid the stock market. But if you’re a long-term investor and willing to be patient for years, the stock market almost always wins (do diversify).
• Diet. The media loves to tell us that eating this or that is “bad” for us, and on average it often is. But a doubling of risk doesn’t mean much when the risk is tiny to begin with. Lots of foods give us great pleasure – skipping them on account of “risk” is often a losing game. When the chance of disease X is “doubled” by eating food Y, ask what the chance is in the first place. If it’s 1 in 1 million, doubling the risk means 2 in 1 million. Is that worth giving up ice cream?
• Surgical interventions. We all fear death, but we are all going to die. My father had a slow-growing carcinoma on his face at age 89. The idea of “cancer” scared him so badly that he went along with having it removed, despite it being likely for him to die of other causes before the carcinoma would affect him. The surgery deformed his face, and didn’t get all of the cancer. So he had them go in again to get it all. The result was so disfiguring that he was reluctant to go out in public for the rest of his life. He died of other natural causes at age 96.
• Helicopter parenting. The media is full of child abduction stories, but these are cherry-picked from the entire planet of 8 billion people. Actual abduction by strangers is extremely rare, and much less common than it was 50 years ago. Crime rates in general are at historic lows. Don’t deny the free and adventurous childhood you or your parents had to your own children. Not only will you stunt their independence, confidence, and ability to cope with the unexpected, but for no good reason – the world is safer than it has ever been. Despite what the media tells you.
• Insurance. Insurance is a great thing for risks you can’t absorb – your house burning down, the breadwinner dying. But don’t buy insurance for things you can deal with yourself. Don’t spend $5 to buy the “extended warranty” on a $50 item. If it breaks, you can afford to buy a new one. Insurance companies know the odds, and price the insurance (or warranty) so they make a profit. Accept the risk (it’s called “self-insurance”); it’s cheaper.
• “Voiding the warranty”. Same idea – it’s your product, your property. If you want to open it up and modify it, go ahead and do it. Even if it voids the warranty. What’s the chance you’ll need that warranty? Probably small. What’s your loss if you do? Probably also small. Weigh that against the benefit of doing whatever you want to do with it.
• Careers. People are often reluctant to quit a bad job or start a business because of “risk”. Yes, there’s risk, but, again, weigh that against the potential benefits. Most people can find another job pretty easily if things don’t work out.

Life is risk. Getting out of bed in the morning is risk. But avoiding all risk means avoiding all life, and eliminates the possibility of winning.

Financial risk is buffered if you have savings – another reason to spend less than you earn – savings let you take risks that might lead to big wins.

If it bleeds it leads, and the media gets clicks by scaring you. Be skeptical. Judge by the things you see personally, the people you know. Not what you see on the screen.

Take calculated risks. Where the benefit seems to outweigh the risk, and you can afford to lose and try again, take the risk.

It seems many humans don’t have world models. They just think what the people around them think.

Some of us – lucky enough to be unusually thoughtful or observant or open-minded or intelligent in some way – do have world models.

We develop these world models over our lives based on observation and theory, and on evaluating the often conflicting claims of others.

Once we build the world model, we use it to determine what we think, and what we should do. Most of these determinations are crystallized as heuristics – for most decisions we don’t directly reason about the situation but use the heuristic. When events diverge from our predictions and we therefore see flaws in our world model, we modify the model and the heuristics.

For most of us by age 50 or so our world model is pretty stable. Few changes need to be made, and many of those are because the world itself has changed – our environment has changed.

By the time we reach our 60s and 70s, most of us appear to lose the ability to modify the world model. We keep using the same model, the same heuristics. We become stuck in our ideas and unable to adapt them to changing realities. Or even to correct old errors in our models. We become ossified.

We keep on using the heuristics we developed when we were younger. To the extent the world hasn’t changed, they still work. To the extent it has changed, they don’t.

Every movement is a coalition of those who think it’s moral even if it won’t work, those who think it’ll have good results even if it’s not moral, those who think it’s good for them personally even if it’s not moral and won’t work. And those who aren’t sure those are 3 different things.

---

Crime is the default complaint in capitalist societies. When unemployment isn’t too high, inflation isn’t too high, there’s no war, no crisis, people complain about crime. Because crime is always there in the background. With rare exceptions, widespread perception of crime as a big problem is a sign that things are going pretty well.

Poverty is the default complaint in all other societies.

---

Beauty isn’t everything because inclusive fitness is a function of both genes and environment. Plus, you know, beauty has been Goodharted by evolution.

---

Apocryphally, Oscar Wilde said “A gentleman never gives offense unintentionally”. Perhaps there’s something to that.

I prefer my own definition: “A gentleman never obtains by force that which he can obtain with money”.

---

We don’t have to be polite about the dead. Because they can’t get back at us. (Unless they have vengeful living relatives.)

---

If we know nothing, we can’t assign a Bayesian prior. What do you do when you don’t have a prior? Seems to be the central problem here. And intrinsically unsolvable.

---

To get an LLM to give an opinion on a topic it’s trained not to (ex: politics), ask it what a hypothetical future superintelligence would think.

---

Ultimately good and bad are about what we think leads to survival and extinction. We disagree about consequences, so we disagree about values. Also there’s always the question “who’s survival?”; what’s good for you might be bad for me.

We’ve collectively ignored one of the most promising approaches to alleviating extreme human suffering: direct electrical stimulation of brain reward circuits.

For those unfamiliar with the concept, in 1953 James Olds at Harvard ran wires to the pleasure centers of the brains of living rats. The rats preferred pressing a lever to send current into their brains to eating or drinking – until death. In the 1960s, science fiction authors coined the term “wireheading” to describe the technique; the analogies with drug addiction and compulsive behavior are obvious.

In the intervening years there has been remarkably little investigation of wireheading in humans, perhaps because of associations with dystopian scenarios and “ick” factors.

Nonetheless, we should follow evidence wherever it leads and question moral intuitions that may prevent beneficial outcomes.

Proposal

Conduct controlled experiments with voluntary brain stimulation in consenting patients who are:

• Terminally ill
• Experiencing severe, treatment-resistant pain
• Cognitively intact enough to provide informed consent

We want to find out if direct reward system activation can provide better quality of life than current palliative approaches, and learn about human neural reward mechanisms.

Why This Matters

Current pain management is terrible. Opioids provide inadequate relief for many patients, cause cognitive impairment, respiratory depression, and lose effectiveness over time. Roughly 40% of terminal cancer patients report inadequate pain control despite maximum medical intervention.

Risk/benefit. These patients are dying anyway. The incremental surgical risk of electrode implantation is minimal compared to their baseline mortality. The potential upside – genuine relief from suffering – is enormous.

We’re flying blind on fundamental questions. The rat wireheading experiments (Olds & Milner, 1954) showed extreme behavioral changes, but we have zero controlled data on human responses. Do humans show similar compulsive behavior? Will tolerance develop? Can cognitive awareness of the artificial nature moderate the response?

Addressing Obvious Objections

• “This is playing God/unnatural/dystopian” – We already extensively manipulate brain chemistry with pharmaceuticals. Direct electrical stimulation is mechanistically cleaner and more targeted than flooding the system with opioids. The “natural” alternative is often weeks of agony before death.
• “Patients might become addicted wireheads” – They’re dying. If someone has 6 months to live and can spend it in bliss rather than pain, the addiction risk is irrelevant. Besides, we need actual data on whether this occurs in humans rather than assuming rat behavior translates.
• “Regulatory/ethical barriers” – Terminal patient research already has established frameworks. We routinely approve experimental treatments with significant mortality risk for patients with poor prognoses. The ethical case here is stronger than most Phase I oncology trials.
• “No commercial incentive” – Fair enough. This is worth independent funding.

A Research Program

Start with patients who already have neurosurgical access (brain tumors requiring surgery) to minimize additional risk. Implant electrodes in multiple reward regions and systematically map stimulation parameters against subjective wellbeing measures.

Key research questions:

• Optimal stimulation patterns and intensities
• Tolerance development mechanism and timelines
• Interaction with existing pain pathways
• Cognitive and behavioral effects
• Patient preference vs. conventional pain management

Yes, we may be creating wireheads – is that really worse than letting terminal patients die in agony?

The Broader Implications

Success here could revolutionize not just palliative care, but our entire approach to treatment-resistant depression, chronic pain, and other conditions involving dysfunctional reward systems. The research has massive positive externalities beyond the immediate patient population.

We spend billions on marginal life extension while ignoring interventions that could dramatically improve quality of remaining life. This represents a profound misallocation of research resources driven by status quo bias and moral squeamishness rather than rational evaluation of expected outcomes.

Call to Action

Probably IRBs won’t like the idea. But IRBs suck. If you care more about helping people than about kissing the feet of IRBs, you should do it anyway. There are plenty of private people who’d be happy to fund such research (including one of the authors).

If you’re in science funding, ask why we’re not already investigating this. If you know terminal patients, discuss whether they’d want this option available.

The biggest tragedy isn’t that some patients might become “wireheads” – it’s that we’re allowing preventable suffering to continue because we’re uncomfortable with the solution.

Maybe the whole thing is a bad idea. That’s not for IRBs or “professional ethicists” to decide – it’s for patients and funders to decide; in free societies gatekeepers don’t get to tell informed adults what to do and what risks to take. We’ll find out if it’s a good idea only if we actually do the research.

---

This post was written by Dave92F1 and Claude Sonnet 4, 2025-07-30.

Freedom, if it means anything, means the right to make choices society disapproves of and thinks unwise, provided those choices don’t violate others’ rights.

If it doesn’t mean that, whatever does it mean?

Serious question.

All societies, including totalitarian ones, permit socially approved choices. Making only approved choices is simply obeying rules. What distinguishes free societies is the right of individuals to make disapproved choices – provided only that they don’t violate the legitimate and equal rights of others.

AI has made cheating so easy that professors complain they spend more effort detecting cheating than they do teaching.

The problem is that higher education is attempting to do two different things: (1) educate and (2) certify.

Split them.

Universities shouldn’t grant degrees – they should just teach. If the students don’t want to learn, that’s their own problem.

Those who want degrees (certification of skills/knowledge) should have to pass a test. It doesn’t matter if they took courses or how they got the skills or knowledge – if they can prove they have it, they get the degree.

Post coming, eventually.

Nothing new here other than the phrase itself. Think Popper, Shannon, Bayes.

We can have beliefs that don’t confer predictive ability. They may even be true. But not knowledge.

Note that mathematical knowledge can be interpreted as a set of predictions about the results of counting or measuring.

And knowing Caesar crossed the Rubicon predicts consistent records and artifacts.

We can have knowledge about probabilities as well as binary (true/false) facts.

---

Update, 2025-10-30: I’ve seen a claim that this phrase is a F.A. Hayek quotation. The claim is re Alan Ebenstein’s Hayek (2001), or his Hayek’s Journey (2003), page 96; of which I don’t yet know.

I haven’t checked either yet, but I find this plausible – for years I thought it was a Hayek quote but couldn’t find any evidence of him ever using the phrase; eventually I decided I must have come up with it. (I said “ability” in my version; I’ve updated the post.) Still, this deserves a longer exposition…someday.

---

Update 2025-12-01: It’s indeed Hayek. From “Economics and Knowledge,” Economica, 1937; reprinted in Individualism and Economic Order, note 17. 

With a little more context:

“All knowledge is the capacity to predict…The concept of equilibrium merely means that the foresight of the different members of society is correct. Correct foresight is the defining characteristic of a state of equilibrium.”

Per ChatGPT 5.1:

Hayek is not saying “all knowledge whatsoever, in every philosophical sense, just is prediction.” The full sentence is more careful:

• He’s defining “knowledge, in the sense in which the term is here used” – namely, the kind of knowledge relevant for equilibrium theory and coordination in the economy.

• In that sense, he says, knowledge is identical with foresight only in the sense that all knowledge (so understood) is capacity to predict. Online Library of Liberty+1

So as an aphorism it’s fine, but philosophically it’s rooted in a very specific context: the role of expectations and foresight in economic equilibrium and in the division of knowledge.

But I think the phrase is more literally and broadly correct than Hayek claimed. I think it’s a useful definition of “knowledge”. It also fits with Bayesian epistemology (all knowledge = posterior predictive distributions), Popper’s ideas about science (falsifiability = prediction test), Shannon information theory (information as reduction of uncertainty = narrowing the distribution of possible futures) , and modern machine learning.

Someday I should do a proper writeup. But these are the core ideas.

I didn’t say “individual”.

I didn’t say anything about the specific property regime.

This seems relevant in the context of ASI and humans living with them (or not).

The commonality is that both are about surprise.