Since there are no currently active contests, we have switched Climate CoLab to read-only mode.
Learn more at
Skip navigation

Community Discussions

Seeking Help - A Game that Optimizes Climate Data for Machine Learning Based Forecasting Systems

Share conversation: Share via:

Lucas Arndt

Nov 21, 2019


1 |
Share via:

Hello Climate CoLab,

Let me start by saying that I have been designing games for a decade. I recently graduated with a BA in Practical Economics, and my Thesis Project was a mission to design a prediction game that, when played, optimizes the data acquisition process for said prediction data to be used in machine learning based forecasting systems. I want to apply this successful thesis project to applications that attempt to create better data that can be used to forecast and plan for environmentally based events.

I am planning to playtest a game where players make predictions about various climate-related aspects. For instance: accrued storm damages in particular regions, the average temperature in particular regions, inches of rainfall, CO2 emissions in particular regions, etc... Anything quantitative and useful to the community could have a game centered around it.

The game I have designed must be tested in order to determine the best scaling of the developed mechanics. The optimization of these mechanics would lead to an optimization of accrued data from players interested in competitive predictions that ultimately make a difference to the world.

In short, I need help creating a barebones UI that can fit around the game infrastructure that I have coded. Any information, questions, or suggestions are welcomed. Below is a more in-depth description of how this game makes better data, and why it hasn't been done yet.


The general idea is that such a game would combine & bolster the proven strength of 'wisdom of the crowd' and the analytical prowess of modern machine learning. Without getting too technical in this initial post, such games do exist but often lack design elegance or an efficient game infrastructure, thus relying heavily on machine learning. Look into 'Cindicator' as a major example of such a prediction game & forecasting system (including relatively poor game design). My design sought to optimize the game structure by creating an infrastructure that gathered more specific, quantitative data per player without sacrificing playability, player welfare (rewards yielded by effort), or total elicited effort (how much effort is put into the combined predictions of all players).


Current prediction games provide a binary option as follows:

Do you think that x unit for y 'market' will surpass quantity next fiscal year?

For instance: Do you think that the average temperature will surpass 90 degrees F next fiscal year?

Rather nonspecific player inputs - you either think 'yes' or 'no.'


My game asks for player-provided specifics as follows:

Predict for y market in the next fiscal year.

For instance: Predict the average temperature in the next fiscal year.


The data provided is more specific in my game. This is a far more complicated task than it seems, for reasons like the ones to follow:


- (note that I have developed answers to these addressed issues)

+ How does one set up a fair scoring system when it is possible that no one in the game predicts correctly, or every single player predicts correctly?

+ How does one determine what scale a specific quantitative prediction game should establish when there are virtually infinite scales from tiny fractions of units of variation to hundreds of thousands of units of variation?

+ How does one determine appropriate specificity & accuracy to deem a prediction 'correct' when there are virtually infinite predictions available?

This list goes on painstakingly.

- (note that I have developed answers to these addressed issues)


I found solutions to the issues surrounding such a game by implementing a set of automatically-scaling mechanics that guide the player through a simple set of inputs while calculating appropriate parameters to define success, failure, game-specific scale, etc...

What I accomplished was a variety of games that, when tested, would indicate ideal mechanical incentive structures that optimize the data gathered per player without sacrificing playability. I have the mechanics built, I just need to figure out how to scale them appropriately. I need to do a big playtest.

I need help developing a widely testable version that can be used on an online platform. What I need is a UI that will work on an online platform in tandem with my already-coded game. From there, the game structure can be optimized to produce the highest quality data possible which will lead to the most useful forecasts possible when implemented in data-driven forecasting engines.


Thank you for your time, comments, and help.