Genetic Algorithms and Participatory Democracy

Laurence Leff, Western Illinois University

Usage Details

Our research team has developed a simulation of a genetic algorithm optimizing a tax code and budget based upon ratings given by subgroups in the demos. At each step, the participants rate members of the candidate pool of budgets 1 to 10. The genetic algorithm sums the ratings on each budget to determine its fitness. The highest rated budgets are crossed over. That is part of one budget, e. g. the tax rates, is combined with the other part of another budget, e. g., its spending allocations. The probability that two budgets will be paired or crossed over is proportional to the sum of the ratings by all the subgroup, as weighted by the number of voters they represent.

Some of the populations will be special interests. As an example, they may be the employees and stockholders of a defense contractor seeking to increase the amount of money spent on defense. The defense contractor would have money for supercomputers to determine a set of votes by these followers that would increase the resulting defense spending; the supercomputer may find that giving a high rating to budgets with relatively low allocations on defense but high allocations for farm programs, lead to a higher final budget for defense than just voting sincerely. That is, all the followers simply rate candidate budgets with high defense spending and low taxes on themselves higher than others budgets.

Two prior simple simulations found that the strategic players get a less favorable outcome than if they simply voted sincerely. The question for this Blue Waters research is simply: If a special interest has access to one of the most power computers on the planet, can they manipulate a genetic algorithm to get a more favorable result than if they simply voted sincerely?