National Alliance for Physician Competence Discovery Workshop


  • After the morning conversation the participants divided themselves into three groups. While a third of the group listened to Kirsten Moy speak about non-profits going to scale, a third of the group listened to Marko Rodriguez speak about distributed decision making, and the other third of the group listened to Bill Rouse speak about complexity in healthcare.
  • Download: Marko's Powerpoint slides.

Collective Decision Making Systems and Prediction Markets

Marko Rodriguez

I’m going to talk about Dynamically Distributed Democracy (DDD), which is an accurate way of determining what the collective’s values are.  It is an example of collective decision making using computers to support the process.

Collective decision making with computer support started in the 1980s with something called GDSS - Group Decision Support Systems.  They wanted to improve on the limitations of the hierarchy to remedy some of the deficiencies in face to face meetings. 

In 2000 Murray Turoff developed the concept of Social Decision Support Systems.  His ideas was to develop a system so that instead of voting every 4 years, how would you make it so everyone in a society or a community could constantly participate?

Another concept is called “collaborative discourse,” which is for people to see the flow of argumentation.  The intent was to try to come to consensus prior to voting.

My work covers all these, under the broad title of Collective Decision Making System (CDMS).  How can we use computers to help humans make decisions?

This presentation complements my paper, “A Survey of Web-based Collective Decision Making Systems,” co-authored with Jennifer H. Watkins.

marko rodriguez
Ranking Artifacts
A key concept is how we rank the value of artifacts.  Google uses a human-driven ranking algorithm.  Web pages are linked, which forms a network.  Google has an algorithm that is able to figure out the centrality of nodes in the network.  Nodes that are more central are ranked higher, and they are based on human value judgments.  Most of the work of Google is done by the humans who build the network by linking web pages together.  For every web page there is a ranking value, which is determined by the humans who are making the links.

How do you categorize?  How do you structure an environment so that humans will get you to solutions? Sometimes you want collaboration, sometimes competition, sometimes categorization. 

Flikr and Delicious are called “folksonomies.”  There is a taxonomy, from specific to general.  Folksonomies are where people label things any way they want.  Patterns of labeling emerge because people tend to converge on labeling conventions.

There are some key concepts pertaining to collaboration.  We all know the Wiki.  Open source software is a huge movement.  Anyone can help develop software.  If the code is good it gets put into the next release.  The software to support open source is good at distributing tasks to people who are good at that function.

Q:  Can you meld Flikr and Wiki?  Can you use a Flikr system to categorize and then develop in a wiki.  A.  Yes, definitely.

Taxonomy Chart Description
The last page of the paper is a chart that shows a taxonomy of these systems.  On the left it shows “problem space.”  You use different tools to solve different kinds of problems. 

Stigmergetic is blackboard communication.  The use of the environment to leave messages.

Dynamically Distributed Democracy
The question I’m interested in for Dynamically Distributed Democracy is, How you do get a decision in a large group?  What would happen if you wanted everyone to have a say in the system and participate in the evolution of the system?  If the group is constantly making decisions, everyone in the country can’t be there at all times to participate. 

You can have asynchronous voting - a time window.

In our current government, if you ignore your opportunity to vote you lose your chance to vote. 

The concept of Social Compression relies on some people being able to represent others.  Any subset of the whole can be a model of the whole.

The concept of Social Compression relies on some people being able to represent others.  Any subset of the whole can be a model of the whole.

Direct Democracy Slide
This simulation of direct democracy shows that if not everyone is participating then there is error in the vote because not every one’s views are represented in the final result.  So we’re looking for a solution to this shortcoming.

Dynamically Distributed Democracy Slide
In a trust-based social network where the network propagates vote power, in case I’m not present I trust others to use my voting power.  The premise is that socially close individuals can be similar in values.

This slide shows a simulation of Dynamically Distributed Democracy.  Everyone starts with one energy step.  The goal is to move all the energy in the system.  Because of the weighting mechanism there is not error in the system.

In direct democracy, if 100% of the population participates there is no error.  But as participation wanes, there is an increase in the error.

With dynamic distribution, if 50% or 20% participate you still get a very accurate reflection of what everyone perceives.  So you don’t need everyone to participate, you need to weight the population effectively. 

DDD in the Real World
Open governance is starting to emerge in society.  If you want anyone to be able to produce policy, you need infrastructure.  Wikis are good infrastructure for this.  Implementation of policy is based on open source principles.  This is a techo-dream system of how a government could be run.

“Smartocracy” is a system that implements this.

Prediction Markets

Information Markets are a general category, and prediction markets are an example.

Robin Hansen of George Mason University developed a prediction market system in response to 9/11.  He suggested that people participate to help make the predictions about future terrorist acts.  It’s a very accurate mechanism, but it was shut down due to social concerns.

The idea of the prediction market is to aggregate individual knowledge into global knowledge.

The premise is that as complexity increases in a system, a single individual cannot understand that entire system.  The idea of the prediction market is to aggregate individual knowledge into global knowledge.  Our brains are engineered for pattern matching, but when we lack language we cannot match.  So we rely on aggregating the knowledge of many to match the complexity of the system with an understanding of the system.

In a prediction market you don’t vote on your preference, you vote on what you think will happen.   If you have knowledge you can make money with that knowledge.

In the US, prediction markets are illegal. 

For a prediction market to work, you need a date, or a specific action that will or won't happen:  Specific outcomes are specified at specific dates. 
You also need:
- a diversified population. 
- a collective of self-interested traders.
- a market and a payout mechanism.

Yahoo uses an internal prediction market to look at new products.

Q:  Could we predict the receptivity to the Good Medical Practice document?
A:  Yes.

We can also use something like this to see which scenario we developed in the first Summity the physician community thinks will really come to pass.

… in the real world
Prediction markets are a useful tool for harvesting information.

Top of Page