We are living in times of unprecedented technological advancements. The pace of innovation has brought artificial intelligence (AI) out of science fiction books into our daily life. From simplifying everyday tasks such as planning a travel itinerary to accelerating scientific breakthroughs, the positive impact of AI is undeniable. But even as AI has reshaped our world, there have been increasing concerns about centralization of AI models, data privacy, and black-box decision making processes. A potential answer to these issues has emerged in decentralized AI (deAI).

While still in its infancy, deAI is a complex heterogeneous system that has many diverse parts such as GPU compute, large-language models, and inference. Unlike objective blockchain systems, where all transactions are recorded on-chain, many critical processes and interactions in deAI occur off-chain. Verifying and attributing these off-chain events on-chain within the broader network then becomes a complex task, often introducing challenges with data integrity, trust, and scalability. 

Enter intersubjective attribution, a novel mechanism where attribution of off-chain events is powered by causality graphs. 

Let’s first look at intersubjectivity: it posits that our experiences and understanding are fundamentally shaped by our interactions with others and the world around us. It's the shared understanding that allows us to communicate, empathize, and collaborate effectively.

In the context of digital systems, intersubjectivity offers a path to a more holistic form of trust. It suggests that true trust isn't just about verifying transactions or enforcing smart contracts - it's about creating shared understanding and mutual recognition among diverse entities in a network.

Meanwhile, causality graphs provide causal information for each event in a network, such as the number and identities of peers in a transaction dissemination graph, and causal dependencies with respect to other transaction invocation and commitment. This decentralized, causal information can help determine a fair order of events in a network.

The framework of intersubjectivity combined with the principles of causality helps create a robust technology that accurately and fairly attributes off-chain values on-chain. This technology is called intersubjective attribution, which is poised to bridge the gap between the dynamic interactions of different intelligences, human and AI.

DeAI applications require immense computational resources and data processing capabilities. Traditional blockchain systems where all events are recorded on-chain currently fail to provide the scalability and low latency that is crucial for the real-time and high-volume events that are characteristic of DeAI applications. 

This issue of scalability and real-time finality can be solved by intersubjective consensus, which enables more flexible and efficient event ordering. Hetu integrates intersubjective attribution into heterogeneous distributed systems, allowing DeAI to scale effectively while maintaining security.

1. Verifiable Logical Clock (VLC): At the heart of Hetu’s innovation is the verifiable logical clock. In a distributed system, event ordering is crucial but can become cumbersome if done via mechanisms that follow a global total order. An example is traditional consensus mechanisms. Hetu attaches a VLC to every event in a network for event ordering, creating a system that is more efficient. These VLCs are small and independently scalable, ensuring hyper-scalability and cost-efficiency in a network. 

2. Causality-Based Ordering: Unlike traditional consensus mechanisms that rely on total global ordering, Hetu uses partial ordering via establishing causal relationships between events. This approach is more flexible and efficient, especially for applications that can operate asynchronously or in parallel.

3. Off-chain Pre-confirmation: Hetu moves consensus off-chain and uses its causality hub to establish the order of events through the VLC system. This allows events to happen off-chain while still maintaining a verifiable order, significantly enhancing scalability and efficiency.

Hetu addresses key challenges in leveraging decentralized systems for DeAI and other applications that have a high volume of events or interactions, handle a large volume of data and offer real-time finality. Let’s explore the benefits Hetu provides for these applications. 

Hetu’s layered structural design separates global on-chain consensus from local off-chain consensus environments. This design provides highly elastic computation and verification capabilities at the local level, significantly expanding the potential for application across heterogeneous networks. In doing so, Hetu is a better choice for applications that need low latency and real-time finality, something blockchain networks are currently struggling to address.

Traditional blockchain networks often struggle with slow transaction confirmation times. Hetu solves this by using computational verification and event timestamping to confirm off-chain non-global consensus events. This approach allows for rapid finality and ordering without the need to wait for global on-chain confirmation. Moreover, this ordering is temporally comparability across various scenarios, ensuring consistency in heterogeneous systems.

Incentivizing participation in decentralized distributed networks is crucial. Hetu’s causality hub enables rapid verification and computation of proof data from different decentralized networks. This features allows non-transferable contributions to be converted into transferable credits or tokens at minimal cost, thereby supporting large-scale incentive systems.

By addressing some key challenges of decentralized distributed systems, Hetu paves the way for a new era of hyper-scalable decentralized applications that can operate at real-time speed.

A key differentiator for Hetu is its use of off-chain consensus and time-stamping for decentralized microservices. While blockchains put everything on-chain, Hetu takes some consensus off-chain, benefiting applications that have a high volume of transactions or events. This approach ensures low latency and hyper scalability, making Hetu perfect for applications such as deAI, decentralized physical infrastructure (DePIN) and decentralized science (DeSci), which require real-time finality and high throughput.

By selectively moving consensus off-chain, Hetu maintains the benefits of decentralized trust while overcoming the scalability limitations that have plagued many blockchain systems. This hybrid approach allows for interactions that are both nuanced and context-aware, bridging the gap between human subjectivity and digital objectivity.

Hetu’s ultimate goal lies beyond technological advancements. We embrace the philosophy of intersubjective trust-minimization to drive breakthroughs in the convergence of humans and AI. By enabling harmonious consensus and collaboration between different intelligences, we not only enhance the capabilities of DeAI systems but also ensure that they develop in a way that is aligned with human values and understanding.  

The development of truly complex systems that can capture the nuances of interactions between different intelligences requires us to delve into philosophy as much as into technological research. It necessitates a deep dive into the nature of trust, phenomenology, and shared understanding. 

Hetu combines innovative approaches like VLCs and causality graphs, allowing for interactions that are nuanced and context-aware. This enables digital interactions that maintain the subjectivity of humans while providing the objective trust, security and verifiability we expect from digital systems. This approach positions Hetu as a foundational layer for deAI.

We invite researchers, developers, and enthusiasts to join us and co-build on our research. Join our Telegram Builders Hub and follow us on Github to contribute to the protocol. Follow us on Twitter and Farcaster to stay up to date with the latest news.