Should AI Scores Determine Voting Power in SciWard?

SciWard currently gives each eligible voter one vote. If we decide to give higher-scoring users more voting power, I see two options. Linear weighting gives each voter a weight equal to their AI score. Square-root weighting gives each voter a weight equal to the square root of that score. I would keep equal voting for bans and reinstatements until testing shows that either proposal leads to better decisions.

How voting can affect who gets paid

SciWard is an experimental funding platform for researchers and open-source developers. Its AI evaluates documented contributions and assigns a score expressed as an estimated fraction of world GDP. The score is a heuristic estimate of the value of the contributions. SciWard uses the original AI score directly to calculate proportional payouts from available funds for eligible recipients. Both proposals apply only to vote weight. How SciWard calculates funding allocations

SciWard’s documented process gives participants seven days to challenge a funding recommendation by identifying a specific concern. A human vote can override the AI recommendation in a disputed case. Serious misconduct can lead to a ban, and a later vote can reinstate the user. A vote can therefore affect payment by changing who is eligible. SciWard’s challenge process · Ban and reinstatement voting

The financial incentive behind an eligibility vote

Suppose a fund has USD 1,000 to distribute and two eligible recipients with equal scores. Each would receive USD 500. If one becomes ineligible and the full pool is distributed to the remaining recipient, that recipient gets USD 1,000. Their score stays the same. A decision about misconduct can change the size of someone else’s payment.

Equal votes can turn a dispute into a popularity contest. A widely used open-source project can draw many supporters, while a researcher maintaining a specialized library may struggle to find people willing to examine the evidence. Low turnout can also give a small group of regular voters disproportionate influence.

Score weighting adds a second risk. An inflated AI score could increase someone’s prospective payout and give them more influence over disputes involving collaborators. And even an accurate score for scientific or software contributions gives us little reason by itself to expect better judgment about disputed authorship or misconduct.

Linear versus square-root voting power

Let s be a voter’s nonnegative AI score, expressed as a fraction of world GDP. Suppose two eligible voters have scores of 0.0001 and 0.000001. The first score is 100 times the second.

Rule	Vote weight	Higher-scoring voter’s weight relative to the other voter’s
One person, one vote (current rule)	1	1 to 1
Linear weighting	s	100 to 1
Square-root weighting	√s	10 to 1

Under linear weighting, one higher-scoring voter in this example can tie 100 voters who each have the lower score. The square root brings that number down to 10. In general, if one score is r times another positive score, the ratio of voting power is r under linear weighting and √r under square-root weighting. Under either proposal, an eligible voter with a score of zero has zero voting power.

How square-root weighting relates to quadratic voting

In quadratic voting, acquiring one, two, or three votes costs one, four, or nine voice credits, respectively. A budget of B credits therefore buys up to √B votes on a single decision, subject to the ballot rules. The SciWard proposal uses the same mathematical relationship to assign a ballot’s weight automatically from the voter’s AI score. In quadratic voting, the participant chooses how many credits to spend to express the strength of a preference. Lalley and Weyl’s paper on quadratic voting

What SciWard should test before changing the rule

• Make disputes reviewable. Require a specific allegation and supporting evidence, give the affected user an opportunity to respond, record the reason for the decision, and provide an appeal heard by different reviewers.
• Address financial conflicts and low turnout. Require parties and close collaborators to recuse themselves. Set a quorum based on the number of unconflicted people voting. Refer the case for independent review if too few can participate. NIH rules on conflicts of interest in grant peer review
• Test both proposals against equal voting. Recount the same mock ballots under all three rules to see where outcomes change. Then run mock rounds in which voters know their weights in advance, since a rule can change who turns up and how they vote. Include low-turnout cases, allegations against highly scored users, and legitimate work by lesser-known contributors. SciWard’s testing and evidence plan

Ask independent reviewers to assess the evidence and flag cases they cannot resolve confidently. Track missed abuse, wrongful exclusions, legitimate payments delayed, and how often a handful of voters decide the outcome. I would change the voting rule only if one of the two proposals consistently improves those decisions under realistic conditions. A smaller gap in voting power makes square-root weighting less concentrated than linear weighting for any given pair of voters. The test is whether it helps the right people get paid. SciWard’s invitation to independent reviewers