1 of 24

Improving the Governance of Digital Platforms with Interactive Marketplace Experiments

2 of 24

Lying Online is Costless and Hard to Limit

No consequence to posting a lie on platforms right now. What about countermeasures?

  1. Fact-checks don’t scale
  2. Banning content → censorship
  3. Reliance on “central authorities” → trust eroding
  4. Multimodal content is overwhelming, believed despite prebunks
  5. Nudges can get overwhelming, effect disappears
  6. Idealogues don’t care
  7. Responsibility of safety placed on listeners

3 of 24

Current interventions suffer from limitations

Arms Race

Discrediting the Raters

Wrong Party Responsible

Cheaper to Produce Fake News

4 of 24

Can we (re)define the Fake News Problem?

Identify

Measure

Mitigate

Who decides what is truth?

Who is responsible to counter it?

Who should measure the impact of falsehoods?

Speakers?

Listeners?

Platforms?

Others?

Speakers?

Listeners?

Platforms?

Others?

Speakers?

Listeners?

Platforms?

Others?

5 of 24

Balance the rights of speakers and listeners on the platform

Platforms need to balance speaker’s right to free speech with listener’s right to selective hearing

Fake news matters only when it causes the listener to commit decision errors

Only speakers prioritized: spam�Only listeners prioritized: censorship

How to find an equilibrium between both?

6 of 24

Speech Markets fail to find equilibrium because decision error is an externality

Speakers (making claims)

�Free Speech

Listeners (decision error)

�Reporting Regulations

Marketplace!

Identify

Measure

Mitigate

Who decides what is truth?

Who is responsible to counter it?

Who should measure the impact of falsehoods?

externality = Side effect that is not reflected in the cost of a good.

7 of 24

Internalizing the externality so marketplaces can find equilibrium

Platforms = two-sided marketplaces.

Fake news (externality) = marketplace failure.

Marketplaces do not self-correct market failures!

Restructure incentives for production of fake news in the marketplace aka internalize the externality!��Make it costly to produce fake news.

8 of 24

Economics offers Nobel Prize winning theories to inform this work

First Fundamental Welfare Theorem: “perfect information to both sides in a complete market with perfect competition results in economic equilibrium.”

However, externalities result in market failures! → We need a mechanism that prices externalities into the cost of goods.

Pigouvian tax: make products more expensive when they produce externalities. But how expensive?

Coase Theorem: allow bargaining for goods that produce externalities for market to reach Pareto-optimal outcome.

Increasing production cost = shifting responsibility onto speaker of falsehoods.

Complete = Negligible transaction costs, every asset has a price. (cf. Adam Smith, Kenneth Arrow and Debreu)

9 of 24

Commercial Marketplace to Test these Theories

Two-sided virtual market

Selling goods online based on ads

Multiple rounds of play

Results revealed to both parties after each round

Feedback = Ratings by Buyers

Points to track score

Lower quality products are cheaper to produce, more profitable!

Buyer has to buy high quality products based on advertisements that may be misleading if product is originally low quality.

Seller has to decide what quality of product to produce (high or low). Advertisements always are of high quality.

10 of 24

Sellers can select true quality of the product accordingly determining whether ad is honest or dishonest

Buyer can decide which product to purchase after looking at all ads in the market.

11 of 24

Results Revealed after each Round

Seller chooses what product to produce and advertise them (honestly / dishonestly)

Multiple different sellers with their own sales strategies enter market

Buyer chooses up to 3 products based on advertisements

Buyer is shown true product quality and whether or not they were misled

Seller makes profits accordingly – low quality more profitable!

12 of 24

Experimental Setup

5 rounds of gameplay

Human players as consumers

Play vs. 7 “bot” producers

Producers exhibit adaptive strategies based on consumer response!

Can we increase social benefit to both sides using warrants?

May draw parallels with Robert Axelrod’s Game of Life design, except with human players.

Buyer

Honest�Seller

Opportunist�Seller

Politician�Seller

Goldfish�Seller

Cheat�Seller

Bait and Switch Seller

Reformed�Seller

13 of 24

Introducing (optional) self-certification or “warrants”

Producers choose to escrow additional amount to “back their ad claims”.

If unchallenged, amount returned to producer after round.

If challenged, adjudication based on honest or dishonest advertising

Rewards honest advertisers with an optional mechanism to back the validity of their claim!

“Warrant” Label increases credibility

Escrowed amount scales with “reach”

14 of 24

Buyer can Challenge Warranted Claims to Win Escrowed Amount

15 of 24

Seller Personas Designed to Reflect Human Tendencies

Always Honest (warrants)

Choose H + warrant

Always Cheat

Choose L

Goldfish

Choose L until 1st purchase, switch to H + warrant until 1st purchase, restart with L

Bait-and-switch

Choose H + warrant until 1st purchase, switch to L until 1st purchase, restart with H + warrant

Reformed Cheat

Choose L until 1st purchase, switch to H + warrant

Politician

Choose H + warrant until 2 purchases, switch to L until purchase, switch to 2H + warrant

Honest Opportunist

Choose H + warrant through penultimate round then L

When selling honestly, it is strictly more profitable to warrant a good / service because warranted goods sell for 10% more money.

16 of 24

Without Warrants: Cheating is Significant in the Market…

Producers try to cheat a lot… of course this is set up by our choices of producers but remember it is adaptive, so it is driven by consumer behavior.

17 of 24

…and it seems like it does pay (sales high) at first glance…

18 of 24

…but gaming the system does not work. Market collapses!

Sales diminish over 5 rounds, for every single producer strategy!

19 of 24

Introduction of Warrants Incentivizes Honesty in Marketplace

~30% better for honest sellers

~600% better for honest sellers

20 of 24

Warrants Market: Honesty Pays, although Reputation mining works

  1. Honesty pays, but opportunism has higher sales in this market.
  2. Reformed sellers are able to sell faster than other sellers, using warrants
  3. Reputation does have an effect on sales, it seems (consistent honesty has higher reward).
  4. Cheating has no reward, sales.

But this is kind of by design? Need to evaluate a more stringent tradeoff (e.g. warranted products cost more, money runs out) and see if this trend still persists.

Honest sellers receive higher sales in warrants marketplace

21 of 24

Summary

Still very early stage results (50 users, 250 rounds of gameplay)

Evidence indicating support for hypothesis that warrants can improve honesty in marketplace

Who decides what is truthful? Author can warrant all types of claims, market can adjudicate.

What about someone who has lots of money to waste on lost challenges? In current market, anyway no penalties for them! Warrants limit reach, reputation much faster.

Open Challenges:

  1. Adjudication can be imperfect in practice
  2. Warrant pricing arbitrary (we used minimum added cost), should scale with externality

Come talk to me!

swapneel@bu.edu

mehtaver.se

22 of 24

Reputation: Ratings are low for cheats

Note: scales on Y-axes are different by 8x

Honest producers are highly rated

Bait and switch leads to poor ratings!

Goldfish (forgetting honesty / lies) also rated poorly

Politicians win by mining reputation periodically!

23 of 24

Reputation: Brand changes frequent for cheats

“Brand changes” are free when reputation hits a low, reflecting how sellers simply change accounts when they cheat buyers in real life

Cheats and goldfish have most brand changes.

Restarting sales without a reputation in theory should affect the volume of sales by sellers

Warrants may offset this delay by demonstrating positive signal from any new seller without a reputation

24 of 24

Can we define the Fake News Problem?

Claims are hard to prove/disprove

Context is challenging to ascertain

Who decides what is truth?

Who counters it?

Context is private

Even Facts change over time

Lorem Ipsum