Written by Kevin Owocki, Co-founder of Gitcoin
Compiled by: Tia, Techub News
Goodhart's Law states that when a metric is used as a goal, it is no longer a valid measurement because the focus is on optimizing the metric rather than the original intention of setting the goal.
Implications for DAOs and public goods funding: In DAOs, and especially in public goods funding mechanisms, systems that evolve over time, use diverse and dynamic metrics, and incorporate community feedback are needed to prevent gaming and maintain effectiveness.
Balance between immutability and evolution: While the cryptocurrency community values immutability, this article advocates for a balance, suggesting that some parts of a DAO should remain immutable, but other aspects should evolve to adapt to new circumstances to avoid Goodhart’s Law.
Goodhart's Law
Goodhart’s Law is a principle in economics and statistics that states: “When a measure becomes a target, it ceases to be a good measure.”
This concept was first proposed by British economist Charles Goodhart in 1975. The reasoning behind this law involves several key ideas:
Incentive distortion: When a particular metric is chosen as a target for policy or performance measurement, it often results in changes in behavior in order to optimize for that metric. This change in behavior can lead to unintended consequences that distort the original purpose of the measure.
Overemphasis on measurable results: People tend to focus on what can be easily quantified. When one metric becomes the goal, there is a risk that other important but less quantifiable aspects will be overlooked, leading to an overly partial and potentially misleading assessment of the project.
Manipulation and gaming: Individuals or organizations may be tempted to manipulate the system to get better scores on targeted indicators, thereby neglecting to do what is truly important. For example, if test scores become the primary measure of educational success, schools may focus on teaching to the test rather than improving overall educational quality.
Reductionism: Goodhart's Law highlights the danger of reductionism in decision making and management, whereby complex systems are reduced to simple numbers or targets. This reductionism can oversimplify reality and lead to decisions that may work for the metric but are bad for the system as a whole.
The law warns against over-reliance on any single metric or indicator to make decisions, especially in complex systems where multiple factors need to be considered. It advocates for a balanced approach and emphasizes the importance of being vigilant about the ways incentives influence behavior.
Goodhart's Law and Public Goods Funding
I think this article will have a significant impact on how we fund public goods in the future.
Why? Public goods funding is not a discrete event but a series of events that occur over time.
We hope that the PGF (Public Goods Funding) round will create tremendous value in your ecosystem.
But if we consider Goodhart’s Law, what actually happens is that the creative value is lower.
In the above diagram, we see that what worked in PGF round (n) does not work in PGF round (n+1).
As measures become targeted, PGF rounds will become less effective, resulting in diminishing returns in value creation.
Goodhart’s Law applies to any mechanism, and it applies to quadratic funding contexts, badge holder review contexts, and any other public goods funding round in adaptive complex systems like the political economy of DAOs.
So what should we do?
I think the conclusion from this is that the best PGF round will be an infinitely evolving game. An evolutionary game is a game theory concept where strategies and behaviors evolve over time based on the success of previous rounds and influenced by natural selection.
As measures become goals and goals become measures, the best systems will move forward in a way that is difficult to manipulate and thus does not create diminishing returns in value creation.
In the new paradigm, our experiment will look a bit like this.
Design measure m+1
How does measure m evolve to measure m+1? We are under evolutionary pressure because the best ecosystems will keep moving forward, but how can we do that?
This is cutting-edge work, so unless you've run dozens of rounds of PGF, you probably don't know the answer yet. But we can reason about it. I think there are a few ways we can think about this.
OODA loop: Each round of PGF is an OODA (Observe, Orient, Decide, Act) loop for the mechanism designer. The designer needs to learn from each round, evaluate the effect of the measurement, and drive its evolution based on these learning results;
Algorithmic randomness: Introducing an element of randomness into the reward or evaluation process to reduce the predictability and effectiveness of game strategies. For example, random checks or audits can be used to ensure compliance without requiring the system to be completely predictable.
In RetroPGF, the voting for the metric element designed in Round 4 has a lot of randomness built into it, as well as unpredictable game mechanics such as arbitration rules and scoring formulas.
Community feedback mechanism: allows participants to report and address concerns about the integrity of community metrics or behaviors. This feedback can be used to continuously refine and improve the system.
Indicators that are difficult to game – Indicators that are difficult or expensive to game will resist (but are not immune to) Goodhart’s Law.
Diversified metrics: Rather than relying on a single metric or indicator, use a diverse set of metrics to assess performance or impact. This helps ensure that different aspects of the desired outcome are captured and reduces the risk of any one metric becoming too dominant.
Dynamic and adaptive metrics: Implement mechanisms that allow metrics to be adjusted or replaced as the system evolves. This adaptability helps prevent manipulation of static metrics and ensures that metrics continue to align with the DAO's fundamental goals. Metrics should also have balancing forces, e.g., new users vs. retained users.
Caps and rotation mechanisms: For key metrics or roles within the system, consider using caps (limits on maximum score or gain) and rotating focus between various metrics or areas. This prevents over-optimization of any single metric and encourages broader contributions across different areas. An example of this policy in practice is: a single metric can never receive more than 20% of the allocation.
I think it's a competitive landscape. If a lot of players start trying to game a metric, then its value collapses.
Immutable => Evolving
For most of the history of the crypto ecosystem, we’ve placed great value on the complete immutability of our protocols.
Usually, there’s a good reason for this. An immutable protocol is one that cannot be broken. We need this in currency protocols — currency protocols, according to cryptocurrency lore, should not be subject to the whim of any one party. That’s why we have uncensorable currencies, unprintable currencies, namely Bitcoin and Ethereum.
But does everything in every protocol need to be immutable?
Public goods funding in the crypto era was born in this age of immutability, but must also transcend it to succeed.
We can have the benefits of immutability in some places (e.g., within a round, we should rely on a trusted neutral protocol where anyone can verify vote counts), while having some variability between rounds.
We must recognize that an emphasis on immutability often leads us to aim for perfect solutions and results in less iterative thinking. Public goods funding experiments can and should take the opposite approach — iterating to local maxima over time. We prefer action and courage to deliver something imperfect as quickly as possible, rather than never delivering something (theoretically) perfect.
in conclusion
Goodhart's Law was proposed by British economist Charles Goodhart in 1975, which clearly states that "when a measure becomes a target, it ceases to be a good measure." This principle emphasizes that when using specific indicators as strategic or performance goals, it often leads to the manipulation of measurable results, thereby ignoring the original intention and reducing the complexity of the system to simplified numbers, which can lead to misleading evaluations.
We believe this has huge implications for the financing of public goods in Decentralized Autonomous Organizations (DAOs), particularly mechanisms like quadratic funding and badge holder review.
We suggest that the best public goods funding (PGF) systems will benefit from being evolvable and adaptive, incorporating strategies such as diverse metrics, dynamic adjustments, and strong community feedback to remain effective and resistant to gaming.
We advocate for a balance between immutable protocols and adaptive strategies to ensure funding of public goods for long-term success.
