AI Language Progress

By Robin Hanson · April 6, 2022 1:00 pm · Comments (0)

Brains first evolved to do concrete mental tasks, like chasing prey. Then language evolved, to let brains think together, such as on how to chase prey together. Words are how we share thoughts.

So we think a bit, say some words, they think a bit, they say some words, and so on. Each time we hear some words we update our mental model on their thoughts, which also updates us about the larger world. Then we think some more, drawing more conclusions about the world, and seek words that, when said, help them to draw similar conclusions. Along the way, mostly as a matter of habit, we judge each other’s ability to think and talk. Sometimes we explicit ask questions, or assign small tasks, which we expect to be especially diagnostic of relevant abilities in some area.

The degree to which such small task performance is diagnostic of abilities re the more human fundamental task of thinking together varies a lot. It depends, in part, on how much people are rewarded merely for passing those tests, and how much time and effort they can focus on learning to pass tests. We teachers are quite familiar with such “teaching to the test”, and it is often a big problem. There are many topics that we don’t teach much because we see that we just don’t have good small test tasks. And arguably schools actually fail most of the time; they arguably pretend to teach many things but mostly just rank students on general abilities to learn to pass tests, and inclinations to do what they are told. Abilities which can predict job performance.

Which brings us to the topic of recent progress in machine learning. Google just announced its PaLM system, which fit 540 billion parameters to a “high-quality corpus of 780 billion tokens that represent a wide range of natural language use cases”, in order to predict from past words the next words appropriate for a wide range of small language tasks. Its performance is impressive; it does well compared to humans on a wide range of such tasks. And yet it still basically “babbles“; it seems not remotely up to the task of thinking together with a human. If you talked with it for a long time, you might well find ways that it could help you. But still, it wouldn’t think with you.

Maybe this problem will be solved by just adding more parameters and data. But I doubt it. I expect that a bigger problem is that such systems have been training at these small language tasks, instead of at the more fundamental task of thinking together. Yes, most of the language data on which they are built is from conversations where humans were thinking together. So they can learn well to say the next small thing in such a conversation. But they seem to be failing to infer the deeper structures that support shared thinking among humans.

It might help to assign such a system the task of “useful inner monologue”. That is, it would start talking to itself, and keep talking indefinitely, continually updating its representations from the data of its internal monologue. The trick would be to generate these monologues and do this update so that the resulting system got better at doing other useful tasks. (I don’t know how to arrange this.) While versions of this approach have been tried before, the fact that this isn’t the usual approach suggests that it doesn’t now produce gains as fast, at least for doing these small language tasks. Even so, if those are misleading metrics, this approach might help more to get real progress at artificial thinking.

I will sit up and take notice when the main improvements to systems with impressive broad language abilities come from such inner monologues, or from thinking together on other useful tasks. That will look more like systems that have learned how to think. And when such abilities work across a wide scope of topics, that will look to me more like the proverbial “artificial general intelligence”. But I still don’t expect to see that for a long time. We see progress, but the road ahead is still quite long.

GD Star Rating
loading...
Tagged as:

Great Filter With Set-Backs, Dead-Ends

By Robin Hanson · April 1, 2022 10:40 am · Comments (0)

A biological cell becomes cancerous if a certain set of rare mutations all happen in that same cell before its organism dies. This is quite unlikely to happen in any one cell, but a large organism has enough cells to create a substantial chance of cancer appearing somewhere in it before it dies. If the chances of mutations are independent across time, then the durations between the timing of mutations should be roughly equal, and the chance of cancer in an organism rises as a power law in time, with the power equal to the number of required mutations, usually around six.

A similar process may describe how an advanced civilization like ours arises from a once lifeless planet. Life may need to advance through a number of “hard step” transitions, each of which has a very low chance per unit time of happening. Like evolving photosynthesis or sexual reproduction. But even if the chance of advanced life appearing on any one planet before it becomes inhabitable is quite low, there can be enough planets in the universe to make the chance of life appearing somewhere high.

As with cancer, we can predict that on a planet lucky enough to birth advanced life, the time durations between its step transitions should be roughly equal, and the overall chance of success should rise with time as the power of the number of steps. Looking at the history of life on Earth, many observers have estimated that we went through roughly six (range ~3-12) hard steps.

In our grabby aliens analysis, we say that a power of this magnitude suggests that Earth life has arrived very early in the history of the universe, compared to when it would arrive if the universe would wait empty for it to arrive. Which suggests that grabby aliens are out there, have now filled roughly half the universe, and will soon fill all of it, creating a deadline soon that explains why we are so early. And this power lets us estimate how soon we would meet them: in roughly a billion years.

According to this simple model, the short durations of the periods associated with the first appearance of life, and with the last half billion years of complex life, suggest that at most one hard step was associated with each of these periods. (The steady progress over the last half billion years also suggests this, though our paper describes a “multi-step” process by which the equivalent of many hard steps might be associated with somewhat steady progress.)

In an excellent new paper in the Proceedings of the Royal Society, “Catastrophe risk can accelerate unlikely evolutionary transitions”, Andrew Snyder-Beattie and Michael Bonsall extend this standard model to include set-backs and dead-ends.

Here, we generalize the [standard] model and explore this hypothesis by including catastrophes that can ‘undo’ an evolutionary transition. Introducing catastrophes or evolutionary dead ends can create situations in which critical steps occur rapidly or in clusters, suggesting that past estimates of the number of critical steps could be underestimated. (more)

Their analysis looks solid to me. They consider scenarios where, relative to the transition rate at which a hard step would be achieved, there is a higher rate of a planet “undoing” its last hard step, or of that planet instead switching to a stable “stuck” state from which no further transitions are possible. In this case, advanced life is achieved mainly in scenarios where the hard steps that are vulnerable to these problems are achieved in a shorter time than it takes to undo or stuck them.

As a result, the hard steps which are vulnerable to these set-back or dead-end problems tend to happen together much faster than would other sorts of hard steps. So if life on early Earth was especially fragile amid especially frequent large asteroid impacts, many hard steps might have been achieved then in a short period. And if in the last half billion years advanced life has been especially fragile and vulnerable to astronomical disasters, there might have been more hard steps within that period as well.

Their paper only looks at the durations between steps, and doesn’t ask if these model modifications change the overall power law formula for the chance of success as a function of time. But my math intuition is telling me it feels pretty sure that the power law dependence will remain, where the power now goes as the number of all these steps, including the ones that happen fast. Thus as these scenarios introduce more hard steps into Earth history, the overall power law dependence of our grabby aliens model should remain but become associated with a higher power. Maybe more like twelve instead of six.

With a higher power, we will meet grabby aliens sooner, and each such civilization will control fewer (but still many) galaxies. Many graphs showing how our predictions vary with this power parameter can be found in our grabby aliens paper.

GD Star Rating
loading...
Tagged as: ,

We Trust The Statusful

By Robin Hanson · March 31, 2022 12:35 pm · Comments (0)

The ancient world heavily favored aristocrats for important social positions. Yes, they were probably on average more competent in many ways, but many have claimed that good family connections were favored more strongly than can be explained by this effect.

You might argue that aristocrats were just conspiring to favor themselves, but it seems that others also shared their preference. For example, many stories (e.g.) describe someone seeking an audience with a high official, out of a belief that higher officials must be less corrupt and more concerned with general welfare, relative to the lower ones. “If only the king would hear, he’d do something.”

It seems to me that humans have generally trusted higher status people more, and that we still do so today. For example, status is the main way we trust doctors, lawyers, CEOs, fund managers, and many other professionals (instead of track records or incentive contracts).

As another example, even though people say that they trust smaller orgs more than big ones, people say they trust those at the top of orgs more than those at the bottom. Yes, top people get more easily spread their messages blaming lower folks for problems, and can more easily repress lower folks who blame them. But it seems most people don’t correct for this, or don’t see it as a big correction.

Furthermore, consider our criminal law system. Whether and how much a possible criminal is accused or punished is influenced by many parties. First police, then prosecutors, then judges, then appellate judges, then executives who can pardon, and finally prison wardens and parole boards who influence prison duration and severity. All of these parties are either politicians, political appointees, or folks appointed directly or indirectly by such appointees, and thus should all have incentives to please such people.

Yet if you ask people who they trust to make choices in this system, they consistently prefer the higher status people. In some recent polls, I’ve found that they trust prosecutors more than police, trust judges more than prosecutors, trust judges more than prison wardens or parole boards, and trust executives whose pardons overrule judges. Furthermore, people only proposed recently to “defund” the lowest status among these groups, and one of the main proposals to make police more trustworthy is to require more years of education, i.e., to raise their status.

Some people say this is because the higher status groups have better incentives. But we have chosen the incentives that each of these parties to be what they are, and we could, if we wanted, give all these other groups the same incentives we now give the most trusted group, judges. The relative status of these groups seems to me a cleaner explanation.

I think we are suffering enormous losses from trusting based on status, instead of using other methods. But it seems very hard to displace our innate trust of the high status enough to get people to consider such alternatives. We don’t seek solutions when we don’t think we have a problem.

Added 6Apr: I failed above to mention that we are reluctant to ask for track records or incentives from our statusful suppliers in part because we see that as weakening the connection by which their status raises our status. We want it to look like we mutually trust each other, as that’s the kind of relation by which associates gain status from each other.

GD Star Rating
loading...
Tagged as: ,

Sensible Antitrust

By Robin Hanson · March 23, 2022 8:35 pm · Comments (0)

Most people distrust for-profit firms, relative to individuals and other kinds of orgs. And they distrust such firms more when they get larger, and serve a larger fraction of their industries. This fear expresses itself in criticism, suspicion, and support for politicians who increase the regulations and taxes on such orgs.

This fear is often justified by pointing to potential harms that might result from selfish firms exercising their market power, i.e., their ability to influence prices. (Somehow they can’t believe government or charities would do similarly.) Most people are thus especially eager to see “antitrust” regulation, which purports to directly focus on and address this issue.

Thus most people are surprised when scholars like me (and others) tell them that, historically, regulation with that label seems to been worse than useless; it has done net harm. While simple theory says that harmful market power would be used to raise prices and restrict supply, antitrust regulation has often punished firms for lowering prices and increasing supply.

But a valid question remains: if theory suggests that market power can be harmful, why don’t we see beneficial regulation to limit such harm? Is government so incompetent that it can’t even on average help to limit the harms from market power?

My main point in this post is just to say: actually we do have a lot of regulation that usefully limits market power. It just doesn’t go under the name “antitrust regulation”. It instead goes under the name “utilities regulation”.

We quite commonly have regulation that controls to varying degrees the organizations that supply water, sewer, gas, electricity, phone, cable TV, internet, roads, mail, and other common city services. This is because of a strong widespread consensus that it wouldn’t go well to allow most such organizations full freedom to choose the prices that they charge for such services. It is often prohibitively expensive to build infrastructure to allow multiple competing suppliers of such services, and a single supplier could charge crazy high prices.

Now it might in fact be feasible and even preferable to have such services be supplied by fully private organizations constrained by long term contracts with their customers. But most everywhere instead solves such problems with utilities regulation. Suggesting a widespread consensus that there is in fact a real problem with market power in these cases, and that regulation is usually a reasonable solution.

So the lesson to learn here is not that governments can’t effectively regulate market power. The lesson is that market power has to get pretty extreme before that approach works. Regulating roads and sewer services, that can make sense. But regulating suppliers of oil, aluminum, cameras, or computer operating systems, when suppliers face actual competitors offering such products to customers, doesn’t make sense.

Yes there is often substantial market power there, and thus some potential for harm, and the logical possibility that smart well-intentioned regulators could improve on unregulated behavior. But given typical abilities of firms to capture regulators, and use their powers as a club to limit competition, it doesn’t actually make sense. The state of the art is: either turn an industry into a regulated utility, or just leave it alone.

GD Star Rating
loading...
Tagged as:

Friends, Factions, And Status

By Robin Hanson · March 16, 2022 12:30 pm · Comments (0)

We have three different kinds of reasons to favor others: friends, factions, and status. First, we favor those to whom we have relatively direct connections: friends, family, co-workers, neighbors, sport teammates, church-mates, etc. Second, we favor other members of our self-promoting groups, where each such group makes up a substantial fraction of our relevant society. Such as groups who share gender, ethnicity, education level, or religion. Third, we favor people we see as having prestige or dominance, and are thus accepted by most everyone as being better than others.

Over time, groups can move between these roles. For example, Christianity started out as small groups who favored each other due to their direct connections. Then it became a repressed minority faction within the Roman Empire, then a large faction, and then the largest dominant fraction, where it repressed others. Over the following centuries Christianity became so dominant, at least in areas near Europe, that it became part of the local concept of status; Christians were higher status than heathens or atheists, and status within Christianity counted as social status more generally. Lately, Christianity has retreated to become more of a strong faction.

Similarly, science started out as small community, then became a minor faction, but by now has become entrenched and dominant in our society. That is, most everyone accepts that being more scientific is better overall for your status. Anti-science factions are repressed factions, when they are allowed at all. For example, we fight over if economics is “science” as a way to fight over it s status.

Another example is prestigious college degrees. Once these were an optional and contested status marker, but now they are almost universally seen as a powerful marker, quite often required for any elite position.

Over the last few centuries, “liberation” movements have succeeded in moving some group differences in the other direction, from status to faction. For example, regarding aristocrats, ethnicities, gender, and sexual preferences. Previously these were all widely accepted as status markers, with some types better than others, but recently they have come more to be seen as just more ways that people can be different. Some now try to induce similar changes in how we treat looks, age, and urban/rural, but they have so far largely failed.

Humans have long worried that allegiances to friends or factions would interfere with loyalty to other shared units. So regarding friends and family favoritism, firms limit nepotism, judges recuse themselves when they have personal connections, and teachers aren’t supposed to grade their friends, family, or lovers.

Regarding faction favoritism, humans have long had neutrality norms to limit such favoritism. As a result, faction favoritism is a common “hidden motive” to which we are often consciously blind, like other hidden motives we discuss in our book Elephant in the Brain.

For example, you are supposed to set aside faction allegiances when you rule as a judge, or grade as a teacher, or review a journal article. As a firm representative or responsible professional, you aren’t supposed to give better or cheaper service to customers who share your factions. In a firm, you are supposed to argue for choices of personnel, projects, or policies based on what is good for the firm as a whole, not for your within-firm factions.

Some legal rules of evidence, and common norms of good debate, can be seen as trying limit factional support. For example, rules against hearsay can discourage factions from getting their members to lie for each other, and rules against revealing prior criminal convictions to jurors may prevent conviction-correlated faction action. And debate norms that discourage group-based insults or appeals to authority or popularity can also discourage factional coordination.

In most ancient societies, there was usually a single dominant power coalition, and this still holds for most democratic cities today. The founders of US democracy worried a lot about democratic polities plagued by factions. They thought they had a fix (many small ones), but they were wrong; in larger democratic polities, we’ve often seen a consistent split into two main roughly-equally-strong political factions.

Which subgroups comprise these two factions changes during rare “political realignments.” And in such polities, people like judges, teachers, or generals who represent the polity as a whole are usually expected to set aside their affiliations with these two political factions when making key decisions.

In such polities, it is more okay to say that you favor particular political candidates because they favor your factions. But even there we prefer factions to say that they are trying to do good for the polity as a whole. Valid factions, we say, merely hold different opinions on what is good for us all.

Yes, we are aware that factions may hypocritically pretend to serve the general good while actually serving themselves. We are also aware that high status groups like the top-college-educated may in part just be self-favoring factions so powerful that they suppress any substantial opposition. Even so, we prefer hypocrisy to more overt factional self-dealing, and we prefer to suppress the factions that we can suppress rather than to suppress none at all.

In the U.S. since WWII, we have seen increasing and now record “polarization” between the main two political factions. And recently we’ve seen the relatively new phenomena of “reverse discrimination”, whereby some groups which were first seen as low in status, and then second were supposed to be treated neutrally as just another faction, are now third to be treated in key selection processes as if they were actually higher in status, to correct for their past lower status. Is this a new political realignment?

I actually think it is even bigger; we are now seeing a political faction make a bid for promotion, from faction to status. They want most everyone to accept that their side is just morally better. This faction has come to dominate the most prestigious high ends of many social high grounds: tech, academia, media, civil service, finance, charity, and more. And this faction has induced those arenas to set aside their usual professional neutrality norms regarding political factions, to strongly favor this faction.

In wider public discourse, this faction has used direct control over media and tech, and also the threat of punishment in other arenas it controls, to silence disliked parties and points of view. The opposing faction now mainly retains control of roughly half of top political positions, of peripheral locations of all sorts, and of professions like construction, engineering, and the military wherein they still proudly apply professional norms of factional neutrality. Which looks to me like a weak and losing position.

The main thing I can see that could derail this fast strong train is factional infighting. Such as between those focused on anti-racism, anti-sexism, anti-capitalism, and pro-immigration. There are plausible cleavages there, but I don’t see them opening much anytime soon; this alliance really does show extreme ideologue/religion/etc levels of passion and bonding. Which has long been the origin of new empires in history.

GD Star Rating
loading...
Tagged as: , ,

Unblinding Our Admin Futures

By Robin Hanson · March 7, 2022 9:35 pm · Comments (0)

Our job as futurists is to forecast the future. Not exactly of course, but at least to cut the uncertainty. And one of the simplest way to do that is to take relatively stable and robust past long term trends and project them into the future. Especially if those trends still have a long way that they could continue before they hit fundamental limits. For example, futurists have tried to apply this method to increasing incomes, leisure, variety, density, non-violence, automation, and ease of communication and transport.

It seems to me that one especially promising candidate for this method is also plausibly the fundamental cause of the industrial revolution: bureaucracy. For centuries we humans have been slowly learning how to manage larger more complex networks and organizations, via more formal roles, rules, and processes. (That is, we have more “admin”.) As a result, our orgs have been getting larger and have wider scope, governments have been doing more, and government functions have moved up to larger scale units (cities to states to nations, etc.).

For example, a twitter poll just found respondents saying 1o-1 that the org they know best has been getting more, as opposed to less, bureaucratic over the last decade. And our laws have been getting consistently more complex.

If formal roles, rules, and processes increase over the next century as much as they have over the last century, that should make our future quite different from today. But how exactly? Yes, we’ll use computers more in admin, but that still leaves a lot unsaid. You might think science fiction would be all over this, describing our more admin future in great detail. Yet in fact, science fiction rarely describes much bureaucracy.

In fact, neither does fantasy, the other genre closest to science fiction. Actually, most stories avoid org complexity. For example, most movies and TV shows focus on leisure, instead of work. And when bureaucracy is included, it is usually as a soul-crushing or arbitrary-obstacle villain. It seems that we’d rather look away than acknowledge bureaucracy as a key source of our wealth and value, a pillar and engine of our civilization.

To try to see past this admin blindspot, let us try to find an area of life that today has relatively few formal rules and procedures, and then imagine adding a lot more of them there. This doesn’t necessarily mean that this area of life becomes more restricted and limited compared to today. But it does mean that whatever processes and restrictions there are become more formal and complex.

Public conversation comes to my mind as a potential example here. The rise of social media has created a whole lot more of it, and over the last few years many (including me) have been criticized for saying things the wrong way in public. The claim is often made that it is not the content of what they said that was the problem, it was the way that they said it. So many people say that we accept many complex rules of public conversation that are often being violated.

Thus I’m inclined to imagine a future where we have a lot more formal rules and processes regarding public conversations. These might not be seen as a limit on free speech, in that they only limit how you can say things, not what you can say. These rules might be complex enough to push us to pay for specialist advisors who help us navigate the new rules. Perhaps automation will make such advisors cheaper. And people of that era might prefer the relatively neutral and fair application of these complex rules to the more opportunistic and partisan ways that informal norms were enforced back in the day.

Now I’m not very confident that this is an area of life where we will get a lot more bureaucracy. But I am confident that there will be many such areas, and that we are so far greatly failing to imagine our more bureaucratic future. So please, I encourage you all to help us imagine what our more admin future may look like.

Added 11a: I’m about to attend an event whose dress code is “resort casual”. Whatever that means. I can imagine such dress rules getting a lot more explicit and complex.

GD Star Rating
loading...
Tagged as: ,

From Norms to Law In War

By Robin Hanson · March 6, 2022 9:20 pm · Comments (0)

Humans have long had norms against “starting” fights. Even so, those who start fights are often able to point to things that the other side did first to “really” start the fight. For example, Putin says that actions by the West forced his hand in Ukraine.

Within a nation, law is usually able to draw a clear enough line to decide who started a fight. But if you think about it, at the global level the world is not very clear on what counts as an “act of war”. Many have said the following are not acts of war: selling weapons, econ sanctions, cyberattacks, troop advisors, and even bombings.

You might agree that we haven’t written down an exact clear rule, yet still feel like that it seems pretty clear in most cases. But due to “automatic norms” this is usually less clear than we think. Me four years ago:

categorization of some of the options as norm violating is supposed to come to us fast, and with little thought or doubt. … we are supposed to be sure of which options to reject, without needing to consult with other people, and without needing to try to frame the choice in multiple ways, to see if the relevant norms are subject to framing effects. We are to presume that framing effects are unimportant, and that everyone agrees on the relevant norms and how they are to be applied. … “ignorance of the norms isn’t plausible; you must have known.” (more more more)

For war, this norm ambiguity is especially unfortunate, as it causes more war. What we should hope for instead is to deal with war less via vague informal norms, and more via formal law.

You might object that without a strong central world government, we can’t really have law; we can only have treaties enforced informally, by a threat of shame in the eyes of a world community. But actually, the world has long seen many different kind of legal systems, including legal systems that are stronger than informal norms, yet using powers short of a strong central government.

For example, in one classic legal system, courts issue rulings without having any further power to enforce their rulings. Their threat is just that if you don’t do what they say, they may officially label you an “outlaw”, after which anyone is free to harm you without fear of legal penalties.

Such a classic legal system could be further strengthened if its subjects were to give it hostages. For example, a financial hub which holds many financial assets of subjects could also serve as a legal system, if those assets held could be forfeit in the case of adverse legal rulings. (And if that hub were run by a distinct non-partisan community proud to serve in its legal role, and sufficiently able protect itself from outside attack.)

Thus it seems possible for the world to have a legal system wherein a widely-used sufficiently-defended financial hub agreed to enforce treaties between the nations whose assets were held there. Then if a nation violated its treaty, and refused to abide by this court’s ruling against it, then this law could declare that nation an outlaw, and grab its held assets.

How is that different from a large world alliance spontaneously agreeing to treat a nation as an outlaw and then grab whatever assets they can? It would be the difference between norms and contract law. An alliance might not be fair. It might instead not protect a once-ally if that were inconvenient, or it might opportunistically use its power to unfairly dump on a nation if that happened to be convenient. In contrast, with a more formal law, we might have more (though hardly infinite) hope for principled consistent non-partisan rulings.

The world doesn’t need to have a strong world government to have a functioning contract law between nations. Treaties can be more than expressions of hope.

GD Star Rating
loading...
Tagged as: , ,

Make Law Like Couches, Not Cars

By Robin Hanson · March 3, 2022 8:20 pm · Comments (0)

Action movies often show fights in complex environments like factories, ships, kitchens, warehouses, or construction sites. In such cases, whomever knows more details of that environment can have big advantages in the fight. However, when people instead try to arrange a “fair fight”, they usually choose simple environments that combatants can know similarly well, like an empty flat walled square or circle.

Lawyers often fill their offices with big shelves full of law books. As if to say “Law is a vast complex machine you wouldn’t want to mess with without access to an engineer like me who knows all its details.” Or more relevant, “The arena of law is as complex a place for a fight as is a factory or kitchen. You don’t want to fight there without a warrior like me who knows all those complex details.”

However, the essence of law is for a judge to hear A complain about B, and then to issue a ruling to reward or punish A or B. And the main point of such law is to induce better behavior via shared expectations of such rulings. For that purpose, what matters about the law is those shared prior expectations; further legal detail beyond that has little social value.

Actually, added legal complexity and detail can hurt, by tempting people to learn more legal details in order to gain strategic advantages. Just as warriors fighting in a kitchen would need to learn kitchen details, people with possible legal conflicts can need to learn about arbitrary legal details, or to hire lawyers who learn them, even if those details do little to help guide prior actions by A or B.

Imagine that two people will hold a verbal debate in some physical space. Law without needless details is like the debaters sitting on a simple couch to do their debate. Such a couch has little other structure besides that which is needed to coordinate their locations and orientations. Which is good.

Now imagine a couch with lots of little pockets holding weapons or controls to make the couch poke people, change shape, get hot, or make noises. Something like a car. If you were to be in a debate on such a complex couch, you’d want to invest in learning those details. For example, you might be able to poke your opponent out of view at just the right moment. Even though that is a social waste.

Is a minimal couch-like law possible? Consider juries. Imagine there is little formal law, so that juries can rule most any way they choose. In this case legal expectations are just expectations over jury rulings. So if A and B know the community from which jurors are chosen well enough, then they know that they have shared legal expectations. And they know that there’s not much either of them can do to gain more info on that. Their law is a couch, not a car.

Of course it is not enough just to have shared legal expectations; one also wants those expectations to do well at taking into account situation details known to both A and B. Thus one problem with a simple jury system is that random juries many not know important situation details that are known by both A and B. So each pair A and B might prefer that a case between be judged by a jury chosen from a community closer to them, so that this jury knows more of their shared context.

But you also couldn’t pick jurors who are too closely connected to A and B, as these might not be willing to function as independent jurors. So, for example, if A and B are both in the movie industry, it might make sense to give them a jury from the movie industry, who could then understand movie practices. But maybe not jurors who are currently working on the very same movie as they.

150 years ago, the US had something closer to this simple jury system, as stated laws were few and vague, juries made most decisions, lawyers were cheap and less often needed, and plea bargaining wasn’t yet much of a thing. Since then, US law has accumulated far more detail. Yet little of this detail seems to be an adaptation to a more complex world; most is just random. And we must pay lawyers who learn this detail if we hope to win at court.

Worse, regulations greatly restrict who can be a lawyer, slower more expensive legal processes add to our costs, and few of us have sufficient assets to pay if we lose. Thus US law has rotted in a great many ways. When will we notice that, and consider big changes?

By the way, one feature that we might want in a legal system is an ability ask it for prior approval for behavior. “Would it be legal or not-negligent if I did it this way?” And you might hope that a very detailed legal system could at least offer this advantage over a simple jury-based version. You’d just look up the relevant detailed law. But in fact our very complex detailed legal system doesn’t offer this feature. You just can’t ask what acts might be legal; you can only do stuff and find out later if you are punished.

Added 11a: Jury decisions can vary. To reduce the impact of that in particular parties, we could  have the consequences for them be set by prediction markets on jury decisions. Those market predictions would be far more consistent across cases.

GD Star Rating
loading...
Tagged as: ,

Are Firms Today’s Low Class?

By Robin Hanson · March 1, 2022 10:10 am · Comments (0)

In the ancient world, there were different classes of people, who were treated differently. Some had more freedoms than others, such as to own property, make contracts, start lawsuits, speak in public forums, or travel on their own. On the street, some had to stay silent, to wear distinctive clothing, to give way to others, and to show on demand who vouched for them.

Today we are proud that we make fewer such distinctions among individuals. (Though we do distinguish kids and convicts.) But that may be less because of our liberality, and more because of the rise of large organizations, of which the ancient world had far fewer. Today we require many distinctions between individuals, for-profit firms, non-profit orgs, and government agencies.

For example, we ban many kinds of discrimination by firms that we allow by individuals, and we hold firms to higher standards of truthfulness. We require more public disclosures by them. And many want to further limit their abilities to speak in public debates.

We today are horrified to hear that an unwed mother might have once been required to wear a “Scarlet A” in public to display her status, and are similarly horrified by rumors of a Chinese “social credit” system to weigh many indicators of social approval and disapproval. (Though we seem okay with credit scores, relating only to our dealings with firms.)

But we seem fine with legally requiring firms to disclose Environmental, Social, and Governance (ESG) scores, which score firms by adding up dozens of particular “do-gooder” indications. A social credit score for firms. So far laws have not yet required us to shun the firms that rate poorly, though many suspect such laws are coming. But such scores today clearly help mobs to coordinate to punish firms, just as Scarlet As do for individuals.

Government agencies, in contrast, often have more privileges than both firms and individuals. For example, US police have “qualified immunity” protecting them from lawsuits. Government, academic and think tank studies are taken at face value, yet firm versions are considered suspect. Non-profits like universities are allowed to collude in ways that firms cannot.

Thus we today still do maintain formal class ladders in custom and law, but more ladders of orgs than of people. Now you might say this is fine, as class ladders are mean to individuals, making them feel bad, but organizations can’t feel bad. But in fact our class ladders of orgs influence how people in those orgs feel.

As individuals today largely gain status via association with orgs, the status of those orgs directly effects the status of individuals. So when we shame or honor an org, we also shame or honor the people associated with it. Formal class ladders continue today.

GD Star Rating
loading...
Tagged as: ,

Rationality As Rules

By Robin Hanson · February 27, 2022 9:15 pm · Comments (0)

Our world is complex, where we try many things, with varying success. And when we look at the distribution of our successes and failures, we often notice patterns. Pattens we then summarize in terms of models and rules of thumb regarding ideal arrangements.

For example, an ideal sentence has a verb and a subject, starts with a capital letter, and ends in a period. An ideal essay has an intro and a conclusion, and each paragraph has an idea sequence of sentence types. In my old world of lisp, ideal computer code has no global variables, each function has only a few lines of code and is documented with text, and functions form a near tree structure with few distant connections.

Ideal rooms have little dust or dirt, few loose items around, and big items are where they appear in a designed floor plan. Ideal job performance follows ideal work schedules and agreed-on procedures. Ideal communication uses clear precise language and is never false or misleading. And so on.

Such simple and easily expressed rules and ideal descriptions can be quite helpful when we are learning how to do things. But eventually, if we are primarily focused on some other outcomes, we usually find that we want to sometimes deviate from the rules, and produce structures that differ from our simple ideals. Not every best sentence has a verb, not every best code function is documented, and the furniture isn’t always most useful when placed exactly according to the floor plan.

However, when we are inclined to suspect each others’ motives, we often turn rules of thumb into rules of criticism. That is, we turn them into norms or laws or explicit rules of home, work, etc. And at that point such rules discourage us from deviating, even when we expect such deviations to improve on rule-following. Yes, it is sometimes possible to apply for permissions to deviate, or to deviate first and then convince others to accept this. But even so, enforced rules change our behavior.

With sufficient distrust, it can make sense to enforce a limited set of such rules. We can lose on net in getting the very best outcomes when people have good motivations, but we can gain via cutting the harm that can result from poor motivations. At least when on average poor motivations tend to move choices away from our usual ideals.

For example, we humans are complex and twisted enough that sometimes we are better off when we lie to others, and when they lie to us. Even so, we can still want to enforce rule systems that detect and punish lies. Yes, this will discourage some useful lies. But that loss may be more than compensated by also discouraging damaging lies from people with conflicting interests. We can want rules that tend to push behavior toward the zero-lie ideal even when that is not actually the best scenario for us, all things considered.

I recently realized that “rationality” is mostly about such ideal-pushing rules. We say that we are more “rational” when we avoid contradictions, when our arguments follow valid logical structures, when our degrees of belief satisfy probability axioms, and when we update according to Bayes’ rule. But this is not because we can prove that such things are always better.

Oh sure, we can prove such things are better in some ideal contexts, but we also know that our real situations differ. For example, we know that we can often improve our situation via less accurate beliefs that influence how others think of us. (Our book Elephant in the Brain is all about this.) And even accuracy can often be improved via calculations that ignore key relevant information. Our minds and the problems we think about are complex, heuristic, and messy.

Yet if we sufficiently fear being maliciously persuaded by motivated reasoning from others with conflicting motives, we may still think ourselves better off if we support rationality norms, laws, and other rules that punish incoherence, inconsistency, and other deviations from simple rationality ideals.

The main complication comes when people want to argue for us to accept limited deviations from rationality norms on special topics. Such as God, patriotism, romance, or motherly love. It is certainly possible that we are built so as to be better off with certain “irrational” beliefs on such topics. But how exactly can we manage the debates in which we consider such claims?

If we apply the usual debate norms on these topics, they will tend to support the usual rational conclusions on what are the most accurate beliefs, even if they also suggest that other beliefs might be more beneficial. But can people be persuaded to adopt such beneficial beliefs if this is the status of the official public debates on those topics?

However, if we are to reject the usual rationality standards for those debates, what new standards can we adopt instead to protect us from malicious persuasion? Or what other methods can we trust to make such choices, if those methods are not to be disciplined by debates? I’m not yet seeing a way out.

GD Star Rating
loading...
Tagged as: