CivicTwin – Rethinking Representation in the AI Era
Bridging the gap between citizens and representatives
Preface
As a preface to this post, I would like to note that my background is not in politics. Whilst I have a keen interest in political systems from an organisational engineering perspective, I do not claim to be an expert. This piece explores one possible—albeit highly improbable—evolution of democratic governance.
The future remains deeply uncertain, artificial intelligence is advancing at an extraordinary pace, and political systems are designed and have evolved to withstand disruptions in industry and society at large. Consequently, the likelihood of anything resembling this scenario materialising is minimal. Nevertheless, it presents an intriguing hypothetical.
If you have any insights, critiques, or alternative perspectives, please leave a comment.
The Current Limitations of Democracy
Since the advent of democracy in ancient Athens, representative governments have been plagued by a myriad of challenges including corruption, bureaucratic inefficiency, unrepresentative decision making, weak accountability mechanisms, over-reliance on partisan politics, and susceptibility to demagoguery.
Over time, political systems have evolved to counter these threats. This adaptation is typically slow, however this historically was a feature rather than a bug. Slow adaptation allows political systems to be disconnected from the more radical shifts in other spheres of a country’s operations.
Eventual evolution is, however, necessary to ensure governments do not become antiquated. There will always be threats to democracy, and they will continue to increase in scale and complexity in tandem with the world at large. If countries today were governed using the same mechanisms as ancient Athens, the world would be much more chaotic and dysfunctional.
Democracies today, face threats from all directions – measurably increasing polarisation (Iyengar & Westwood, 2015), erosion of public trust due to misinformation, voter suppression, and the growing influence of populist and authoritarian movements.
CivicTwin: An Aggregator of Human Values
We propose a decentralised system for aggregating individual ethical frameworks to address the limitations of current political systems. This system, CivicTwin, functions as a predictive model for individual legislative preferences, built using a combination of identity verification, self-reported data, and machine-learning inference.
Upon enrolment, users undergo a verification process to authenticate their identity and residency. Following this, they complete a structured questionnaire designed to extract a high-dimensional representation of their ethical and political values. From this input, a digital twin is instantiated—an algorithmic representation of the user’s ideological perspectives. This model is trained to approximate the user's voting behaviour by retroactively predicting the user’s responses to historical legislation. Users engage in a feedback loop by reviewing and correcting these predictions, allowing for iterative refinement of the underlying model.
To enhance predictive fidelity, the system incorporates auxiliary demographic and socioeconomic inputs, including income level, number of dependents, marital status, and education level. These parameters serve as additional priors, refining the digital twin’s ability to infer voting preferences across legislative contexts.
The questions used for backtesting would be randomly selected from a pool of questions. This would mean that, for any given question, a set of the total user base has responded, which provides a wide breadth of data for the model. This would allow the model to anticipate how a user may respond to any given question based on how other similar users responded (the same way Spotify or Netflix can make recommendations based on the activity of similar users).
Any proposed legislation introduced in parliament would be uploaded to the platform's database. For each user, the system would assess the legislation against their digital twin, predicting how the user would vote given sufficient time to analyse and comprehend the details of the proposal. A rationale for the predicted vote would be generated, explaining the model’s decision-making process. Users retain the ability to override the prediction if they believe the model has inaccurately represented their stance, with the correction being fed back into the system to refine and recalibrate the digital twin.
Users would be provided with analytics on the predictive accuracy of their digital twin across various policy issues. They could selectively delegate their vote to the platform on specific matters (e.g. where the digital twin has demonstrated a robust track record at matching the user’s actual preferences).
Ultimately, the platform is not designed to replicate the precise political positions of each individual but instead functions as an axon, increasing the efficiency of communication between the user and their political representative by leveraging advanced predictive modelling.
I think a good analogy for this system is autonomous vehicles. It is commonly argued that, to make roads safer, self-driving vehicles don’t need to be perfect; they only need to outperform the average human driver. Similarly, CivicTwin does not need to perfectly represent every user. Its goal is simply to provide more accurate data to political representatives than what currently exists, thereby enhancing the effectiveness of the political system.
Additional Features:
Users may delegate proxy votes to digital instantiations of prominent political figures, enabling these representations to vote on their behalf for specific issues. For instance, a user might designate a digital representation of Bernie Sanders to cast votes on economic matters, and a digital Angela Merkel for immigration-related issues.
Each piece of legislation uploaded to the platform would be accompanied by an executive summary, outlining the direct impact of the proposal on the user based on the personal data provided. The summary would also include an analysis of expected positions of major political parties, accompanied by rationales for their positions. Additionally, relevant news articles would be provided to offer context and further insight.
To preserve the integrity of the voting process, a chat feature will not be included on the platform. Engaging in real-time discussions could lead to undue influence, coercion, or the spread of misinformation, undermining the personal and independent nature of voting.
Correcting Voter-Representative Disconnect
The primary issue addressed by CivicTwin is the misalignment between political representatives and their constituents. The incentive structures guiding representatives are often opaque and complex. For most politicians, achieving longevity in office requires more than simply delivering outcomes to voters—it also necessitates aligning with party interests. As political polarisation deepens, representatives are increasingly hesitant to defy party lines. As a result, politicians are increasingly focused on framing decisions in a way that appeases voters rather than genuinely representing their interests.
CivicTwin seeks to realign the interests of representatives with the actual needs of their constituents, ensuring that voter intent is accurately represented in parliamentary decision-making.
The Path to CivicTwin
A shift of this magnitude within a democratic system would likely originate from an independent non-incumbent representative. Established political parties and incumbent independents are unlikely to adopt and integrate such a system due to the reputational risks associated with its disruptive nature.
The first candidate to successfully leverage this innovation would likely represent an electorate with high technological literacy. For such a candidate to secure electoral support, the majority of voters would need to have a foundational understanding of artificial intelligence, large language models, and have confidence that the system would effectively represent and advocate for their values.
Should the system's viability be demonstrated, it is probable that other incumbents would follow suit. However, major political parties would remain hesitant in the medium-term, as adopting such a system undermines their very raison d'être.
In the long term, even established parties may be compelled to adopt a system akin to CivicTwin in response to mounting public pressure for more accurate and transparent representation.
Previous Attempts
Historically, there have been two notable attempts at implementing direct democratic systems in Australia, Online Direct Democracy – (Empowering the People!) and Flux.
Online Direct Democracy – (Empowering the People!) operated from 2007 to 2019. The core concept was that for each bill presented in parliament, an online poll would be conducted, allowing individuals on the Australian electoral roll to vote on each bill. The elected Member of Parliament (MP) would then cast their vote based on the majority outcome (greater than 55%). In cases where no majority was reached, the MP would abstain from voting.
The reasons for the party’s failure to gain substantial traction remain unclear, though several factors likely contributed. These may have included limited public awareness of the party and general resistance to adopting new processes and technologies, particularly when it comes to something as critical as political representation.
A fundamental issue with the Online Direct Democracy – (Empowering the People!) model lies in its disregard for the geographical boundaries inherent to the existing political system. Senators are elected to represent specific states and territories, yet this platform allowed any Australian citizen, regardless of their electorate, to vote on legislation through the platform.
CivicTwin, by contrast, recognises the value of the existing political infrastructure, particularly the separation of representation based on geographic constituencies. This system has proven to be robust, ensuring that political decisions reflect local interests. CivicTwin works within these existing frameworks, bridging the gap between voters and MPs to facilitate more accurate representation.
Another notable initiative in this space was Flux, which proposed a form of liquid democracy. The key difference between Flux and Online Direct Democracy was that Flux allowed users to forgo votes on one bill to then use on future legislation. Additionally, users could delegate their votes to others, enabling them to specialise and vote only on matters they understood.
This model introduces a significant issue: by allowing users to delegate votes, it transforms the platform into a market. Humans have a tendency to try to exploit markets, and this could, in the long-term, potentially undermine the integrity of the platform. Like Online Direct Democracy, Flux struggled to maintain membership and was eventually de-registered in 2022.
CivicTwin’s Differentiation
Even if one of these platforms secured enough votes to win an election, they would likely struggle to make decisions that align with the long-term interests of constituents. Humans are quite short-term oriented, and politicians can often persuade the public to support policies that may initially appear to be misaligned with their values but ultimately serve the broader population's long-term welfare.
CivicTwin faces a similar challenge. To address this, the platform would incorporate two key features:
Expert Analysis & Insights – For highly contentious issues, CivicTwin could present expert opinions and in-depth analysis from credible sources, helping users make more informed decisions.
Demographic Voting Transparency – Users could view how different demographic groups will likely vote on specific bills, along with their justifications, fostering a deeper understanding of the broader societal impacts.
Assumptions
The successful implementation of CivicTwin relies on several critical assumptions:
Public Interest, Engagement, and Trust – The platform must garner widespread public support and confidence in both its purpose and execution.
Voters will only accept AI-driven decision-making if it is transparent, explainable, and demonstrably accurate.
Adoption of a system like CivicTwin is unlikely unless there is substantial evidence that the existing political system is failing to produce optimal governance outcomes.
Model Neutrality & Bias Mitigation – The underlying models powering CivicTwin must be rigorously tested to ensure they do not systematically favour specific ideologies or values. Any perception of bias could undermine trust and lead to rejection of the system.
Stability & Adaptability of Individual Values – Whilst voters’ core values remain relatively stable (especially later in life), CivicTwin must incorporate mechanisms to detect and adapt to shifts in user beliefs over time to ensure it remains representative of the electorate.
Accurate Representation Improves Governance – The most significant assumption is that greater democratic representation leads to better governance. While democracies are generally more stable than autocratic regimes long-term due to broad representation, this principle may not scale indefinitely.
Traditional political systems rely on compromise and negotiation, with politicians making trade-offs that they later justify to their electorate.
A system that enables perfectly representative democracy could lead to increased legislative gridlock, heightened political conflict, and the rise of short-term populist policies, rather than more effective governance.
Limitations
1. Public Misunderstanding of Government Policy – Many government actions, such as taxation, interest rate adjustments, and military spending, are essential for economic and social stability but are often misunderstood or distrusted by the public. A system driven by direct representation may struggle to account for policies that are unpopular in the short-term but necessary for long-term governance.
2. Adoption & Representation Challenges – For CivicTwin to be truly representative, a critical mass of local constituents would need to join. In the early stages, adoption is likely to be low, leading to skewed or incomplete public sentiment analysis, reducing the system’s accuracy in reflecting voter preferences.
3. Privacy & Data Security Risks – CivicTwin would require users to verify their identity through official documents such as driver's licenses and proof of address to prevent duplicate accounts. However, this would result in a centralised repository of highly sensitive personal data, creating a potential target for misuse or data breaches.
4. Risk of Deepening Political Polarisation – While CivicTwin aims to improve representation, it could inadvertently exacerbate political divisions rather than bridge them. Politicians are skilled at presenting policy trade-offs in ways that appeal to multiple factions, whereas a fully transparent, direct representation system may entrench ideological divides by reinforcing existing partisan preferences.
5. Cybersecurity & System Integrity – Any political technology platform must contend with significant cybersecurity risks, including hacking, data manipulation, and foreign interference. Robust security protocols, independent oversight bodies, and multi-layered verification mechanisms would be required to ensure trust and system integrity before widespread public adoption.
Interesting concept and one worth exploring. I agree with the fundamental principle, that our current conception of democracy, which relies on elected representatives, is facing extraordinary challenges. There is a misalignment between reps and the demos that needs to be solved.
In matters of democracy, however, I tend to favor old-fashioned solutions. My preference would be to return to sortition; instead of elected representatives, a number of individuals would be selected at random to pass laws. This ensures cognitive diversity and reduces the misalignment problem (if you don't have campaigns at all, no need for "campaign finance" laws.")
The challenge, of course, is writing laws. A large group of people cannot collaboratively write law. I have a potential solution for this though, using randomly "interest panels" that seek the outside assistance of experts. It's here, however, where I think AI could play a huge role. Acting as research experts/assistants and helping concerned citizens "red team" proposals before they go to the sortition-selected legislature.
That way, humans are also in charge but AI broadens the capacity of what these humans can do. You might be interested in my essay on this: https://www.lianeon.org/p/imagining-our-martian-government