Building Block: Bittensor
Will Ogden Moore
Last Updated: 8/16/2024 | 13 min. read
Bittensor is a platform that helps facilitate the development of open and global artificial intelligence systems through the use of decentralized networks and economic incentives.
Bittensor stands at the forefront of two of the most groundbreaking, transformative trends in software: blockchain and artificial intelligence (AI). While Bitcoin helped create the crypto industry as the first peer-to-peer money system and a digital store of value and Ethereum helped to expand the ecosystem with decentralized applications, Bittensor represents an entirely new and distinct use case. It intends to harness the properties of permissionless, public blockchains and economic incentives to develop advanced AI software through an open, decentralized community (rather than a centralized company).
Today, AI development is highly centralized, with significant power concentrated in a few large tech companies. As AI continues to evolve into a more powerful and essential tool, we run the risk that the immense power of AI is controlled by a few entities and is not aligned with human values and the broader society. In contrast, Bittensor is a platform that economically incentivizes open collaboration of AI development through its native token, TAO. Through use of a public blockchain, Bittensor could potentially help democratize ownership and increase transparency around AI systems as well as align decisions regarding AI development with the interests of society. Bittensor aims to create the "Internet of AI," envisioning a future with many interconnected AI ecosystems, or subnets, that form a global decentralized AI platform. This platform would help enable anyone, anywhere, to easily build, deploy, and access artificial intelligence applications by connecting to the Bittensor network.
The Token
TAO is the native token of the Bittensor network and ownership of the TAO token represents a piece of ownership in the ecosystem (Exhibit 2). The TAO supply schedule directly mirrors that of Bitcoin, with a maximum supply of 21 million and a halving approximately every four years. Bittensor’s first halving event is expected to occur in August 2025.
Bittensor is intended to apply Bitcoin-style incentives to AI development, with the TAO token used as an incentive for network participants to perform their intended functions. These participants include network validators and subnetwork (subnet) owners, subnet validators, and subnet miners (these terms are defined in the next section). Beyond incentive rewards, TAO is primarily used today as a deposit fee for subnet owners registering their subnet. In the near future, as the nascent Bittensor network matures, potential additional use cases for TAO include (i) as gas fees for network transactions, (ii) as decision power over which subnets to allocate TAO emissions, and (iii) general network governance decisions. In the longer term, Bittensor may monetize its network by charging end-user applications that use its subnets; this would potentially generate value accrual back to the TAO token.
Exhibit 2: TAO token basics
The Network and Technology:
Bittensor employs a wide variety of subnets tailored to perform distinct machine-learning tasks. For example, one subnet is tailored to AI image generation, another is for AI music generation, and another is for detecting AI-generated deep fakes. Each subnet involves three primary types of actors: the subnet owner, subnet miners, and subnet validators. On a given subnet, miners compete to produce the “best” model outputs while validators assess which miners are the “best” performing (see below). Though elements of this process vary from subnet to subnet, the general idea is outlined below:
How It Works
End users prompt the network through a consumer-facing application. This is akin to asking ChatGPT a question.
Subnet miners run AI models on the relevant subnet and compete to produce the best output for a given prompt. For example, in the chatbot subnet, miners would compete over the best text-based answer to a user’s question.
Validators rank the miner responses by output quality, and the top-ranked response is given back to the end user who asked the question.
(Examples provided for illustrative purposes only.)
Validators determine miner performance through a novel process known as Yuma consensus. This consensus mechanism aggregates the rankings of each validator, weighted by their amount of TAO staked, to produce a collective ranking list of miner performance.
The broader Bittensor blockchain operates under a “proof of authority” consensus mechanism where certain nodes are granted the authority to order on-chain transactions and help maintain the network's integrity. Bittensor’s blocks store updated state changes and token balances to reflect new emissions to network validators, as well as subnet owners, miners, and validators.
Use Cases
Bittensor has a wide range of potential use cases, with each subnet representing a different example. Examples include:
Image Generation Subnet: Tailored for AI models that specialize in creating high-quality, generative images.
Chatbot Subnet: Optimized for AI models that specialize in natural language processing and allow for consumers to access highly responsive virtual assistants.
Deepfake Detection Subnet: Leverages advanced generative and discriminative AI models within the Bittensor network designed to detect AI-generated images.
Factors to Consider
Permissionless Access to Developing and Using Powerful Technology: As AI continues to evolve into a more powerful and essential tool, there could be increasing regulations or restrictions around who can build or access these applications. Bittensor offers an alternative where permission is not needed to access resources to develop and use AI.
Economic Incentives Promoting Equitable AI Development: Compared to centralized alternatives, Bittensor can help provide greater access to AI resources such as compute, storage, and data for independent AI developers. It can also help enable AI researchers and open-source AI developers to monetize their contributions and potentially fund their operations. In success, Bittensor’s open and distributed ecosystem could help provide balance with the closed-source models developed by tech giants and help ensure that the economic benefits of AI are more widely shared.
Growing Adoption and Recognition: Bittensor has gained early traction, with over 40 subnets dedicated to specific AI tasks and has gained endorsements from prominent tech and AI leaders.[7] Companies are raising venture funding to build subnets and applications on Bittensor, indicating growing investor and developer interest in the ecosystem and the potential for Bittensor to grow network effects.
Adoption and Network Growth: The longevity of Bittensor depends on attracting a critical mass of developers and AI projects to build on the platform. If Bittensor fails to achieve significant adoption, the network may struggle to reach its full potential. In addition, given the nascency of the network, the majority of network resources are focused on the infrastructure level and subnet activity. Over time, Bittensor will need to expand the quantity and quality of the end-user applications using the Bittensor network to help increase both token value accrual and its relevance to everyday consumers.
Level of Decentralization and Network Resilience: Bittensor’s operations depend on the smooth functioning of a wide network of participants. Any disruptions, such as technical failures, bugs, or attacks on the network, could impact its performance and reputation. Bittensor will also need to increase overall network decentralization and distribute voting power over TAO emissions allocations more broadly across the network.
Implementation of incentive design: In order to reach its full potential, we believe Bittensor will need to ensure that subnet owners design the right incentive mechanisms for their subnets and that the network properly allocates emissions to the right subnets over time.
Competing Networks: Bittensor faces competition from AI-related crypto assets that attempt to address AI development with token incentives such as Allora, Sentient, Sahara AI, as well as other assets that cover various AI-related use cases such as Filecoin and Gensyn among others. This list may also grow over time as this intersection matures.
[1] AI Universe as defined by the Grayscale Research. Assets must have a minimum market capitalization of $500mm. AI Universe is rebalanced quarterly and was rebalanced at 4/1/2024 and 7/1/2024. The assets in the AI Universe include NEAR, FET, RNDR, FIL, TAO, THETA, AKT, AGIX, WLD, AIOZ, TFUEL, GLM, PRIME, OCEAN, ARKM, and LTP.
[2] Compute refers to the processing power and resources required to train, deploy, and run AI models
[3] For illustrative purposes only
[4] For illustrative purposes only
[5] Markets and Markets. Data as of 2023, projected through 2030.
[6] Grayscale Investments as of August 8, 2024. Represents “Universe” of AI-related crypto assets based on Grayscale Research Methodology.
[7] 45 subnets as of August 14, 2024 (source)
Share This
Like what you see?
Subscribe below for the latest research from Grayscale.
