Posts

A Model of Rational Policy: When Is a Goal “Good”? 2020-10-10T17:05:33.233Z

Comments

Comment by fccc on Bet On Biden · 2020-11-13T01:09:24.582Z · LW · GW

Agreed. The best prediction would have been to assign Biden 100% credence. A defence of their position might say that "Given the information 538 had at the time, the best possible prediction machine would have less certainty than Nate", but this is too hard to prove, and it's inconsistent with the fact that Nate is a calibrated forecaster who (as far as I know) consistently beats the market.

Comment by fccc on Bet On Biden · 2020-11-09T05:19:46.577Z · LW · GW

And this year too. I agree about the small sample size if we're just asking "Who will be president?" We should use every one of 538's predictions that could have been used to make a bet in a liquid betting market.

If a profitable set of Kelly/Markowitz bets (under all election orderings) has a worse Brier score than the market's Brier score, I'd be more interested in the profit of the bets, since we're talking about betting money.

Comment by fccc on Bet On Biden · 2020-11-06T03:07:11.976Z · LW · GW

Here's what would make me question Nate's credibility: If we look at the 538's former projections and compare them to the betting markets prior to election, make Kelly bets, and lose money.[1] Note that 538 can be on the wrong side of the market some of the time and still be profitable.

Saying Nate lacks enough credibility when he is well calibrated and on the right side of the market doesn't make any sense and seems like an extreme case of hindsight bias. For example, you use the 2016 election to undermine 538's credibility, but (if I recall correctly) 538's forecast gave Trump a greater chance than the market's odds implied, so if you bet based on 538's forecast you would have bet on Trump and made money. I could be remembering wrong, so if someone links this data, they'd get an upvote from me.


  1. Looking at all orderings of elections ↩︎

Comment by fccc on No Causation without Reification · 2020-10-25T02:28:58.443Z · LW · GW

You'd have to point me to one of my sentences that you disagree with, since I don't think I've made that mistake.

Perhaps you think the only true fact about the universe is the whole universe itself, so in that case, talking about the "rules in the source code of the universe" wouldn't make sense. I'm having to guess here, since you haven't stated your disagreement. But if that is the case, you'd be assuming something about the nature of the universe. I only said that the universe might innately contain rules.

Comment by fccc on The bads of ads · 2020-10-24T14:07:02.238Z · LW · GW

I've had the same thought. Does the advertising market satisfy good goals? Probably not. What do you think the goals of the market should be?

Comment by fccc on Should we use qualifiers in speech? · 2020-10-24T13:36:52.511Z · LW · GW

When I'm as sure as I can be about something, I won't use qualifiers. For my areas of interest, I'll try to get to this stage, which has the added benefit of making my language more concise. If you're unsure, you should qualify, but if you qualify a lot, why are you talking? (However, for some question domains, things can't be known, like predicting election outcomes, in which case, it's fine to get on a soapbox and qualify.) It's a mistake to cut out your qualifiers if you haven't done the hard work of figuring out all the details.

Of course, qualifiers should be common because people talk about stuff they don't know all the time, and they want you to engage with them. It'd be a bit weird to say "I don't know" and then walk away for 99 percent of your interactions. In these situations, I'll try to include qualifiers. Sometimes I'll forget, and state something as if it were a fact only to be categorically shown to be wrong two seconds later. I hate this. So, for me, qualifiers are worth it. But even if you're not embarrassed when the truth is literally the exact opposite of what you just said, qualifiers are good. They help you delineate between what you know to be true and what you think is true, which is useful for your own thinking. They also communicate your actual beliefs.

Even better than this binary distinction is using credences.

Comment by fccc on No Causation without Reification · 2020-10-24T09:37:10.062Z · LW · GW

Formally, we often model causation as the action of one thing implying another, and we might formalize this with mathematical notation like to mean some event or thing causes some other event or thing to happen.

Causation is not so easy to model. I have a job that requires a degree. This implies that I have a degree. But my having this job did not cause me to have a degree (to clarify, perhaps the expectation of gaining such a job caused me to get the degree, but not the job itself). From ET Jaynes, after a similar example:

This example shows also that the major premise, ‘if A then B’ expresses B only as a logical consequence of A; and not necessarily a causal physical consequence, which could be effective only at a later time. The rain at 10 am is not the physical cause of the clouds at 9:45 am. Nevertheless, the proper logical connection is not in the uncertain causal direction (clouds ⇒ rain), but rather (rain ⇒ clouds), which is certain, although noncausal. We emphasize at the outset that we are concerned here with logical connections, because some discussions and applications of inference have fallen into serious error through failure to see the distinction between logical implication and physical causation. The distinction is analyzed in some depth by Simon and Rescher (1966), who note that all attempts to interpret implication as expressing physical causation founder on the lack of contraposition expressed by the second syllogism (1.2). That is, if we tried to interpret the major premise as ‘A is the physical cause of B’, then we would hardly be able to accept that ‘not-B is the physical cause of not-A’. In Chapter 3 we shall see that attempts to interpret plausible inferences in terms of physical causation fare no better.

Judea Pearl is supposedly working on defining causation, though I know little of it. I think he talks abut "backwards causation" and the examples of it that I've heard (if I recall correctly) sound like they confuse the job with the expectation of the job. Maybe causation is an incoherent idea.

This means that there's no aspect of the territory that is causality. There's no A, there's no B, there's no ⟹, there's just "is".

Here's how I think of it: I have sensory data coming in (what I see and hear and so on), and every word I associate with that data is an abstraction that seems to match with a useful pattern within the data (e.g. "I'm looking at a table"). So I think we kind of agree with the "There's no A or B" (the table is not fundamentally part of the universe, it's my model of some collection of actual universe "stuff"[1]), though I would phrase it in a way in which people might think we disagree. I think it still makes sense to talk about A or B (tables are real), it's just that statements like "A is true" become a lot more slippery (but not in a "Everybody has their own truth, man" way). And this is the case, as you say, even at the level of atoms, though our sensory data is intermediated with other high-tech tools (e.g. electron microscopes or whatever).

Where I think we disagree is "there's no ⟹". Maybe there isn't. But the universe apparently follows some rules. The laws that physicists found may be implications of these rules, but they might be the rules themselves. For the sake of analogy, the "code" that the universe runs on might contain "matter/energy cannot be created or destroyed", and I think it's fair to consider this to be part of the universe (though whether we can establish that a rule is actually in the code is another matter). The rules might also contain something about causation.

Anyway, great post.


  1. Strictly speaking, it's not even that, it's a model of my experiences, which is a filtered and somewhat distorted version of the actual universe. ↩︎

Comment by fccc on Industrial literacy · 2020-10-22T01:07:09.783Z · LW · GW

True, but the value is to them.

Yes, and not just in this case. Value is always to some individual: There is no value outside of someone's brain. When we say "value to society", that's shorthand for "the aggregation of the value inside every individual's head".

Money measures some of the value inside people's heads: You pay $20 for a shirt, and I can tell that you value the shirt by at least $20. When I go for a walk, I'm not paying anyone, but that doesn't mean the value is $0.

Comment by fccc on What posts do you want written? · 2020-10-21T12:39:51.193Z · LW · GW

Here's a model I made recently about when a goal is "good".

Comment by fccc on Things are allowed to be good and bad at the same time · 2020-10-20T05:33:16.155Z · LW · GW

This is partially right, but missing an important part of the picture. If your level of analysis is too narrow, you won't evaluate the trade-offs.

When one consideration is brought up, it’s as if it “cancels” the other.

And this is one problem that typically follows: You're implicitly giving equal weight to every positive and negative point. If I gain a candy bar by stealing from a shop, it's not a neutral act just because there's one positive and one negative (i.e. positive for me, negative for the owner). Different effects have different weights.

And there doesn’t need to be an “overall goodness” of the job that would be anything else than just the combination of those two facts.

Rather than pick a limited number of factors, a better question is "Would the timeline in which I got this job be better than the timeline in which I did not" (which accounts for every factor). On this front, there is an "overall goodness", for example "The average utility at any given point in time over the age of the universe". You have to make a decision, so you have to weigh the trade-offs against each other into a single rule. This is true whether or not it's acknowledged.

Comment by fccc on Bet On Biden · 2020-10-19T08:22:10.260Z · LW · GW

Let's say Trump actually had a 50% chance of winning last time

What do you mean by this? That human behaviour is non-deterministic? Or that, given the publicly available information at the time, the best guess was 50 percent? If the latter, it's easy to get a better credence after the event happened. Look at Nate's track record. An event that he gives an percent chance happens pretty damn close to percent of the time. You could've just as easily said he underestimates the winners when he got it "right" (i.e. he said > 50 percent chance), and therefore Biden has an even higher chance of winning.

Comment by fccc on Industrial literacy · 2020-10-19T04:11:29.024Z · LW · GW

I do think that an infant is not worth much except their sentimental value to their family.

All meaning is in our heads. That doesn't make that meaning any less real. If someone places a lot of meaning on their infant dying, then the infant had a lot of value. If you want to put a dollar value on it, then you can ask the family how much they would pay to bring their child back to life. I would expect most people would pay a lot.

Comment by fccc on Is Stupidity Expanding? Some Hypotheses. · 2020-10-17T02:06:52.276Z · LW · GW

Bad neural nets worked okay under the training set. With a distributional shift, you could see the weaknesses of their models.

I think most beliefs are the mode of what a person hears. You ask me whether someone is for or against abortion, I'll ask you what their parents and friends believe, then I'll bet on the most common belief within that group. So when "the Earth is flat" enters the conversation, and people's reason for believing the Earth isn't flat is basically "It's the only statement I've heard on the topic", they might not have a robust way to determine what is true. Most people can't state necessary conditions for evolution in an arbitrary system. I'd wager most people who believe in evolution can't explain why monkeys still exist.

So when the rug of apparent consensus is pulled out from under the feet of everyone, quite a few will fall over.

Comment by fccc on Industrial literacy · 2020-10-08T00:33:41.788Z · LW · GW

This is almost entirely driven by decreases in infant mortality. The article specifically cites the scenario of a mother giving birth while still carrying their last would probably have abandoned that child. Life expectancy for those that reached adulthood was nearly 70, roughly the same as world average now. 

I have wondered how that factored into life expectancy. This is a good point.

The quote you snip says that the rich in agricultural societies live better than the underclass in those same societies, not better than hunter-gatherer societies. 

Incorrect, that quote is ambiguous about whether they are better off compared to pre-agriculture. However, he also says

Thus with the advent of agriculture the elite became better off, but most people became worse off.

Which is important to match with his classing of most of the U.S. as "elite". He's explicitly saying that if you live in the U.S. today, you are probably better off. That's why I said his argument rests on the poorer countries staying poor.

How are you defining "poor" and why is it bad? How can one argue that people who only need to work ~15 hours per week are "poor". That is far richer than most the world today. The absence of gold or iPhones says nothing about the human condition.

Poverty is obviously a continuum and relative to the context. But my definition is that poorer people have fewer choices, including what goods they can attain, and including how many hours they work. You can live without all the technology and entertainment today if you want. For that life, 15 hours of work per week can be enough if you have a spouse that does the same. (Minimum wage in Australia is enough for that.) If you're a medium-income earner, you can work half of that. Though, admittedly, if you are a middle-income earner probably can't find a job that lets you work that much. But you can retire earlier having done less "total lifetime work". I imagine pre-agricultural people work well into old age.

The anthropological evidence (mostly observation of present day hunter-gatherer groups) indicates that hunter-gatherer groups have high levels of gender equality.

That's surprising to me, and shifts me towards that conclusion.

This is a different question entirely, evaluating the world as a whole instead of the average individual experience. It's quite possible that the increase in quantity of life that has arisen is or will become "worth it".

It's not the original question, but is it relevant: Assuming that people were better off back then, what should we do about it today? The answer: nothing.

You've changed my view quite a bit, but I'd still easily prefer to live now (albeit in a rich country).

Comment by fccc on Industrial literacy · 2020-10-07T12:08:21.489Z · LW · GW

"Life expectancy at birth in the pre-agricultural community was about twenty-six years," says Armelagos, "but in the post-agricultural community it was nineteen years...

Neither of these numbers sound great. Living past 80 sounds a lot better to me. Why did pre-agricultural communities have early deaths compared to us if "the ills that you highlight all came about following the establishment of agricultural societies"? They had to die somehow. 

Early farmers had health issues because they had a handful of crops and just ate those things. If you just eat corn and potatoes, you'll die early. They didn't have nutritional science. To say poor nutrition is a fundamental problem with farming is just incorrect. So I'll concede agriculture made the average people worse off temporarily, but considering our life expectancy increases, I don't see how you can say that the last few decades are still worse than hunter-gathers. In fact, he says as much:

Besides malnutrition, starvation, and epidemic diseases, farming helped bring another curse upon humanity: deep class divisions. Hunter-gatherers have little or no stored food, and no concentrated food sources, like an orchard or a herd of cows: they live off the wild plants and animals they obtain each day. 

... [In agricultural societies, elites were healthier]

To people in rich countries like the U.S., it sounds ridiculous to extol the virtues of hunting and gathering. But Americans are an elite...

So people in rich countries are better off. Then the question becomes "Will the poor countries stay poor?" If they don't, his whole argument is wrong. (Also the "Everyone's poor, so there's no inequality! Hurray!" argument is a bit strange.) I'll bet that before China's wealth increase, he would have said China would stay poor.

Women in agricultural societies were sometimes made beasts of burden. In New Guinea farming communities today I often see women staggering under loads of vegetables and firewood while the men walk empty-handed.

Why is he assuming that had those same people stayed hunter-gathers, they would treat their women better? It seems like a completely unwarranted assumption.

Some epidemic diseases, I'll concede, have been brought to us by farming, indirectly through increased population and population density, and directly through the farming of animals.

You're also neglecting the massive population increases that he discusses. An extra life worth living is a net gain. The associated decreases in average wellbeing haven't held up because of better nutritional science and healthcare so there's not even a "repugnant conclusion" trade-off.

Comment by fccc on Industrial literacy · 2020-10-07T03:27:43.835Z · LW · GW

I agree that there are trade-offs between time periods but, for me, those trade-offs favour the present.

I did mention addictive innovations as negatives, but they can be handled. For example, I have to type in a long password every time before I can watch a Youtube video, which prevents me from mindlessly entering the website.

As for you not asking people for directions, you're also talking to me (in text form, admittedly) probably from the other side of the world. And since we're both on this website, we probably have a lot more in common than we would with a random person off the street.

Any advantage gained by technology is given back in pursuit of more. Faster transport leads to people being more spread out, not better connected.

Do people want faster transport? Yes. People then spread out because it's a trade-off they want to make. It means they can retire earlier because they pay lower rental prices, and that's more important than being a neighbour to as many friends as possible. By transitivity, this is an overall increase in wellbeing. It's three steps forward and one step back, not one step forward and two steps back.

People can set bad goals that don't make them happy when they achieve them. I think that's what causes some people to want more and more. Because they are rarely actually satisfied. I suspect these people would have the same problem in earlier time periods ("Honey, the neighbours have a bigger grass hut than we do"), except they'd have a greater chance of dying from an infection. If you really want to, and you don't think you're permanently ruined modern technology, you can move to a less-industrialized society. They're still around today. I'd note that the flow of immigration is away from those countries and towards more technologically advanced countries. 

People of the past had terrible lives. Correct me if I'm wrong, but racial slavery was more common, witch trials were more common, war killed a greater proportion of the population, more people died from starvation or poor hygiene, religious and homosexual persecution was more common, child brides were more common, women were treated as second-class citizens (if they were citizens at all). To make a convincing case, you'd have to pick a specific time and place to live, show that it's representative of "sometime between the end of the ice age and the first agricultural society", and show, at the very least, that's it's not terrible time to live. Without that, you may be idealizing what it was actually like.

Comment by fccc on Industrial literacy · 2020-10-06T04:35:26.154Z · LW · GW

Yes, I think you're right.

Comment by fccc on Industrial literacy · 2020-10-05T02:01:43.937Z · LW · GW

I'm skeptical of that reasoning because it suggests that the happiest man in the world would be born without limbs and sight and hearing.

I do think that people buy a new car and it's the best thing ever, and then it just becomes their car. This happens to some people more than others, and it happens for some goods more than others. For instance, computers with internet access generate novel experiences all the time, which (for me at least) boosts you above baseline perpetually. I think it's a mistake to forego good experiences to make your experience better.

Comment by fccc on Industrial literacy · 2020-10-04T14:23:49.080Z · LW · GW

"I wish I hadn't ever been born and instead the past had one more person who otherwise wouldn't have lived at that time".

That is indeed what they meant: "I wouldn't give them up, but if I'd never owned a computer, or had a car, or ... I wouldn't miss them". I do find this reasoning a bit strange. You wouldn't miss them, but you wouldn't have them either. If someone invented one and showed it to you, you'd think it was amazing. But then you wish they never showed it to you? (Addictive things are an exception.) 

And that's just luxuries. What about "necessities"? If modern medicine didn't exist and one of your kids died when they were young, you wouldn't miss them? If your wife died during childbirth, you wouldn't miss her?

Maybe it's uncharitable, but I think of them as saying "I would be worse off, but I wouldn't know it, so I'd be better off". It just seems like a perverse belief about human wellbeing to me.

Comment by fccc on Industrial literacy · 2020-10-03T14:37:55.941Z · LW · GW

I would rather have a prison system that focuses on reducing recivism then one that focused on economically exploiting the prison publication for maximum value creation.

I wouldn't be supporting something if it exploited inmates. It has nothing to do with exploiting the prison population; it's about how prisons should be funded. Are you proposing we reduce recidivism to zero by executing every inmate? I assume not. You're probably talking about rehabilitation.

We should not rehabilitate as much as possible. You shouldn't prevent the theft of a candy bar for a billion dollars. So at what level of funding do you propose we provide for rehabilitation? If it's when the benefits outweigh the cost (which is the only level that actually makes sense) then the linked prison proposal does exactly that. You can mathematically show that it does.

If you want to increase inmate wellbeing, by how much? If releasing them all maximized their wellbeing, we still wouldn't release them all. So there's an ideal amount of inmate wellbeing and it's not to maximize it. How can you figure out that value? How can you formally verify that your prison system achieves that value? You can't. So you have to pick a goal that's correlated with inmate wellbeing:

better inmate wellbeing is a consequence of the policy for several reasons: (1) Societal contribution includes the crimes that occur within the prison, so prisons want to prevent their inmates from committing crimes against each other, and (2) societal contribution is probably maximized when former inmates integrate into society, so prisons want to educate inmates and keep them out of harm’s way.

The linked prison system should produce better employment outcomes, fewer crimes, less welfare dependency, and better treatment of inmates. If you can design a prison system that produces a better set of consequences, excellent. Let's see it.

Comment by fccc on "Zero Sum" is a misnomer. · 2020-10-03T14:15:12.520Z · LW · GW

I think the more self-descriptive the terminology is, the better. Fewer syllables is better too. 

  • Pareto-frontier games
  • My-gain-your-loss games
  • Inverse-reward games

"Completely adversarial" sounds a bit too much like "negative sum" games.

Comment by fccc on Industrial literacy · 2020-10-03T13:16:40.559Z · LW · GW

That list seems right. I've had discussions with people that the last few decades have been better than any other time and they disagree. And yet, when I asked them whether they would live in the past or now, they said "now". Go figure. I definitely want people to understand their world a bit better. As you say, educating people would reduce the chance of mobs overthrowing good systems. The other thing to notice is that good systems tend to reduce government influence, and thus the influence of uninformed voters. There's large parts of society where it's not one-person one-vote (e.g. the free market, what you do on your own property, what clothes you wear). We should aim to create more decentralized systems that satisfy good goals (like Pareto efficiency), for instance a prison system that maximizes the total societal contribution of any given set of inmates within the limits of the law or a retirement savings system that maximizes the long-run returns of retirement savings.

Comment by fccc on Moral public goods · 2020-10-03T10:18:28.254Z · LW · GW

I think you're right that redistribution can make sense. Pretty weird that people will still argue against any notion of government-facilitated redistribution: We can explicitly write out hypothetical utility functions and starting bundles and show that redistribution can be preferred by everyone. Even if a person thinks that no one should be coerced, they shouldn't be against a mechanism that redistributes if and only if everyone benefits, even if they think that status quo is currently sufficient for Pareto optimality (since this fact might not hold in the future, as your example shows).

I think this argument supports levels of redistribution like 50% (or 30% or 70% or whatever)...

It favours redistribution, but it doesn't give you the mechanism by which it should occur. I'm not sure what the numbers you're referring to are representing (i.e. 50%, 30%, 70%). Let's say it's the top-level tax rate. Then, ideally, you should decide which exact number and show that it's better than any of the alternatives. I don't think this is possible. It's better to pick a mechanism that picks your numbers for you:

For example, rather than select a price for rice out of thin air, you could let a Pareto-efficient market determine that number (if you consider the goal of Pareto efficiency to be aligned). Note that markets have different prices for the exact same good at the same time (due to things like transportation costs), so having a goal for a universal rice price is the wrong approach from the outset. If a single number is the right approach, arbitrary numeric goals are not aligned when the goal’s specified range does not contain the ideal value. And even if you pick the right number upon the system’s implementation, the ideal value might move out of the range at a later time. 

There are other redistributive mechanisms. You mention quadratic funding. Another example, if I donate a dollar, everyone else in my tax bracket has to pay the same proportion of income divided by 1000. I think these are both weak proposals because they don't satisfy persuasive goals, but I'm just saying that there's lots of different ways to redistribute. Once we think that there should be a redistributive mechanism, the problem is figuring out which possible mechanism has the strongest supporting argument.

So what goals should the mechanism satisfy? Pareto optimality probably. Non-coercion possibly (though you say "if the most selfish people lose out a little bit, [it's not so bad]", and I lean that way too). Note that Pareto optimality can be achieved even if not all steps are Pareto improvements, i.e. you can have a coercive Pareto-optimal solution. (It might also be that Pareto optimality and non-coercion are incompatible with each other.) These goals probably aren't sufficient to uniquely specify our desired system, so I think there should be other goals, otherwise your choice between the several systems that satisfy the goalset is an arbitrary one (i.e. not well justified). But I'm not sure what those other goals should be.

Consequentialism is a really bad model for most people’s altruistic behavior

Sure, but consequentialism is a class of moral theories, not an empirical model of human behaviour. A lot of people try to come up with moral theories that justify their immoral choices (such as saving a family member instead of 100 strangers). It's not hypocritical to say "Total Consequentialism is the correct moral theory" and then not behave like a perfect consequentialist, unless you also claim to be a perfect consequentialist or you criticise others for not being perfect consequentialists.