Tag Archives: Innovation

Radical Signals

Many people tout big outside-the-Overton “radical” proposals for change. They rarely do this apologetically; instead, they often do this in a proud and defiant tone. They seem to say directly that their proposal deserves better than it has gotten, and indirectly that they personally should be admired for their advocacy.

Such advocacy also tends to look a lot like costly signaling. That is, advocates seem to go out of their way to pay costs, such as via protests, meetings, writing redundant boring diatribes, accosting indifferent listeners at parties, implying that others don’t care enough, and so on. But it so, what exactly are they signaling?

If you recall, costly signaling is a process whereby you pay visible costs, but make sure that those costs are actually less when some parameter X is higher. If you get a high enough payoff from persuading audiences that X is high, you are plausibly willing to pay for these costly signals, in order to produce this persuasion. For example, you pay to go to school, but since school is easier if your are smart and conformist, going to school shows those qualities to observers.

Here are six things you might show about a radical proposal:

Investment – It is a good financial investment. You pay costs to initiate or improve a business venture or investment fund that includes variations on this proposal. Doing so is less costly, and even net profitable for you, if this turns out to be a profitable project. By visibly paying costs, you hope to convince others to join your investment.

Popularity – It will eventually become more popular. You lend your time, attention, and credibility to a “movement” in favor of this proposal. This effort on your part may be rewarded with praise, prestige, and attention if this movement becomes a lot more popular and fashionable. You hope that your visible support will convince others to add their support.

Morality – You, and the other supporters of this proposal, are unusually moral. You pick a proposal which, if passed, would impose large costs in the service of a key moral goal. For example, you might proposal a 90% tax on the rich, or no limits on encryption. Others have long been aware of those extreme options, but due to key tradeoffs they preferred less extreme options. You show your commitment to one of the values that are traded off by declaring you are willing to lose big on all the other considerations, if only you can win on yours.

Conformity – You are a loyal member of some unusual group. You show that loyalty by burning your bridges with other groups, via endorsing radical proposals which much put off other groups. This is similar to adopting odd rules on food and dress, or strange religious or ideological beliefs. Once a radical proposal is associated with your group for any reason, you show loyalty to that group by supporting that proposal.

Inventive – You are clever enough to come up with surprising solutions. You take a design problem that has vexed many, and offer a new design proposal that seems unusually simple elegant, and effective. Relative to someone who wanted to show effectiveness, your proposal would be simpler and more elegant, and it would focus on solving the problems that seem most visible and vexing to observers, instead of what are actually the most important problems. It would also tend to use theories that observers believe in, relative to theories that are true.

Effective – If adopted, your proposal would be effective at achieving widely held goals. To show effectiveness, you incur costs to show things that are correlated with effectiveness. For example, you might design, start, or complete related theoretical analyses, fault analyses, lab experiments, or field experiments. You might try to search for problematic scenarios or effects related to your proposal, and search for design variations that could better address them. You might search for plans to do small scale trials that can give clearer cheaper results, and that address some key potential problems.

In principle showing each of these things can also show the others. For example, showing that something is moral might help show its potential to become popular. Still, we can distinguish what an advocate is more directly trying to show, from what showing that would indirectly show.

It seems to me that, among the above options, the most socially valuable form of signaling is effectiveness. If we could induce an equilibrium where people tried to show the other things via trying to show effectiveness, we’d induce a lot more useful effort to figure out what variations are effective, which should help us to find and adopt more and better radical proposals. If we can’t get that, inventiveness seems the second best option.

GD Star Rating
loading...
Tagged as: ,

Automation As Colonization Wave

Our automation data analysis found a few surprising results. We found that labor demand is inversely correlated with education. As if, when facing a labor shortage for a particular kind of worker, employers respond in part by lowering education requirements. And even though more automation directly lowers demand for a job, it seems that labor demand changes, relative to labor supply changes, becomes a smaller factor for jobs where automaton rises more.

But the most interesting surprise, I think, is that while, over the last twenty years, we’ve seen no noticeable change in the factors that predict which jobs get more automated, we have seen job features change to become more suitable to automation. On average jobs have moved by about a third of a standard deviation, relative to the distribution of job automation across jobs. This is actually quite a lot. Why do jobs change this way?

Consider the example of a wave of human colonization moving over a big land area. Instead of all the land becoming colonized more densely at same rate everywhere, what you instead see is new colonization happening much more near old colonization. In the U.S., dense concentrations started in the east and slowly spread to the west. There was little point in clearing land to grow stuff if there weren’t enough other folks nearby to which to sell your crops, and from which to buy supplies.

If you looked at any particular plot of land and asked what factors predict if it will be colonized soon, you might see those factors stay pretty constant over time. But many of those factors would depend on what other land nearby had been colonized recently. In a spatial colonization wave, there can be growth without much change in the underlying tech. Instead, the key dynamic can be that there are big time delays to allow an initial tech potential to become realized via spreading across a large landscape. A colonization wave can be growth without much tech change.

Now think about the space of job tasks as a similar sort of landscape. Two tasks are adjacent to other tasks when the same person tends to do both, when info or objects are passed from one to the other, when they take place close in place and time, and when their details gain from being coordinated. The ease of automating each task depends on how regular and standardized are its inputs, how easy it is to formalize the info on which key choices depend, how easy it is to evaluate and judge outputs, and how simple, stable, and mild are the physical environments in which this task is done.

When the tasks near a particular task get more automated, those tasks tend more to happen in a more controlled stable environment, the relevant info tends to be more formalized, and related info and objects get simpler, more standardized, and more reliably available. And this all tends to make it easier to automate such tasks. Much like how land is easier to colonize when nearby land is more colonized.

Among the job features that predict automation in our analysis, the strongest is: Pace Determined By Speed Of Equipment. This feature clearly fits my story here; it says you coordinate your task closely with a task done by a machine. Many others fit as well; here is more from our paper:

Pace Determined By Speed Of Equipment picks out jobs that coordinate closely with machinery, while Importance of Repeating Same Tasks picks out jobs with many similar and independent small tasks. Variety picks out an opposite case of dissimilar tasks. The job features Wear Common Safety Equipment and Indoors Environmentally Controlled pick out tasks done in calm stable environments, where machines function better, while Hearing Sensitivity picks out less suitable complex subtle environments. In jobs with frequent Letters and Memos, such memos tend to be short and standardized. Jobs with more Advancement are “results oriented”, with more clearly measurable results. Simple machines tend to be bad at Thinking Creatively, Innovation and Mathematics. Physical Proximity picks out jobs done close to humans, usually because of needed human interactions, which tend to be complex, and where active machines could risk hurting them.

We have long been experiencing a wave of automation passing across the space of job tasks. Some of this increase in automation has been due to falling computer tech costs, improving algorithms and tools, etc. But much of it may simply be the general potential of this tech being realized via a slow steady process with a long delay: the automation of tasks near other recently automated tasks, slowly spreading across the landscape of tasks.

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

Automation: So Far, Business As Usual

Since at least 2013, many have claimed that we are entering a big automation revolution, and so should soon expect to see large trend-deviating increases in job automation levels, in related job losses, and in patterns of which jobs are more automated.

For example, in the October 15 Democratic debate between 12 U.S. presidential candidates, 6 of them addressed automation concerns introduced via this moderator’s statement:

According to a recent study, about a quarter of American jobs could be lost to automation in just the next ten years.

Most revolutions do not appear suddenly or fully-formed, but instead grow from precursor trends. Thus we might hope to test this claim of an automation revolution via a broad study of recent automation.

My coauthor Keller Scholl and I have just released such a study. We use data on 1505 expert reports regarding the degree of automation of 832 U.S. job types over the period 1999-2019, and similar reports on 153 other job features, to try to address these questions:

  1. Is automation predicted by two features suggested by basic theory: pay and employment?
  2. Do expert judgements on which particular jobs are vulnerable to future automation predict which jobs were how automated in the recent past?
  3. How well can we predict each job’s recent degree of automation from all available features?
  4. Have the predictors of job automation changed noticeably over the last two decades?
  5. On average, how much have levels of job automation changed in the last two decades?
  6. Do changes in job automation over the last two decades predict changes in pay or employment for those jobs?
  7. Do other features, when interacted with automation, predict changes in pay or employment?

Bottom line: we see no signs of an automation revolution. From our paper‘s conclusion:

We find that both wages and employment predict automation in the direction predicted by simple theory. We also find that expert judgements on which jobs are more vulnerable to future automation predict which jobs have been how automated recently. Controlling for such factors, education does not seem to predict automation.

However, aside perhaps from education, these factors no longer help predict automation when we add (interpolated extensions of) the top 25 O*NET variables, which together predict over half the variance in reported automation. The strongest O*NET predictor is Pace Determined By Speed Of Equipment and most predictors seem understandable in terms of traditional mechanical styles of job automation.

We see no significant change over our time period in the average reported automation levels, or in which factors best predict those levels. However, we can’t exclude the possibility of drifting standards in expert reports; if so, automation may have increased greatly during this period. The main change that we can see is that job factors have become significantly more suitable for automation, by enough to raise automation by roughly one third of a standard deviation.

Changes in pay and employment tend to predict each other, suggesting that labor market changes tend more to be demand instead of supply changes. These changes seem weaker when automation increases. Changes in job automation do not predict changes in pay or employment; the only significant term out of six suggests that employment increases with more automation. Falling labor demand correlates with rising job education levels.

None of these results seem to offer much support for claims that we are in the midst of a trend-deviating revolution in levels of job automation, related job losses, or in the factors that predict job automation. If such a revolution has begun, it has not yet noticeably influenced this sort of data, though continued tracking of such data may later reveal such a revolution. Our results also offer little support for claims that a trend-deviating increase in automation would be accompanied by large net declines in pay or employment. Instead, we estimate that more automation mainly predicts weaker demand, relative to supply, fluctuations in labor markets.

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

Designing Crime Bounties

I’ve been thinking about how to design a bounty system for enforcing criminal law. It is turning out to be a bit more complex than I’d anticipated, so I thought I’d try to open up this design process, by telling you of key design considerations, and inviting your suggestions.

The basic idea is to post bounties, paid to the first hunter to convince a court that a particular party is guilty of a particular crime. In general that bounty might be paid by many parties, including the government, though I have in mind a vouching system, wherein the criminal’s voucher pays a fine, and part of that goes to pay a bounty. 

Here are some key concerns:

  1. There needs to be a budget to pay bounties to hunters.
  2. We don’t want criminals to secretly pay hunters to not prosecute their crimes.
  3. We may not want the chance of catching each crime to depend lots on one hunter’s random ability. 
  4. We want incentives to adapt, i.e., use the most cost-effective hunter for each particular case. 
  5. We want incentives to innovate, i.e., develop more cost-effective ways to hunt over time. 
  6. First hunter allowed to see a crime scene, or do an autopsy, etc., may mess it up for other hunters. 
  7. We may want suspects to have a right against double jeopardy, so they can only be prosecuted once.
  8. Giving many hunters extra rights to penetrate privacy shields may greatly reduce effective privacy.
  9. It may be a waste of time and money for several hunters to simultaneously pursue the same crime. 
  10. Witnesses may chafe at having to be interviewed by several hunters re the same events.

In typical ancient legal systems, a case would start with a victim complaint. The victim, with help from associates, would then pick a hunter, and pay that hunter to find and convict the guilty. The ability to sell the convicted into slavery and to get payment from their families helped with 1, but we no longer allow these, making this system problematic. Which is part of why we’ve added our current system. Victims have incentives to address 2-4, though they might not have sufficient expertise to choose well. Good victim choices give hunters incentive to address 5. The fact that victims picked particular hunters helped with 6-10. 

The usual current solution is to have a centrally-run government organization. Cases start via citizen complaints and employee patrols. Detectives are then assigned mostly at random to particular local cases. If an investigation succeeds enough, the case is given to a random local prosecutor. Using government funds helps with 1, and selecting high quality personnel helps somewhat with 3. Assigning particular people to particular cases helps with 6-10.  Choosing people at random, heavy monitoring, and strong penalties for corruption can help with 2. This system doesn’t do so well on issues 4-5. 

The simplest way to create a bounty system is to just authorize a free-for-all, allowing many hunters to pursue each crime. The competition helps with 2-5, but having many possible hunters per crime hurts on issues 6-10. One way to address this is to make one hunter the primary hunter for each crime, the only one allowed any special access and the only one who can prosecute it. But there needs to be a competition for this role, if we are to deal well with 3-5.

One simple way to have a competition for the role of primary hunter of a crime is an initial auction; the hunter who pays the most gets it. At least this makes sense when a crime is reported by some other party. If a hunter is the one to notice a crime, it may make more sense for that hunter to get that primary role. The primary hunter might then sell that role to some other hunter, at which time they’d transfer the relevant evidence they’ve collected. (Harberger taxes might ease such transfers.)

Profit-driven hunters help deal with 3-5, but problem 2 is big if selling out to the criminal becomes the profit-maximizing strategy. That gets especially tempting when the fine that the criminal pays (or the equivalent punishment) is much more than the bounty that the hunter receives. One obvious solution is to make such payoffs a crime, and to reduce hunter privacy in order to allow other hunters to find and prosecute violations. But is that enough?

Another possible solution is to have the primary hunter role expire after a time limit, if that hunter has not formally prosecuted someone by then. The role could then be re-auctioned. This might need to be paired with penalties for making overly weak prosecutions, such as loser-pays on court costs. And the time delay might make the case much harder to pursue.

I worry enough about issue 2 that I’m still looking for other solutions. One quite different solution is to use decision markets to assign the role of primary hunter for a case. Using decision markets that estimate expected fines recovered would push hunters to accumulate track records showing high fine recovery rates. 

Being paid by criminals to ignore crimes would hurt such track records, and thus such corruption would be discouraged. This approach could rely less on making such payoffs illegal and on reduced hunter privacy. 

The initial hunter assignment could be made via decision markets, and at any later time that primary role might be transferred if a challenger could show a higher expected fine recovery rate, conditional on their becoming primary. It might make sense to require the old hunter to give this new primary hunter access to the evidence they’ve collected so far. 

This is as far as my thoughts have gone at the moment. The available approaches seem okay, and probably better than what we are doing now. But maybe there’s something even better that you can suggest, or that I will think of later. 

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

Dreamtime Social Games

Ten years ago, I posted one of my most popular essays: “This is the Dreamtime.” In it, I argued that, because we are rich,

Our descendants will remember our era as the one where the human capacity to sincerely believe crazy non-adaptive things, and act on those beliefs, was dialed to the max.

Today I want to talk about dreamtime social games.

For at least a million years, our ancestors wandered the Earth in small bands of 20-50 people. These groups were so big that they ran out of food if they stayed in one place, which is why they wandered. But such groups were big and smart enough to spread individual risks well, and to be relative safe from predators.

So in good times at least, the main environment that mattered to our forager ancestors was each other. That is, they succeeded or failed mostly based on winning social games. Those who achieved higher status in their group gained more food, protection, lovers, and kids. And so, while foragers pretended that they were all equal, they actually spent much of their time and energy trying to win such status games. They tried to look impressive, to join respected alliances, to undermine rival alliances, and so on. Usually in the context of grand impractical leisure and play.

As I described recently, status is usually based on a wide range of clues regarding one’s impressiveness, and the relative weight on these clues does vary across cultures. But there are many generic clues that tend to be important in most all cultures, including strength, courage, intelligence, wit, art, loyalty, social support etc.

When an ability was important for survival in a local environment, cultural selection tended to encourage societies to put more weight on that ability in local status ratings, especially when their society felt under threat. So given famine, hunters gain status, given war warriors gain status, and when searching for a new home explorers gain status.

But when the local environment seemed less threatening, humans have tended to revert back to a more standard human social game, focused on less clearly useful abilities. And the more secure a society, and the longer it has felt secure, the more strongly it reverts. So across history the social worlds of comfortable elites have been remarkably similar. In the social worlds such as Versailles, Tales of Genji, or Google today, we see less emphasis on abilities that help win in larger harsher world, or that protect this smaller world from larger worlds, and more emphasis on complex internal politics based on beauty, wit, abstract ideas, artistic tastes, political factions, and who likes who.

That is, as people feel safer, local status metrics and social institutions drift toward emphasizing likability over effectiveness, popularity and impressiveness over useful accomplishment, and art and design over engineering. And as our world has been getting richer and safer for many centuries now, our culture has long been moving toward emphasizing such forager values and attitudes. (Though crises like wars often push us back temporarily.)

“Liberals” tend to have moved further on this path than “conservatives”, as indicated by typical jobs:

jobs that lean conservative … [are] where there are rare big bad things that can go wrong, and you want workers who can help keep them from happening. … Conservatives are more focused on fear of bad things, and protecting against them. … Jobs that lean liberal… [have] small chances that a worker will cause a rare huge success … [or] people who talk well.

Also, “conservative” attitudes toward marriage have focused on raising kids and on a division of labor in production, while “liberal” attitudes have focused on sex, romance, and sharing leisure activities.

Rather than acknowledging that our status priorities change as we feel safer, humans often give lip service to valuing useful outcomes, while actually more valuing the usual social game criteria. So we pretend to go to school to learn useful class material, but we actually gain prestige while learning little that is useful. We pretend that we pick lawyers who win cases, yet don’t bother to publish track records and mainly pick lawyers based on institutional prestige. We pretend we pick doctors to improve health, but also don’t publish track records and mainly pick via institutional prestige, and don’t notice that there’s little correlation between health and medicine. We pretend to invest in hedge funds to gain higher returns, but really gain status via association with impressive fund managers, and pay via lower average returns.

I recently realized that, alas, my desire to move our institutions more toward “paying for results” is at odds with this strong social trend. Our institutions could be much more effective at getting us the things we say we want out of them, but we seem mostly content to let them be run by the usual social status games. We put high status people in change and give them a lot of discretion, as long as they give lip service to our usual practical goals. It feels to most people like a loss in collective status if they let their institutions actually focus too much on results.

A focus on results would probably result in the rise to power of less impressive looking people who manage to get more useful things done. That is what we’ve seen when firms have adopted prediction markets. At first firms hope that such markets may help them identify the best informed employees. But are are disappointed to learn that winners tend not to look socially impressive, but are more nerdy difficult inarticulate contrarians. Not the sort they actually want to promote.

Paying more for results would feel to most people like having to invite less suave and lower class engineers or apartment sups to your swanky parties because they are useful as associates. Or having to switch from dating hip hunky Tinder dudes to reliable practical guys with steady jobs. In status terms, that all feels less like admiring prestige and more like submitting to domination, which is a forager no-no. Paying for results is the sort of thing that poor practical people have to do, not rich prestigious folks like you.

Of course our society is full of social situations where practical people get enough rewards to keep them doing practical things. So that the world actually works. People sometimes try to kill such things, but then they suffer badly and learn to stop. But most folks who express interest in social reforms seem to care more about projecting their grand hopes and ideals, relative to making stuff work better. Strong emotional support for efficiency-driven reform must come from those who have deeply felt the sting of inefficiency. Perhaps regarding crime?

Ordinary human intuitions work well for playing the usual social status games. You can just rely on standard intuitions re who you like and are impressed by, and who you should say what to. In contrast, figuring out how to actually and effectively pay for results is far more complex, and depends more on the details of your world. So good solutions there are unlikely to be well described by simple slogans, and are not optimized for showing off one’s good values. Which, alas, seems another big obstacle to creating better institutions.

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

Status Apps Are Coming

A person’s social status is a consensus of those nearby on that person’s relative social value and power. Which factors count how much in this status vary across societies and subcultures. (They probably vary more at high status levels.) Most people spend a lot of effort in private thought and in conversations trying to infer the status of they and their associates, and trying to raise the status of their allies and to lower that of their rivals.

Typically a great many considerations go into estimating status. Such as the status of your ancestors and current associates, and your job, income, residence, health, beauty, charisma, intelligence, strength, gender, race, and age. Most anything that is impressive or admirable helps, such as achievements in sports and the arts, looking sharp, and seeming knowledgeable. Most anything that is disliked or disapproved hurts, such as often (but not always) applies for violence, rudeness, unreliability, and filth.

Today we generate shared status estimates via expensive gossip and non-verbal communication, but someday (in 20 years?) tech may help us more in this task. Tech will be able to see many related clues like who talks to who with what tone of voice, who looks how at who, who invites who to what social events, who lives where and has what jobs, etc. Given some detailed surveys on who says who has what status, we may build accurate statistical models that predict from all that tech-accessible data who would say who has what status in what contexts.

Or new social practices might create more directly relevant data. Imagine a future app where you can browse people to see numerical current estimates of their status (perhaps relative to a subculture).  You can click up or down on any estimate to indicate that you consider it too low or too high. Some perhaps-complex mechanism then takes prior estimates, background tech data, and these up/down edits to generate changes in these status estimates, and also changes in estimates of edit source reliability. All else equal, people who contribute more reliable/informative status edits are probably estimated to have higher status.

I don’t know how exactly such an algorithm could or should work. But I’m confident that there are many variations that could work well enough to attract much participation and use. Many people would be tempted to use these status estimates similarly to how they now use the status estimates that they generate via gossip and subtle social clues. They might even use them in even more places than they use status today, if these new estimates were considered more reliable and verifiable.

I’m also confident that governments, firms, and other organizations would be eager to influence these systems, as they’d see some variations as being more favorable to their interests. Yes, that creates a risk that they may push for bad variations, though don’t forget that our informal systems today also have many flaws. For example, many people use false rumors and other underhanded status tricks to hurt rivals and help allies, tricks that may be harder to get away with in a more transparent system.

Yes, this may look like a dystopia in many ways. But it is probably coming whether you like it or not, and this change may offer great opportunities to improve our status systems. For example, today we have many anti-discrimination policies that seem to be crude and awkward attempts to fix perceived problems with our current status systems. A more fine-grained, data-driven, and transparent status system might allow more effective and better targeted fixes. So it seems worth thinking now a bit more about how such systems could and should work, before some big government or tech firm imposes a system that quickly gets entrenched and hard to change.

GD Star Rating
loading...
Tagged as: ,

Why Not Hi-Tech Forms?

A half century ago, when people tried to imagine a future full of computers, I’m sure one of the most obvious predictions they made is that we today wouldn’t have to work so hard to fill out forms. Filling out forms seemed then to be a very mechanical task, based on explicit mechanical rules. So once computers had enough space to store the relevant data, and enough computing power to execute those rules, we should not longer need to fill out most tedious parts of forms.

Oh sure, you might need to write an essay for a school application, or make a design for the shed when you ask your homeowner’s association permission to build a shed. But all that other usual tedious detail, no.

Now this has in fact happened for businesses, at least for standard forms and for big business. In fact, this happened many decades ago. Most of them wrote or bought programs to fill out standard forms that they use to talk to customers, to suppliers, and to government. But for ordinary people, this mostly just hasn’t happened. Oh sure, maybe your web browser now fills in an address or a credit card number on a web form. (Though it mostly gets that wrong when I try it.) But not all the other detail. Why not?

Many poor people have to fill out a lot of forms to apply for many kinds of assistance. Roughly once a year I’m told, at least. They see many of these forms as so hard to fill our that many of them just don’t bother unless they get help from someone like a social worker. So a lot of programs to help the poor don’t actually help many of those who are eligible, because they don’t fill out the forms.

So why doesn’t some tech company offer a form app, where you give all your personal info to the form and it fills out most parts of most forms for you? You just have to do the unusual parts. And they could have a separate app to give to orgs that create forms, so they can help make it easier for their forms to get filled out. Yes, much of the effort to make this work is more in standardization than in abstract computer algorithms. But still, why doesn’t some big firm do it?

I suggested all this to a social worker I know, who was aghast; she didn’t want this tech firm knowing all these details, like her social security number. But if you fill out all these forms by hand today, you are telling it all to one new org per year. Adding one firm to the list to make it all much easier doesn’t seem like such a high cost to me.

But maybe this is all about the optics; tech firms fear looking like big brother if they know all this stuff about you. Or many legal liability would fall on these tech firms if the form had any mistakes. Or maybe privacy laws prevent them from even asking for the key info. And so we all suffer with forms, and poor folks don’t get the assistance offer to them. And we all lose, though those of us who are better at filling out forms lose less.

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

Pay More For Results

A simple and robust way to get others to do useful things is to “pay for results”, i.e., to promise to make particular payments for particular measurable outcomes. The better the outcomes, the more someone gets paid. This approach has long been used in production piece-rates, worker bonuses, sales commissions, CEO incentive paylawyer contingency fees, sci-tech prizes, auctions, and outcome-contracts in PR, marketing, consulting, IT, medicine, charities, development, and in government contracting more generally. 

Browsing many articles on the topic, I mostly see either dispassionate analyses of its advantages and disadvantages, or passionate screeds warning against its evils, especially re sacred sectors like charity, government, law, and medicine. Clearly many see paying for results as risking too much greed, money, and markets in places where higher motives should reign supreme.

Which is too bad, as those higher motives are often missing, and paying for results has a lot of untapped potential. Even though the basic idea is old, we have yet to explore a great many possible variations. For example, many of social reforms that I’ve considered promising over the years can be framed as paying for results. For example, I’ve liked science prizes, combinatorial auctions, and:

  1. Buy health, not health careGet an insurer to sell you both life & health insurance, so that they lose a lot of money if you are disabled, in pain, or dead. Then if they pay for your medical expenses, you can trust them to trade those expenses well against lower harm chances.
  2. Fine-insure-bounty criminal law systemCatch criminals by paying a bounty to whomever proves that a particular person did a particular crime, require everyone to get crime insurance, have fines as the official punishment, and then let insurers and clients negotiate individual punishments, monitoring, freedoms, and co-liabilities. 
  3. Prediction & decision markets – There’s a current market probability, and if you buy at that price you expect to profit if you believe a higher probability. In this way you are paid to fix any error in our current probabilities, via winning your bets. We can use the resulting market prices to make many useful decisions, like firing CEOs. 

We have some good basic theory on paying for results. For example, paying your agents for results works better when you can measure the things that you want sooner and more accurately, when you are more risk-averse, and when your agents are less risk-averse. It is less less useful when you can watch your agents well, and you know what they should be doing to get good outcomes.

The worst case is when you are a big risk-neutral org with lots of relevant expertise who pays small risk-averse individuals or organizations, and when you can observe your agents well and know roughly what they should do to achieve good outcomes, outcomes that are too complex or hidden to measure. In this case you should just pay your agents to do things the right way, and ignore outcomes.

In contrast, the best case for paying for results is when you are more risk-averse than your agents, you can’t see much of what they do, you don’t know much about how they should act to best achieve good outcomes, and you have good fast measure of the outcomes you want. So this theory suggests that ordinary people trying to get relatively simple things from experts tend to be good situations for paying for results, especially when those experts can collect together into large more-risk-neutral organizations.

For example, when selling a house or a car, the main outcome you care about is the sale price, which is quite observable, and you don’t know much about how best to sell to future buyers. So for you a good system is to hold an auction and give it to the agent who offers the highest price. Then that agent can use their expertise to figure out how to best sell your item to someone who wants to use it.

While medicine is complex and can require great expertise, the main outcomes that you want from medicine are simple and relatively easy to measure. You want to be alive, able to do your usual things, and not in pain. (Yes, you also have a more hidden motive to show that you are willing to spend resources to help allies, but that is also easy to measure.) Which is why relatively simple ways to pay for health seem like they should work. 

Similarly, once we have defined a particular kind of crime, and have courts to rule on particular accusations, then we know a lot about what outcomes we want out of a crime system: we want less crime. If the process of trying to detect or punish a criminal could hurt third parties, then we want laws to discourage those effects. But with such laws in place, we can more directly pay to catch criminals, and to discourage the committing of crimes. 

Finally when we know well what events we are trying to predict, we can just pay people who predict them well, without needing to know much about their prediction strategies. Overall, paying for results seems to still have enormous untapped potential, and I’m doing my part to help that potential be realized.

GD Star Rating
loading...
Tagged as: ,

Why Age of Em Will Happen

In some technology competitions, winners dominate strongly. For example, while gravel may cover a lot of roads if we count by surface area, if we weigh by vehicle miles traveled then asphalt strongly dominates as a road material. Also, while some buildings are cooled via fans and very thick walls, the vast majority of buildings in rich and hot places use air-conditioning. In addition, current versions of software systems also tend to dominate over old older versions. (E.g., Windows 10 over Windows 8.)

However, in many other technology competitions, older technologies remain widely used over long periods. Cities were invented ten thousand years ago, yet today only about half of the population lives in them. Cars, trains, boats, and planes have taken over much transportation, yet we still do plenty of walking. Steel has replaced wood in many structures, yet wood is still widely used. Fur, wool, and cotton aren’t used as often as they once were, but they are still quite common as clothing materials. E-books are now quite popular, but paper books sales are still growing.

Whether or not an old tech still retains wide areas of substantial use depends on the average advantage of the new tech, relative to the variation of that advantage across the environments where these techs are used, and the variation within each tech category. All else equal, the wider the range of environments, and the more diverse is each tech category, the longer that old tech should remain in wide use.

For example, compare the set of techs that start with the letter A (like asphalt) to the set that start with the letter G (like gravel). As these are relatively arbitrary sets that do not “cut nature at its joints”, there is wide diversity within each category, and each set is all applied to a wide range of environments. This makes it quite unlikely that one of these sets will strongly dominate the other.

Note that techs that tend to dominate strongly, like asphalt, air-conditioning, and new software versions, more often appear as a lumpy change, e.g., all at once, rather than via a slow accumulation of many changes. That is, they more often result from one or a few key innovations, and have some simple essential commonality. In contrast, techs that have more internal variety and structure tend more to result from the accumulation of more smaller innovations.

Now consider the competition between humans and computers for mental work. Today human brains earn more than half of world income, far more than the costs of computer hardware and software. But over time, artificial hardware and software have been improving, and slowly commanding larger fractions. Eventually this could become a majority. And a key question is then: how quickly might computers come to dominate overwhelmingly, doing virtually all mental work?

On the one hand, the ranges here are truly enormous. We are talking about all mental work, which covers a very wide of environments. And not only do humans vary widely in abilities and inclinations, but computer systems seem to encompass an even wider range of designs and approaches. And many of these are quite complex systems. These facts together suggest that the older tech of human brains could last quite a long time (relative of course to relevant timescales) after computers came to do the majority of tasks (weighted by income), and that the change over that period could be relatively gradual.

For an analogy, consider the space of all possible non-mental work. While machines have surely been displacing humans for a long time in this area, we still do many important tasks “by hand”, and overall change has been pretty steady for a long time period. This change looked nothing like a single “general” machine taking over all the non-mental tasks all at once.

On the other hand, human minds are today stuck in old bio hardware that isn’t improving much, while artificial computer hardware has long been improving rapidly. Both these states, of hardware being stuck and improving fast, have been relatively uniform within each category and across environments. As a result, this hardware advantage might plausibly overwhelm software variety to make humans quickly lose most everywhere.

However, eventually brain emulations (i.e. “ems”) should be possible, after which artificial software would no longer have a hardware advantage over brain software; they would both have access to the same hardware. (As ems are an all-or-nothing tech that quite closely substitutes for humans and yet can have a huge hardware advantage, ems should displace most all humans over a short period.) At that point, the broad variety of mental task environments, and of approaches to both artificial and em software, suggests that ems many well stay competitive on many job tasks, and that this status might last a long time, with change being gradual.

Note also that as ems should soon become much cheaper than humans, the introduction of ems should initially cause a big reversion, wherein ems take back many of the mental job tasks that humans had recently lost to computers.

In January I posted a theoretical account that adds to this expectation. It explains why we should expect brain software to be a marvel of integration and abstraction, relative to the stronger reliance on modularity that we see in artificial software, a reliance that allows those systems to be smaller and faster built, but also causes them to rot faster. This account suggests that for a long time it would take unrealistically large investments for artificial software to learn to be as good as brain software on the tasks where brains excel.

A contrary view often expressed is that at some point someone will “invent” AGI (= Artificial General Intelligence). Not that society will eventually have broadly capable and thus general systems as a result of the world economy slowly collecting many specific tools and abilities over a long time. But that instead a particular research team somewhere will discover one or a few key insights that allow that team to quickly create a system that can do most all mental tasks much better than all the other systems, both human and artificial, in the world at that moment. This insight might quickly spread to other teams, or it might be hoarded to give this team great relative power.

Yes, under this sort of scenario it becomes more plausible that artificial software will either quickly displace humans on most all jobs, or do the same to ems if they exist at the time. But it is this scenario that I have repeatedly argued is pretty crazy. (Not impossible, but crazy enough that only a small minority should assume or explore it.) While the lumpiness of innovation that we’ve seen so far in computer science has been modest and not out of line with most other research fields, this crazy view postulates an enormously lumpy innovation, far out of line with anything we’ve seen in a long while. We have no good reason to believe that such a thing is at all likely.

If we presume that no one team will ever invent AGI, it becomes far more plausible that there will still be plenty of jobs tasks for ems to do, whenever ems show up. Even if working ems only collect 10% of world income soon after ems appear, the scenario I laid out in my book Age of Em is still pretty relevant. That scenario is actually pretty robust to such variations. As a result of thinking about these considerations, I’m now much more confident that the Age of Em will happen.

In Age of Em, I said:

Conditional on my key assumptions, I expect at least 30 percent of future situations to be usefully informed by my analysis. Unconditionally, I expect at least 5 percent.

I now estimate an unconditional 80% chance of it being a useful guide, and so will happily take bets based on a 50-50 chance estimate. My claim is something like:

Within the first D econ doublings after ems are as cheap as the median human worker, there will be a period where >X% of world income is paid for em work. And during that period Age of Em will be a useful guide to that world.

Note that this analysis suggests that while the arrival of ems might cause a relatively sudden and disruptive transition, the improvement of other artificial software would likely be more gradual. While overall rates of growth and change should increase as a larger fraction of the means of production comes to be made in factories, the risk is low of a sudden AI advance relative to that overall rate of change. Those concerned about risks caused by AI changes can more reasonably wait until we see clearer signs of problems.

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

Who Likes Simple Rules?

Some puzzles:

  • People are often okay with having either policy A or policy B adopted as the standard policy for all cases. But then they object greatly to a policy of randomly picking A or B in particular cases in order to find out which one works better, and then adopt it for everyone.
  • People don’t like speed and red-light cameras; they prefer human cops who will use discretion. On average people don’t think that speeding enforcement discretion will be used to benefit society, but 3 out of 4 expect that it will benefit them personally. More generally people seem to like a crime law system where at least a dozen different people are authorized to in effect pardon any given person accused of any given crime; most people expect to benefit personally from such discretion.
  • In many European nations citizens send their tax info into the government who then tells them how much tax they owe. But in the US and many other nations, too many people oppose this policy. The most vocal opponents think they benefit personally from being able to pay less than what the government would say they owe.
  • The British National Health Service gets a lot of criticism from choosing treatments by estimating their cost per quality-adjusted-life-year. US folks wouldn’t tolerate such a policy. Critics lobbying to get exceptional treatment say things like “one cannot assume that someone who is wheel-chair bound cannot live as or more happily. … [set] both false limits on healthcare and reducing freedom of choice. … reflects an overly utilitarian approach”
  • There’s long been opposition to using an official value of life parameter in deciding government policies. Juries have also severely punished firms for using such parameters to make firm decisions.
  • In academic departments like mine, we tell new professors that to get tenure they need to publish enough papers in good journals. But we refuse to say how many is enough or which journals count as how good. We’d keep the flexibility to make whatever decision we want at the last minute.
  • People who hire lawyers rarely know their track record at winning vs. losing court cases. The info is public, but so few are interested that it is rarely collected or consulted. People who hire do know the prestige of their schools and employers, and decide based on that.
  • When government leases its land to private parties, sometimes it uses centralized, formal mechanisms, like auctions, and sometimes it uses decentralized and informal mechanisms. People seem to intuitively prefer the latter sort of mechanism, even though the former seems to works better. In one study “auctioned leases generate 67% larger up-front payments … [and were] 44% more productive”.
  • People consistently invest in managed investment funds, which after the management fee consistently return less than index funds, which follow a simple clear rule. Investors seem to enjoy bragging about personal connections to people running prestigious investment funds.
  • When firms go public via an IPO, they typically pay a bank 7% of their value to manage the process, which is supposedly spent on lobbying others to buy. Google famously used an auction to cut that fee, but banks have succeed in squashing that rebellion. When firms try to sell themselves to other firms to acquire, they typically pay 10% if they are priced at less than $1M, 6-8% if priced $10-30M, and 2-4% if priced over $100M.
  • Most elite colleges decide who to admit via opaque and frequently changing criteria, criteria which allow much discretion by admissions personnel, and criteria about which some communities learn much more than others. Many elites learn to game such systems to give their kids big advantages. While some complain, the system seems stable.
  • In a Twitter poll, the main complaints about my fire-the-CEO decisions markets proposal are that they don’t want a simple clear mechanical process to fire CEOs, and they don’t want to explicitly say that the firm makes such choices in order to maximize profits. They instead want some people to have discretion on CEO firing, and they want firm goals to be implicit and ambiguous.

The common pattern here seems to me to be a dislike of clear formal overt rules, mechanisms, and criteria, relative to informal decisions and negotiations. Especially disliked are rules based on explicit metrics that might reject or disapprove people. To the extent that there are rules, there seems to be a preference for authorizing some people to have discretion to make arbitrary choices, regarding which they are not held strongly to account.

To someone concerned about bribes, corruption, and self-perpetuating cabals of insiders, a simple clear mechanism like an auction might seem an elegant way to prevent all of that. And most people give lip service to being concerned about such things. Also, yes explicit rules don’t always capture all subtleties, and allowing some discretion can better accommodate unusual details of particular situations.

However, my best guess is that most people mainly favor discretion as a way to promote an informal favoritism from which they expect to benefit. They believe that they are unusually smart, attractive, charismatic, well-connected, and well-liked, just the sort of people who tend to be favored by informal discretion.

Furthermore, they want to project to associates an image of being the sort of person who is confidently supports the elites who have discretion, and who expects in general to benefit from their discretion. (This incentive tends to induce overconfidence.)

That is, the sort of people who are eager to have a fair neutral objective decision-making process tend to be losers who don’t expect to be able to work the informal system of favors well, and who have accepted this fact about themselves. And that’s just not the sort of image that most people want to project.

This whole equilibrium is of course a serious problem for we economists, computer scientists, and other mechanism and institution designers. We can’t just propose explicit rules that would work if adopted, if people prefer to reject such rules to signal their social confidence.

GD Star Rating
loading...
Tagged as: ,