Profile cover photo
Profile photo
Steven Glover

Post has attachment

Post has shared content
Metrics as a basis for a democracy of behavior
Lately I've been contemplating the governance ramifications of accurate data tracking for players of online games and more broadly the implications for large online societies. 

Online games can be seen as hyper-local forms of governance.  They legislate new policies, execute those policies (often enforced by code) and they have basic judicial systems to handle edge cases.  Most existing games are run by enlightened corporate dictators.  Citizen action is limited to public (often censored) discussion via public discussion boards or in game channels.  Unlike the real world, Citizens may choose at any point to cease participating in the government at the cost of losing access to their in-game assets.  There are minor variations such as Eve's purely advisory council of players, but the general patterns has held rather steady. 

In the past five years, we've seen the wide spread adoption of a major disruptor in this governance status quo:  detailed metrics and A/B testing. 

Within a game, ever single user event can be mined, correlated with historical data and cross checked again previous predictions.  Our technology for pulling patterns out of the maelstrom of big data is only growing more capable. 

Cultures are messy things to study.  Even a single social situation with a handful of participants may have dozens of interpretations.   In aggregate, over thousands of individual samples, clear patterns emerge.  These can be studied, reinforced and used to inform future policy. 

Democracy of behavior
Key to all of this is the concept that player behavior is more honest than player discussion.  Listening to the forums, one might imagine that users are saying to a definitive voice that the selling of an item for cash is utterly unacceptable.  Yet the data shows that 90% of players both purchase the item and are happy with their purchases.  In essence, players 'vote' with their behavior.  

In theory this helps avoid some of the pitfalls of current governments, particularly the tendency for rational scientific evidence to be dismissed in the face of politically powerful demagoguery or silent backroom deals. 

A 'ballot' then becomes a series of behavioral tests performed on statistically significant populations of users.   If you want to know if a particular policy is effective, you implement both A) the policy and B) an alternative to the policy.  Then you see which one works better as recorded by certain key metrics. 

How you spend your time and energy due to your contextual and internal psychological tendencies yield your ultimate vote. 

*Legal architectures that embrace regular change* 
Essential to all this is a low cost of implementing new laws and variations on those laws:  Digital laws work well in our virtual realities since well written code is quite possible to amend or refactor and the spread across millions of people.  There is an entire field of research around the study of architectures and patterns that enable flexible updating and iterating upon functioning legal code. 

If Facebook can post new laws twice a day, there is no reason why an agile online government cannot do the same. 

Legislative data priesthood
Statistics are elitist.  Understanding and interpreting them requires a technical expertise that is uncommon in the everyman.  You need to be comfortable with manipulating numbers and at the same time think holistically about root causes.  They are not something that the current unwashed masses can understand by consulting their 'truthiness'-sensitized gut.  

This barrier to entry naturally concentrate power in the hands of the few that understand how create tests, run tests and interpret the results.  As a result, one would expect early scientific democracies to headed by a near dictatorial data priesthood. 

Areas of corruption
As with any form of governance, power will corrupt those in charge.  This is inevitable.  Another area of research is identifying areas of corruption and creating systems that guard again the inevitable power grabs. 

Selection of policies: If you can propose a policy that is compared against a known poor strawman policy, you increase the chances that the numbers show your desired outcome. 

Selection of metrics: If you select metrics tailored to a particular outcome, you can generate perverse incentives or make certain good policies look bad. 

Pollution of test groups by outside sources.  By rallying public support, you can ensure that players behave in a manner to influence the outcome of a supposedly scientific test. For example, telling players to boycott a particular item would result in far lower sales for that item and make it look worse than it otherwise would. 

Bad data.  The most obvious hack is to delete, corrupt or manipulate data.   Relatively minor recording errors can result in flipped results. 

Judicial investigator
In a data-centric world, you end up with a rich and convoluted trail.  A technically competent third party isolated from the influence of the data priesthood can investigate claims of corruption. 

More importantly, they can judge if the current actions contribute to the spirit of some higher social ideals.  For example, a scientific democracy might constitutionally hold that all policy are benchmarked against the ideal of a fair and reasonable opportunity to achieve happiness.  This is a fuzzy concept, but a test that involves a politically motivated wiping of faction assets would trigger the opinion that such a policy was unconstitutional. 

Though there would remain a considerable cultural disturbance (emotional histories can never be erased), a rollback helps restore the previous economic and systemic reality. 

 Governance as a conversation
This may feel rather dystopian to some.  However, on a practical level, the process of governance is always a conversation between the policy makers and the behavior of the players.  If the policy makers screw up, the players can always trump them by leaving or by behaving in a manner that yields poor metrics.  Arguably this is vastly more power put in the hands of the masses than anything we have in place today. 

These techniques are not amendable to all problems. 

Problems that require population majorities.  Sometimes it not possible to create a meaningful sample of a policy because it requires a large mass to participate.  Testing a hypothesis around the impact of a reduction in fossil fuels on global warming may need most carbon producing countries to limit production in order to generate a meaningful signal of cause and effect. 

Irreversible policies: Some policies (such as sending deathsquads to all players that said the word 'socialist' in private chat) are difficult to test and then retract later.  Policies should be rated on not just their cost of implementation, but also the difficulty of reversal. 

Long term dynamics: Short term tests can optimize for local maxima.  Longer term tests allow for cultures to stabilize around perhaps better end results. 

Undervaluing things for which we have poor metrics:  How do you measure 'love' in a population sample?  Or the communal benefits of having a cafe down the street where the barristas know your name? There is an immense risk that those fuzzy, yet hugely meaningful attributes will be steamrolled in a quest to improve key metrics. 

Online worlds are petri dishes that we use to test scientific democracies.  Perhaps the simplest realization is that this is happening right now whether we agree with it or not. As 1) more of our lives are mediated by machines, 2) more of our everyday interactions are bound by law in the form of code and 3) policies are shaped by our aggregated behavioral data, what was once only virtual becomes meaningfully real.  It is worth starting a conversation. 
Add a comment...

Post has attachment
Wait while more posts are being loaded