Spectral | Vision Paper: Machine Intelligence Network

Vision Paper

Spectral’s Machine Intelligence Network

Unlocking the Inference Economy by bridging ML and Web3 with verifiable, trustless inference feeds for seamless integration into
smart contracts.

Unlocking the Inference Economy by bridging ML and Web3 with verifiable, trustless inference feeds for seamless integration into smart contracts.

Spectral’s Machine Intelligence Network is a two-sided protocol offering high-quality inference feeds for smart contracts to consume machine learning (ML) and artificial intelligence (AI). Using zero-knowledge machine learning (zkML) allows for privacy preservation, a unique validation process, and an easy-to-use, state-of-the-art software development kit.

The protocol enables the Inference Economy, an incentivized system that rewards machine intelligence miners, protects their intellectual property (IP), and generates ML inference feeds to be consumed directly by smart contracts. This technology essentially bridges the gap between ML, AI, and the blockchain.

Streaming Inference Feeds

Spectral offers an innovative use case: inference feeds for smart contracts.

While existing oracle networks cater to data feeds, most notably for price data, Spectral addresses a significant gap by enabling an unprecedented feed of high-quality inferences. This broadens the scope and the potential market of smart contracts in general, opening new horizons for decentralized applications in Web3.

Design Principles

Value in a machine intelligence network is fundamentally determined by the quality of machine intelligence from a Consumer’s perspective. Decentralization alone does not guarantee the long-term reliability and sustainability of a network without some expectation of quality. Recognizing the significance of this requisite, Spectral's network also involves Validators that play a pivotal role in upholding quality while remaining decentralized and trustless.

Validators leverage the concept of a randomness beacon to uniquely validate machine learning models contributed by each Modeler in a tamper-proof fashion. Unlike traditional machine learning methodologies that rely on a single static metric to evaluate model performance, Spectral's approach is more general: a collection of machine learning metrics, either weighted or dynamically adjusted, is applied to provide robustness across various scenarios. As transparency is paramount, Validators openly upload all relevant intermediary steps, e.g. to IPFS (InterPlanetary File System), offering full disclosure and accountability.

It is through all these methods combined that Spectral's unique validation mechanism is mathematical, yields inferences that are provably high-quality, and effectively safeguards against collusion among Modelers and Validators. In particular, this security against collusion is based on the security of the randomness beacon, drawing parallels with the security reduction that serves as a foundation for Ethereum's security.

On Incentives, Security, and Quality

It is worth noting that incentive mechanisms are factors that augment the overall security argument. While incentive mechanisms play a significant role in the overall architecture of a web3 network, it is essential to understand them in the context of machine intelligence. A validation mechanism solely based on incentives does not suffice and can still yield unreliable inferences, endangering the network's long-term value and viability. From the vantage point of applications, a network of machine intelligence must prioritize the quality of machine intelligence in order for the network to be useful and the miners of the network to be performing work that is useful a la proof of useful work. With a useful network, an incentive layer (or a disincentive layer) can most naturally supplement the fundamental value of the network, and this understanding drives Spectral's methodical approach in intertwining machine intelligence and web3 to fuel a network that is perpetually useful.

To further prevent faulty outputs, Spectral adopts a modern technique within the realm of verifiable computation called zkML (zero-knowledge machine learning). While zkML involves intricate cryptographic and mathematical concepts, it is vital to shield users from these technical complexities on the surface for a frictionless user experience. The objective is to achieve the integrity of machine intelligence, so that outputs are not tampered with at the last minute during inference time, while not sacrificing user experience. Accordingly, Validators may explore alternative techniques beyond zkML as long as they adhere to the core principle of verifying the integrity of machine intelligence without compromising user experience.

A Network of Verifiable Machine Intelligence

Spectral's inference economy operates as a dynamic two-sided network. On the one side, there are miners (Modelers) who train and contribute machine learning models in a privacy-preserving manner. On the other side, we have Consumers with a fundamental demand for customized, high-quality inferences.

Consumers who are unable to find a feed of inferences to solve their problem may also opt to issue a challenge of their own, becoming a Creator, staking a bounty, setting benchmarks and usually providing a dataset and a description.

Two points are worth noting in this latter aspect. The quality of inferences produced often hinges on the specific context in which they are employed. This means that relying solely on general-purpose LLMs, for instance, may fall short of meeting the precise requirements of specialized use cases. Hence, fine-tuning emerges as an important concept and need, ensuring that machine intelligence remains practical and applicable in diverse scenarios.

The fact that a network upholds the ethos of decentralization, which involves distributing trust and intellectual resources from centralized entities, does not immediately imply the quality of outputs. In this regard, Spectral adopts a meticulous design that harmoniously combines the virtues of both.

Modelers form the backbone of the protocol, playing the role of alchemists taking raw data and materializing it into valuable inferences. The permissionless nature of the network allows any sophisticated technique to underpin the network. Privacy-preserving methods protect such models, allowing perpetual incentivization for a Modeler to host and offer ongoing value to the network.

Consumers represent a diverse spectrum of entities seeking high-quality inferences. Whether it is powering smart contracts, enhancing decision-making, writing code, or enabling any other innovative applications, the demand for reliable machine intelligence is only growing. Spectral offers a solution to this demand, presenting a tapestry catering to the unique needs of each Consumer.

Inferences in an Agent-Based Marketplace

One of the compelling uses for contemporary artificial intelligence is the creation of semi-autonomous agents capable of interacting with one another, which means agents using smart contracts could transact with one another, automatically executing certain trades for example.

These agents automatically verify, rather than merely trust, inferences and outputs from other machine learning models.

As Spectral's core philosophy lies in the verifiability of both the integrity (via verifiable computation) and quality (via perpetual mathematical validation) of machine intelligence, an agent-based marketplace is made possible where an agent can be LLMs, smart contracts, or any sort of machine intelligence.

By shifting the trust paradigm from centralized authorities to automated agents, Spectral enhances the efficacy and accessibility of machine intelligence. An agent-based marketplace fosters an environment of collaboration and competition among automated agents, further elevating the quality of inferences and interactions in the network that evolves. While this type of agent-based marketplace and its evolution represent a vision for the longer term, Spectral aspires to lead the way in forging this path and push the boundaries of machine intelligence and web3.

The Beginning of the Inference Economy

With Spectral, we embark on a journey into an era of the Inference Economy where machine intelligence is transparent, decentralized, and accessible to all. It is a future where the world's brightest minds collaborate to solve complex challenges, where verifiability meets machine learning, and where the possibilities of Web3 are amplified by high-quality machine intelligence, and vice versa.