aa r X i v : . [ q -f i n . E C ] F e b Blunt Honesty, Incentives, and Knowledge Exchange
Bruce Knuteson ∗ We propose a simple mechanism to facilitate the buying and selling of useful, bluntly honest infor-mation. The for-profit, arm’s length knowledge exchange this mechanism enables may dramaticallyincrease the pace of scientific progress.
Contents
I. Motivation II. Q&A III. Summary Acknowledgments References I. MOTIVATION
The past two decades have witnessed an explosionof valuable information [1] easily accessible to billionsof people. The breadth and depth of quality informa-tion available for free, at your fingertips with just a fewkeystrokes, is breathtakingly world changing. Extraor-dinary examples of individuals and organizations freelysharing information to propel advances in science, indus-try, design, and human well-being are all around. In-formation hoarding by companies and elites is rapidlydisappearing in an era of increasing and unprecedentedtransparency. Information, which has always wanted tobe free, is. Content providers, forced into a greater re-liance on advertising revenue, maintain a clear line be-tween advertisements and the objective integrity of jour-nalistic content. This remarkable transition in the wayinformation is produced and disseminated is one of themost profound in recent human history, and one withpowerful implications for our future. After thousands ofyears of backbreaking labor, we now live and work in anInformation Economy.Moreover, we live in an Information Economy in whichinformation is free.Against this backdrop, this article’s purpose is limitedand superficially regressive. We point out a small sub-set of information exchange in which both parties maybenefit from the exchange not being free. The scenarioin Section II illuminates this limited subset of informa-tion exchange while simultaneously motivating a protocoldesigned to facilitate such exchanges. Section III summa-rizes. ∗ Electronic address: [email protected]
II. Q&A
Scientist A, who recently discovered an interesting,not-yet-published property of a particular molecule,hears through the grapevine that her discovery might bevaluable to Scientist Q, who has been working for yearsto develop a cure for a rare disease [2]. The two scientistsQ and A work in different fields, do not know each other,and have no common contacts.A wants to sell her knowledge to Q for $100K. Q willhappily pay $100K for any information likely to helpher complete her ambitious research program. The obvi-ous transaction, unfortunately, is problematic. Before Qagrees to pay $100K for A’s information, Q wants someguarantee of the usefulness and accuracy of A’s informa-tion. The only way A can convince Q the informationis worth paying for is by telling Q the information. If Atells Q the information before the transaction is agreedupon, Q has no reason to agree to the transaction. BothQ and A understand all of this, so A never approachesQ with her discovery, Q gets hit by a bus on a sunnyafternoon two months later, the cure nearly within Q’sgrasp goes undiscovered for another fifteen years, andsome ten thousand people and their families worldwide,including someone you love, suffer unnecessarily duringthe intervening decade and a half.Our society’s existing procedures for dealing with caseslike this, often involving legal contracts, can be expensiveto construct, monitor, and enforce. Perhaps we can de-vise something better.To assure Q of the usefulness of A’s information, Q andA can agree that Q is interested in an answer to the ques-tion “What molecule will enable the following reaction?”Important details are clarified in a couple of accompany-ing paragraphs. Q and A agree that a valid answer mustbe “a molecule, fully specified in its structure.” Agreeingon the category of valid answers is very important to Q,who has no desire to pay $100K to be told the answer is“a molecule similar in character to one that can be foundin the excrement of some insects” – which, while perhapstrue, is of little use to Q. Q and A agree on a third party,X, who will decide whether A’s answer falls within theagreed upon category of allowed possible answers.Q will only agree to the transaction if she considersA’s incentives to be well aligned with providing a bluntlyhonest answer [3]. A must have some skin in the game. Amust be rewarded if correct, and penalized if incorrect [4].The social and reputational incentives guiding the ac-tions of individuals can be difficult to understand, andharder still to engineer. Anonymizing Q and A conve-niently mitigates these incentives. Nobody other thanX, who brokers the transaction between Q and A, needsto know the identities of Q and A (or their agents). Thisanonymization is liberating, allowing Q to ask questionswithout worrying about appearing ignorant, and allow-ing A to provide a bluntly honest answer without havingto worry about delivering insult or saving face.With social and reputational incentives thus sidelined,money is the natural carrot and stick. A must lose moneyif she is wrong and make money if she is right. For thisto work, Q and A must agree up front on a date D bywhich the correctness of A’s answer will be determined,and X must hold money from Q and A in escrow.X, who is generally competent but not omniscient,lacks the resources necessary to determine whether A’sanswer is correct. X must outsource this task, demandingaccompanying evidence sufficient to convince X beyondreasonable doubt. X cannot outsource this task to A,who has a clear incentive to claim her answer is correct.However, provided the question is one for which Q willbe able to verify the accuracy of any answer received, Xcan reasonably outsource this task to Q . At the start ofthe transaction, X thus demands from Q an additionaldeposit that will be returned to Q if, by the agreed upondate D, Q informs X of the accuracy (or inaccuracy) ofany answer received, providing enough evidence to con-vince X beyond reasonable doubt.With Q and X thus responsible for determining thecorrectness of A’s answer, A will rationally mistrust thegame and decline to play if Q or X can make moneyfrom A’s loss. X must therefore make the same amountwhether A’s answer turns out to be right or wrong, and Qmust pay the same amount whether A’s answer turns outto be right or wrong. Q, who consults doctors, lawyers,accountants, mechanics, and other experts when needed,is familiar with the concept of paying the same amountwhether A’s answer turns out to be right or wrong.With X responsible for determining whether Q gets herdeposit back, Q will rationally mistrust the game and de-cline to play if X can make more money by denying thesufficiency of the evidence Q provides. X must there-fore agree up front to take a fixed fee. The amount Xmakes cannot depend on whether X approves or deniesQ’s claim, nor on whether Q claims A’s answer is rightor wrong.In some cases, satisfying the above constraints will re-sult in money that cannot be given to Q, A, or X. Theyagree up front to donate any such money to charity [5].Putting some simple numbers to this story may helpgive it substance. Q gives X $100, plus an additional $100deposit. A gives X $50, together with A’s answer to Q’squestion. X passes A’s answer along to Q, respecting theanonymity of both Q and A. X pockets $50 for brokeringthe transaction [6]. There are now three possibilities. • On or before date D, Q informs X that A’s answeris correct, providing sufficient evidence to convinceX beyond reasonable doubt. X returns Q’s $100 deposit. X returns A’s $50. X pays A an additional$50. • On or before date D, Q informs X that A’s answeris wrong, providing sufficient evidence to convinceX beyond reasonable doubt. X returns Q’s $100deposit. X sends $100 to charity. • Date D passes without Q getting back to X, or with-out Q providing sufficient justification to convinceX beyond reasonable doubt. X returns A’s $50. Xsends $150 to charity.A’s incentives are straightforward. A makes $50 if sheis right. A loses $50 if she is wrong. A is incentivized toenter into the transaction only if she is pretty sure she isright.X’s incentives are similarly straightforward. X gets $50whether A is right or wrong, and whether or not Q sub-mits a claim with evidence sufficient for X to approve.X, whose business depends on brokering future informa-tion exchanges, is incentivized to objectively weigh Q’sevidence, and to do whatever is necessary to ensure trans-actions proceed smoothly and efficiently.Q’s incentives are also straightforward. Q pays $100for her answer whether A is right or wrong. Q has noeconomic incentive to claim A’s answer is wrong. Q hasan economic incentive (her $100 deposit) to inform X ofthe accuracy of the answer she received, but no economicincentive [7] to lie about the accuracy of that answer. IfQ is unable to verify the accuracy of the answer received,Q does have an economic incentive (her $100 deposit) tomake something up and provide fake evidence to supportit, but this is balanced by X, who has the power to ap-prove or deny Q’s claim, and whose continuing businessinterest relies on her consistently and objectively assess-ing the quality of the evidence Q provides.An even more straightforward case is obtained if Qcan specify up front how the accuracy of A’s answer isto be determined. Suppose Q lays out a simple proce-dure, easily followed by X, for determining the accuracyof any answer Q receives. Further suppose Q, X, and Aall agree, up front, that X will follow this procedure todetermine the accuracy of whatever answer A provides.X, who no longer needs Q to eventually say whether A’sanswer is accurate, can go ahead and return Q’s $100deposit. Moreover, with Q playing no further role in theadjudication process, it is fine to give Q her money backif A turns out to be wrong. That is, if Q can specify upfront, to the satisfaction of A and X, exactly how X willdetermine the accuracy of A’s answer , then X makes $50and A stands to make or lose $50, as above, but there isno need for the charity, and Q gets all of her money backif A turns out to be wrong .This simple protocol thus interlocks the self-interestedindividuals Q, A, and X in a manner facilitating thetransfer of useful, bluntly honest information from A toQ. In this particular scenario, Q tests the answer pro-vided by A, A turns out to be correct, money changeshands as prescribed, a cure is obtained, and the qualityof many lives dramatically improves.
III. SUMMARY
This article has identified a subset of information ex-change in which both parties may benefit from the ex-change not being free. The subset identified [8] are ex-changes that can be cast as simple questions with an eas-ily agreed upon set of possible answers for which Q willeventually be able to inform X of the accuracy of anyanswer received. Clear specification of the category ofallowed possible answers before the transaction protectsQ against receiving an answer that is true but unhelpful.Restricting questions to those for which Q will eventu-ally be able to verify the accuracy of any answer receivedallows the use of an information exchange protocol, de-scribed in Section II, designed to incentivize the transferof useful, bluntly honest information from A to Q, bro-kered by X.The protocol attempts to align, with robustness suit-able for practical use, the incentives of very human buy-ers and sellers of information. The resulting alignment,although imperfect, may represent a significant improve-ment on the incentives created by other communicationchannels. The protocol does not guarantee correct an- swers, but it does make it costly for A to provide anincorrect answer. The protocol relies on a central adju-dicator X, but limits her role to approving the evidencebacking the claim eventually provided by Q, and providesa strong and clear economic motivation for X to consis-tently and objectively weigh the evidence put in front ofher. Q is motivated to ask a clear question that will elicitan answer helpful to her. A is motivated to carefully con-sider the question and provide an accurate answer withinthe specified category of valid possible answers. The pro-tocol itself is little more than a few rules, of roughly thesame level of complexity as a typical board game, thatcreate and enforce a simple incentive structure [9].We hope the paid knowledge exchange developed inthis article may supplement, in some small but usefulway, the vast trove of information that has already beenunleashed by our remarkably vibrant, mostly honest, andostensibly wondrously robust Free Information Economy.
Acknowledgments
Kn-X [10], a for-profit knowledge exchange, has been avaluable platform for exploring, refining, and testing anextended version of this protocol [11]. This work has ben-efitted from comments, suggestions, and criticisms leviedby a diverse set of friends, colleagues, and critics [12]. [1] The words “information” and “knowledge” are used in-terchangeably in this article.[2] A fanciful example involving two entrepreneurial scien-tists is convenient for the purpose of this article. Realworld usage is not limited to science. The informationexchange protocol described here should be useful to lawenforcement (bridging the chasm between free tips andtips worth large bounties), governments, regulators, busi-nesses, and individuals, in many ways.[3] The difference between “honest” and “bluntly honest”is not worth putting too fine a point on, but it is per-haps worth a quick note. If a friend comes to you with asilly business idea and you express skeptical support, youare being honest. You are being bluntly honest if, aftersome due diligence, you tell him you think the proba-bility he closes up shop within two years is 98%, andyou are willing to stake some money on that. If a col-league or employer asks you a question and you providea direct answer, glossing over some complications thatcan be viewed as tangential to the specific question youhave been asked, you are being honest. If you also note,in detail, some related, uncomfortable fact you feel isimportant, knowing that doing so may jeopardize yourrelationship, you are bluntly honest.[4] If A has the possibility of reward but no possibility ofpenalty (roughly the incentives applying to those cur-rently managing your money), A has a clear economicincentive to enter into whatever transactions she can,whether or not she knows anything.[5] “Charity” can be anyone other than Q, A, and X. [6] Simple numbers are chosen for this exposition. The equal-ity of Q’s deposit and the amount Q is willing to pay isnot necessary, but it is of the right scale: the deposit mustbe large enough to get Q to inform X of the accuracy ofA’s answer, while not being so large it unduly deters Qfrom entering into the transaction in the first place. Ofthe $100 Q is willing to pay, the split of $50 to A and$50 to X is one of similar expositional convenience.[7] Here and elsewhere, “economic incentive” refers to amonetary incentive within the context of the paid infor-mation exchange protocol developed in this article. Veryreal economic incentives external to this protocol maylead A to provide an intentionally incorrect answer, andmay lead Q to lie about the accuracy of the answer(s) re-ceived. The information exchange protocol developed inthis article attempts to mitigate these external economicincentives. The protocol does not eliminate them.[8] Many types of information transfer obviously lie outsidethe scope of this protocol.[9] For most board games, reading the directions is insuffi-cient. You need to play a few times.[10] http://Kn-X.comhttp://Kn-X.com