Automated Justice

Introduction

Human conceptualization and institutionalization of Justice has always been predicated on devising mechanisms that could ensure and enforce neutrality. The concept of judicial neutrality implies that decision-making agents in the legal systems (judges, juries etc) must show no political or cultural bias when reaching verdicts. Whether such neutrality is possible in systems that rely on human judgment is a topic of contentious debate.

Decentralized Justice

In the last decade, a new technological paradigm entered the discourse, blurring the lines between software, law and enforcement: smart contracts. With their self-executing nature, smart contracts catalyzed viral narratives a la “code is law” and established itself as a digital mutation in the trajectory of legal scholarship. The evolutive nature of immutable, impersonal self-execution caught the attention of many scholars as a significant turning point (a revolution?).

On the flip side of deterministic self-execution, lies contextual decoupling where, given the satisfaction of required inputs, a regular user and a hacker are treated with the same indifference by smart contracts. As hackers took home billions of dollars in thousands of exploits, web3 developers and designers started to recognize the value of human interpretation and judgement.

In 2017, Kleros’ whitepaper was released (here’s the 2019 version), describing a game theoretic framework that enabled decentralized courts as a service for web3 use-cases which required human judgment and prediction. In order to incorporate this into on-chain systems, Kleros Court is structured as a Schelling-point-based coordination game, where randomly selected token holders are drawn as jurors to vote on the case at hand based on a set of rules and evidence. Incoherent and incorrect rulings are kept at bay thanks to a robust appeal system and crypto-economic incentives, and the majority vote of the jury becomes the ruling of the case.

Since then Kleros found many applications in the space such as DeFi insurance claim arbitration, eCommerce dispute resolution, DAO governance, identity verification, data curation and social recovery of assets. An overview of the applications can be found here.

Autonomous Inference

Simultaneously, the generative AI wave has also reached the shores of legal tech, causing quite a stir already. By radically democratizing computational reasoning, costly legal processes which were conventionally beyond the reach of the underprivileged, suddenly become possible to automate. As in every other field where AI is anticipated to provide significant value, there are incumbents who perceive the advancements as an existential threat.

Regardless of such tidal oscillations, there are conceptual convergences which cannot be ignored, between justice, neutrality and access. Independent researchers and developers are coming together to generate empirical knowledge around speculative applications.

In the meantime, Web3 is scaling beyond its chronic limitations thanks to moonmath and as AI takes center stage in shaping technological imaginaries. It is within this intersectionality that we are building Giza. By enabling the transpilation of machine learning languages into a verifiable smart contract language, Giza makes it possible for (simple) ML models to come on-chain and profoundly augment the capabilities of smart contracts.

At the heart of these capabilities lie Orion, an open-source ONNX runtime built in Cairo 1.0. ONNX Runtime is a cross-platform machine-learning model accelerator, with a flexible interface for integrating hardware-specific libraries. ONNX Runtime can be used with models from PyTorch, Tensorflow/Keras, TFLite, scikit-learn, and other frameworks. By bridging ONNX to Cairo, Orion provides a runtime implementation for verifiable ML model inferences using STARKs.

With the ability to verifiably contextualize, classify and relate, a new paradigm of smart contract engineering is underway.

Announcing the partnership

At the intersection of these groundbreaking projects we see the future iterations of “decentralized justice” as well as plenty of compelling use-cases. Underlying the convergence between automated intelligence and decentralized justice, lies the ability to scale para-legal and judicial processes and constitute a force for fairness, mitigating human bias where possible. Making smart contracts more intelligent and context-aware could potentially expand the design space and application areas for decentralized arbitration.

To explore this intersectional design space, we are excited to announce the formation of a partnership with Kleros towards tackling collective research, protocol integrations, open experimentation, and publishing.

Let’s take a quick look at tangible research questions which have motivated this alliance.

Slashing Disputes

Since blockchains are pseudonymous, applications built on top of them frequently require additional trust mechanisms to ensure the integrity of their functionality. This problem is generally referred to as the Nothing-at-Stake problem and has attracted significant amount of mechanism design research. Staking, the process of locking up digital currency, is now the “industry standard” for disincentivizing malicious behavior on decentralized networks. But for use cases that involve nuanced interpretation and are difficult to quantify, processes like decentralized arbitration are required.

As Giza is set to roll out a protocol for serving ML smart contracts, establishing a trust layer that signals the integrity of these models to the end users becomes relevant. In the case where staking is incorporated to provide such signal, Giza must establish a reliable mechanism for flagging and slashing in order to constitute sufficient disincentives for byzantine behavior.

This partnership will explore the suitability of decentralized arbitration towards facilitating a performant curation market for decentralized AI.

Proof of Humanity

Another canonical challenge of decentralized socio-technics is the Sybil attack vector, whereby permissionless networks are exposed to theoretically infinite duplicate identities, limiting the social and economic use cases significantly. Various approaches have been developed to render duplicate accounts unfeasible or costly, such as biometric authentication or web of trust.

Kleros has pioneered a new solution by deploying their decentralized court system towards the authentication of user-submitted videos where jurors make bets on whether a certain submission is fake or from a real human being.

On-chain ML can significantly increase efficiency and reduce the costs of human judgment, thereby enabling higher scales for the deployment of social authentication layers. By incorporating auxiliary data sources such as wallet transaction history and image recognition, ML models can provide scores that can inform the decision-making of human agents where they are relevant.

Decentralized Alignment and Control

Control and alignment are extremely relevant as self-improving machines play ever more critical roles in high-stakes systems and infrastructures. So far prominent AI companies have pleaded for the trust of regulators, citing their responsible approaches and their risk awareness. However there is dire need for substantial open experimentation around tangible propositions for mechanisms of alignment and control.

Proofs of concepts integrating Giza’s ML Smart Contracts with Kleros’ Decentralized Courts, can allow us the perfect sandbox for mechanistic experimentation around alternative design patterns which incorporate human systems to regulate machine outputs and decisions. This is a field that profoundly excites both teams and is an area where further research and publications are intended.

Conclusion

Institutions that are not subject to creative destruction tend to become irrelevant in good cases and pathological in worse ones. Therefore viability of institutional systems depends on adaptivity, therefore the capability to metabolize new knowledge.

We are deeply excited to make this announcement public as we believe decentralized ML and coordination games constitute such an evolutionary ground for social systems at large. In this journey of open speculation and experimentation, we could not be more grateful to have partners with exceptional and synergistic insights and experience.

To take part, connect with us on our channels for updates and join our Discord community.