Coherent Futures

There is very little careful thinking about the far future.  Yes, many people talk about the far future as if they cared, but typical interest rates suggest few care much, and most far future discussion seems to be symbolic talk about current issues.  Furthermore, when people do actually directly discuss the far future, they tend to engage in an extreme far-think, vs. a more practical near-think:

[NEAR] All of these bring each other more to mind: here, now, me, us; trend-deviating likely real local events; concrete, context-dependent, unstructured, detailed, goal-irrelevant incidental features; feasible safe acts; secondary local concerns; socially close folks with unstable traits.
[FAR] Conversely, all these bring each other more to mind: there, then, them; trend-following unlikely hypothetical global events; abstract, schematic, context-freer, core, coarse, goal-related features; desirable risk-taking acts, central global symbolic concerns, confident predictions, polarized evaluations, socially distant people with stable traits.

While our future vision should fade into an increasingly vast and uncertain fog of possibilities, far future fans instead fragment into factions, each confident in a very different view of the important future issues.  Factions use such different assumptions that they rarely build on each others’ work, or even engage others in debate.  Only they really “get it” you see, and few others ever seriously consider their arguments.  Extreme far-thinking apparently produces extreme disagreement.

Such fragmentation may be acceptable when searching a large space for rare combinations, but it is severely dysfunctional for advising common actions.  We instead need to find ways for the few people who actually care about the far future to work together via a division of labor.  But how can we do that?  Just tell each faction to reconsider that they might be mistaken?

I’ve said that with a million to spend, I’d build fire-the-CEO decision markets, revolutionizing corporate accountability via a powerful demo that can inspire related projects.  But if my purpose was not to demo or develop the tech, or to generally improve society, but just to get the info that I personally and directly cared the most about, I’d instead make real-money combinatorial markets on the far future (say one to ten decades out).

I’ve developed a combinatorial betting tech that lets a few or many users edit an always-coherent joint probability distribution over all value combinations of some set of base variables.  Far futures base variables might include the years of important tech milestones, population, wealth, or mortality values at particular future dates, etc.  Each user edit would be backed by a bet, a bet invested in assets paying competitive interest/returns.  This combo bet tech worked well in published lab tests, several firms have used it, and I’m now working with Consensus Point to deliver a robust commercial implementation.  More on the tech here, here, and here.

Such markets could fend off fragmented far-thinking futurism in several ways

Incentives Since bets change how much money your descendants (or maybe you) will have in the far future, betting puts you in a more practical frame of mind.  You would edit estimates with an eye more for realism, versus an inspiring story.  Yes incentives on far future bets can be much weaker than incentives on near term bets, but the relevant comparison is across mechanisms on the same topic, not across topics; it is just hard for any mechanism to give strong accuracy incentives on far future topics.

Engagement When your faction pushes up the probabilities of its favored scenarios, and then others push those probabilities back down again, you can’t as easily tell yourself they are just unthinkingly dismissing your views.  Since you know they face stronger incentives, their actions carry a stronger signal for you to consider.

Specialization Participants would tend to focus their edits on particular topics, such as demographics, environment, war, nanotech, genetic engineering, surveillance, or AI.  Equally important, they could specialize in particular relations between topic areas.  Someone might, for example, specialize in how nanotech and AI relate, e.g., how the arrival of one makes the other more or less likely soon after.  Then when nanotech estimates change, AI estimates would automatically change.  In this way we could work together to build a coherent joint distribution.

Far future factions may say “Sure this could work in principle, but in practice so few get it that your market prices would be useless noise.  Yes, our special future insight would allow us to increase our long term investment returns by a few orders of magnitude, but that is peanuts compared to all that is at stake.”

But coherent futures markets can also help factions 1) gain converts, and 2) get help from non-converts.  If, as you suspect, other factions know in their hearts their evidence is weaker, market odds will tilt toward you, convincing some on the margin to join your efforts.  Furthermore, your faction can always focus on what these markets say conditional on their favored assumptions, and that factional view should be substantially improved by others’ efforts to improve the joint distribution on topics away from your faction’s focus.

For example, my guess is that the most important far future issues to think through now center on whole brain emulations.  Since few futurists think much about ems now, I’d expect few to focus on em variables at first.  But I’d expect speculators who did consider ems to mostly agree that ems will more likely than not appear within a century, at least if even more dramatic events do not happen first, and that ems have quite dramatic and rapid implications for other future topics.  This should draw more interest into ems.

More important, the few people who focus on ems simply do not have the time to consider much about how ems interact with other changes. People focusing on other topics, such as war, environment, or AI, should help refine the market’s estimates of how ems would affect those topics, and how they would affect ems.  They would do this both by updating unconditional estimates on those topics, and by updating the relations between those topics and more em-related topics.

Few people really care much about the distant future, and sadly those few are mostly fragmented into far-thinking factions.  Combinatorial betting markets could offer them incentives toward a more practical frame of mind, stronger signals to induce engagement, and a framework to let them collaborate via a consensus joint probability distribution over all important future topics.

It really is possible.  We have the tech, but not yet the will.  I expect this will have to wait for a patron with a stronger loyalty to the far future than to any particular far-thinking fragmented faction.  I’ve been looking for six years now, and will continue for another sixty years, if necessary.

GD Star Rating
loading...
Tagged as: , ,
Trackback URL:
  • anonymous

    Far-future prediction markets are quite likely to be illiquid due to scarcity of information and knowledge. An illiquid prediction market can only provide uncertain probability intervals, and extending such uncertain representations to deal with joint combinatorial events seems to be an open research problem (see these recent preprints on arxiv: [1] [2] [3]).

    How would your combinatorial prediction markets deal with these issues?

  • http://entitledtoanopinion.wordpress.com TGGP

    anonymous, your third link was just to this very OB post.

  • anonymous

    my mistake, see [3]

  • http://profile.typekey.com/robinhanson/ Robin Hanson

    anonymous, info scarcity is not a plausible cause of illiquidity and the markets I’ve designed have guaranteed liquidity. The papers you cite are about general difficulties of efficiently representing uncertainty – they don’t apply with any extra force to my mechanism or application.

  • John P.

    Robin,

    I’m a fairly recent math/comp sci graduate working in finance. I’m interested in implementing prediction markets in the real world, and was wondering if you had any advice on doing so professionally, recreationally, or academically. Are there any open source prediction markets that are having any success? Do I need a PhD to be taken seriously? What sort of formal or informal education is necessary to contribute meaningfully?

  • http://profile.typekey.com/aroneus/ Aron

    I love the promise of prediction markets, but I’m not sure why you would focus on long-term or even medium-term predictions when discussing them (is this pandering to the perceived blog audience?). I’d aim for rounding out their versatility and work first in the general time horizons where they seem to provide the most useful feedback.

    Nitpick: affect is the verb. effect is the noun.

  • Cyan

    Aron, it does appear that Robin wanted “affect” where he wrote “effect”, but “effect” can be a verb too.

  • http://profile.typekey.com/sentience/ Eliezer Yudkowsky

    I was happy to engage in this bet with Unknown at odds of $10 now to $1000 inflation-adjusted then. Unknown happened to phrase his predicted event in such fashion as to contain no loopholes from my perspective: that really just Can’t Happen unless my whole understanding of mind is flat wrong.

    But I would not have been willing to engage in that bet if I had needed to post $1000 of collateral until the end of time. It’s not that I discount the future so highly, but rather that I have what I think are “unusually good investments” of my current resources whose “dividends” (paid off in existential risk reduction, which I value) the S&P 500 can’t match. In the fantastic event that Unknown’s prediction comes true and I don’t have $1000 in the bank at that time, I’ll take out a loan. But meanwhile I need the resources I have.

    At least I’m betting on the far future – though in retrospect, we both should have taken bids! (As it is the market price was fiat-established by my offer, which isn’t fair.)

    But if collateral is not posted, then credit risk (what if I get hit by a truck between now and then?) will widen spreads to the point that no trades occur, except in extreme cases like the one above. If I’d demanded $100 for $1000 inflation-adjusted would the bet have still occurred?

    I’m not willing to make an indefinite series of trades like the one I did with Unknown, because I need to have confidence that I can pay off the net amount of all outstanding bets if any possible series of impossible events comes true. (And if I weren’t the sort of person who would abide by that, others wouldn’t bet with me, or would demand higher payments for credit risk.)

    However, I am willing to make at least one more bet of the type shown above – $1000 inflation-adjusted payout if someone builds a superintelligence (which must at least be strictly transhuman in the sense of dominating a human on any test of intelligence) without worrying at all about Friendliness, and the Earth is still there afterward. So I am willing to sell that bet to the highest bidder who offers immediate payment, which would give us a fairer view of the “market odds”. I leave it to Robin Hanson to say whether I should accept bids now, or do so later under some other conditions. I am not, however, willing to post collateral – though I will take out a loan or hire myself out or do whatever else is required to pay off, should the event occur.

  • http://thomblake.com Thom Blake

    Eliezer,

    I think the bet as it stands is a little ambiguous. “Without worrying at all about friendliness” seems a bit far-fetched if you a)consider it “worrying at all” if the designer ever even thought about the issue or read a blog post about it, and b)haven’t established how you’ll know whether “worrying at all” happened.

  • http://profile.typekey.com/robinhanson/ Robin Hanson

    John, the availability of college graduates, or even Ph.D.s, willing to work in this area is simply not a limiting factor. The limit is mostly finding customers willing to pay for more accurate forecasts. Most everyone in this area is “hired” on commission – to get paid they had to find a customer.

    Aron, as I’ve said to build interest, validate, and develop a tech, yes you’d want to focus on short term applications. But then once you have an info tech you use it to get the info you want. The info I want is about the far future.

    Eliezer, I’ll happily start the bidding at $10, but the markets I have in mind would better disentangle the probability and interest rate factors that your winning bid would represent. They would indeed require collateral, but I expect many in your camp would be willing to participate, if you couldn’t.

  • http://profile.typekey.com/robinhanson/ Robin Hanson

    Aron and Cyan, I made the spelling correction you suggest.

    I retract my bid – it is too hard to disentangle the factors.

  • http://profile.typekey.com/sentience/ Eliezer Yudkowsky

    Thom, I’d consider the obvious interpretation to be that if the designer states afterward that they underwent some type of training program or explicit programming operation to shape the AI to have end-in-itself X and that’s why it now wants to do X, where X is anything remotely like “don’t kill humans” or “care what sentient beings want”, that would count as “worrying about Friendliness”. If X is “to make money” or “to maximize this reward button” then that would not count as “worrying about Friendliness”, even if the team tries to teach the AI that “you can best ensure that you make money by following the laws” or some such. I am willing to be charitable toward Unknown in interpreting this factor; it’s not like I expect to survive in borderline cases.

  • http://profile.typekey.com/sentience/ Eliezer Yudkowsky

    Robin, it’s not clear to me what would count as “disentangling the factors”. Do you mean how to adjust the payoff amount for inflation, or how to account for default risk? Shouldn’t the overlap between my bid and your ask be wide enough to account for these? – Anyway, it’s not clear to me what you mean by this.

  • http://profile.typekey.com/robinhanson/ Robin Hanson

    Eliezer, to estimate the value of X times payoff now versus in a future state where Eliezer and I live to see an unfriendly AI, I need to estimate the date an AI will appear, the chance it will be unfriendly, the chance Eliezer will be alive then, the chance I’ll be alive then, and the zero and full risk interest rates between now and then, and the correlation between these events and full market risk.

  • anon

    Sorry for the offtopic, but I’ve got a word stuck on the tip of my tongue. How does one call the tendency to justify an action taken solely by emotion for making up reasonable-sounding stories for taking the action?

  • ShardPhoenix

    @anon:

    Rationalization

  • anon

    thank you very much, ShardPhoenix!

  • http://jamesdmiller.blogspot.com/ James D. Miller

    John P. -

    If you want to help build prediction markets you should go here:

    http://zocalo.sourceforge.net/

  • frelkins

    @John P.

    math/comp sci graduate working in finance.”
    Useful but not mandatory. I have a degree in Ancient Greek and attended the Jack Kerouac School of Disembodied Poetics.

    implementing prediction markets in the real world
    Of course.

    any advice on doing so professionally” A
    Wherever you are working now, look at the low-hanging fruit, probably software projects or sales revenue forecasts. The business case is easy to make, even I could do it. But you have to put in the work first.

    open source prediction markets that are having any success
    I second Chris Hibbert‘s Zocalo.

    Do I need a PhD to be taken seriously
    Academically, yes. Practically, no. Do you want to write papers about markets or run them in a corporation? I don’t have a Phd, I don’t think Bo Cowgill has one, I don’t think Jeff Severts’ has one, etc. Some people who run them do. But if you don’t have one, I recommend that you find a friendly economics or business professor ASAP if only to comfort senior management.

    informal education is necessary to contribute meaningfully:
    Start reading. Read all of Robin’s papers; all of Tetlock’s papers; all of Wolfers & Zitewitz; read Bo’s paper; read Jed’s stuff; read Berg’s papers at Iowa; subscribe to the Journal if you can. I first decided we needed a market here after a meeting in July 2006 – I presented the idea to an executive, then the head of development and he said, yes it might interest him – so I started reading and read from mid-July to mid-October.

    The executive then looked at my 1-page summary and asked for the formal business case, which I started writing in Nov. 2006 with the aid of Dave Pennock and Chris Hibbert. I was fortunate at that time because the Yahoo confab on markets was in NYC. The executive liked the business case, which I then shopped around for support & comment.

    After incorporating all that feedback, I presented the business case with value-of-information trees and ROI to the Board in mid-Jan. 2007. I also had already contacted 3 vendors and demo’d each solution briefy to the Board.

    Approval was instantaneous, and I launched the market in Mar. 2007. If I can do it, everyone can do it. Gentlemen, start your markets now!

  • http://profile.typekey.com/bayesian/ Peter McCluskey

    Robin, the artificially provided liquidity you have in mind doesn’t imply enough trading volume from profit-oriented traders to provide informative prices, which is one of the things people have in mind when they worry about liquidity. It’s implausible that info scarcity is a complete explanation for why some markets fail to attract enough trading, but my intuition tells me it’s not completely wrong. Do you think you have a plausible model of what causes trading levels to be adequate that doesn’t involve what information is available?

    Eliezer, I’d be tempted to accept your bet, but I’m reluctant to analyze how Eliezer[2008]‘s intent to interpret charitably will constrain Eliezer[2030]‘s decisions. I’d want wording that I’d expect a neutral third party could interpret in a fairly predictable way, and I don’t think I can produce such wording.

  • http://hanson.gmu.edu Robin Hanson

    Peter, I’m not following you. If you start a subsidized market maker at price X and then someone changes the price to Y, you’ve gained the info that someone thinks Y is a better estimate than X. More trades give more info, but even one trade gives some info. And zero trades tell you that no one who looked thought they knew of a better estimate. Of course you might have hoped for even more trades, and then blame too low a subsidy or too little publicity.

  • Jeff

    frelkins,

    Would you be willing to post your one-page memo and business case somewhere for people interested? Or perhaps put together a 5-page briefing memo with links? You’ve done the work to read for 3 months, can you make that work put in work for others by cutting down their need to research/read to 1 month?

  • D. Alex

    I’ll take Eliezeer’s bet. I’ll start at $0.01 (since there is no higher bidder atm). Eliezeer, please advise me when the bidding is over, so I can paypal the money. Also, please prepare a digitally signed note stating your (or your “successor(s)”) undertaking to pay – should be a fun excercise in itself.

    I will consider what is a fair value for such a bet from my perspective (to sum up, I think the possibility of AI which would dominate any ONE human in any GENERAL test of intelligence, but cannot exponentially self-improve is pretty high) and – time permitting – will post my reasoning in due course.

    D. Alex

  • frelkins

    @Jeff

    Alas, I cannot do so; I hope you understand! However, I am trying to get my hair to the prediction market conference in SF this coming Jan. There I will be able to talk about our market, our vendor – with whom I’m quite proud to be associated – and my experiences running the market.

    If I can manage to get there, I will be happy to talk to everyone who needs help. I have also tried to hold meet-ups here in NYC but no one seems much interested recently due to the financial crisis. Hopefully that will soon change.

    Finally, my 2007 speech is available, altho’ some corporate details are a little different due to positive restructuring – still, the main story is still valid and I hope inspiring. The speech at least I can freely share altho’ it is a little large at 26mg of QT.

  • http://profile.typekey.com/bayesian/ Peter McCluskey

    Robin: yes, I was imprecise. By “informative prices” I meant prices that provide information that could plausibly be considered more valuable than the subsidy used to generate them.
    What you get from observing zero trades is nearly worthless without a fair amount of information about the traders who thought about trading. If the initial price is set to a price that reflects the readily available information, I expect you’ll often get zero trades. If you set the initial price somewhere else, the first few trades will give you information that is readily available.

  • http://www.spaceandgames.com Peter de Blanc

    I’ll raise the bid to $20.

  • http://profile.typekey.com/robinhanson/ Robin Hanson

    Peter M., the subsidy level I set is my offer price for info. If no one accepts my offer, I lose nothing, and if someone does accept it, I’ve gained info at the price I set. Of course there are also fixed costs associated with setting up such a system, so I’d want to set my price high enough to produce enough volume to cover those fixed costs as well.

  • http://profile.typekey.com/sentience/ Eliezer Yudkowsky

    de Blanc, you surprise me. Why, or are you just making markets?

  • http://profile.typekey.com/1227585392s27146/ Philip Goetz

    “However, I am willing to make at least one more bet of the type shown above – $1000 inflation-adjusted payout if someone builds a superintelligence (which must at least be strictly transhuman in the sense of dominating a human on any test of intelligence) without worrying at all about Friendliness, and the Earth is still there afterward.”

    I think you can delete the ambiguous clause “and the Earth is still there afterward.” The continued existence of Eliezer, the other bettor, and currency, should suffice.

  • http://profile.typekey.com/bayesian/ Peter McCluskey

    Robin: yes, if you ignore questions of whether the information is affordable, then there isn’t any problem.

  • http://profile.typekey.com/robinhanson/ Robin Hanson

    Peter, my post described all these different factions with very different perspectives. At first one could just get representatives from several groups to contribute their intuitive judgments, weighed by their willingness to bet. A coherent summary of even that would be quite valuable I think. Could even that not be induced via sufficient subsidy and publicity?

  • http://rhollerith.com/blog Richard Hollerith

    look at the low-hanging fruit, probably software projects

    If the source code for the software project is freely available, then the futures contract can be specified in a formal language, that is, it can consist of a computer program that accepts the source code as input and outputs text explaining who owes who what amount of money. The advantage of using a formal language for a contract is that the the cost of adjudicating the contract does not rise much as the complexity of the contract grows because (briefly) the chances that a judge or jury or even a lawyer is needed is much lower than when the contract is written in a natural or informal language like most legal contracts are. There have been at least a half dozen papers (written by programming-language researchers and a legal scholar) on formal languages in which to write contracts, and I can provide references to anyone who contacts me back channel. I can also provide detailed examples of very complex formal futures contracts about source code that will tend to be economically valuable.

    I think it would prove economically very valuable to make it possible for professional computer programmers to make money like derivatives traders do; it would provide a middle way between the programmer’s being a contractor (or employee) and the programmer’s being a stockholder in a “startup” (a young corporation). It is also an answer to the question of how society can drastically increase the number of programmers making a good living working on open-source software: make a market where people can trade bets on the future availability of specific economically-valuable modifications of the software: the programmer makes a large bet that such and such modification of such and such software will become available by such and such date, then personally uses his/her programming skills to make the modification available. There is a “free rider” problem in that if two programmers have bets on the availabilility of the same modification, then it is in the interests of each to have the other one do all the work of making the modification available, and I am not enough of an economist or mechanism designer to know how pernicious or tractable that problem is.

  • Pingback: Robin Hanson has convinced Concensus Point to support combinatorial prediction markets. | Midas Oracle .ORG