Learn more about Bittensor, a protocol for decentralized subnets focused on machine learning, and one of the leading projects in the decentralized AI sector.
In November 2022, Sam Altman’s OpenAI unveiled ChatGPT to the world, bringing generative artificial intelligence (AI) and natural language processing to the mainstream. Over a year later, AI still remains a hot topic — thanks in part to Altman’s firing and subsequent rehiring
by the board of OpenAI.
The shocking news of Altman’s departure sparked rumors of a powerful artificial intelligence breakthrough, which OpenAI researchers reportedly
warned the board about. Some believe that the discovery is related to artificial general intelligence (AGI), autonomous systems that are better at humans in most economically valuable tasks.
With such powerful technology, many have also raised the potential dangers posed by AGIs. One concern posed by many is that this highly intelligent technology lies in the hands of large, centralized organizations, such as OpenAI and its largest stakeholder Microsoft, and other technology companies developing AI, such as Google’s Bard and Elon Musk’s xAI.
This has led to a surge in decentralized
AI projects, with many believing that blockchain and cryptocurrencies could play a crucial role in the development of artificial intelligence. For instance, Arthur Hayes argued
that AGI agents will choose Bitcoin as the currency to transact in. One such project leading the sector in decentralized AI is Bittensor, a decentralized machine learning focused protocol.
is developed by the non-profit organization, OpenTensor. Originally designed as a Polkadot parachain
named Finney, the protocol decided to launch its own chain in March 2023 to reduce its reliance on the Polkadot ecosystem. The current chain, aptly named Nakamoto, is modeled after the Bitcoin
Bittensor seeks to decentralize both access to machine learning models as well as training of machine learning models in a censorship resistant
manner. At the moment, training machine learning models require an immense amount of resources which only corporations, such as Google and OpenAI, are able to afford.
Bittensor aims to be the marketplace
for machine intelligence, where intelligence that is contributed to the network by honest contributors is valued and rewarded by a peer-to-peer
network in its native token, TAO.
Moreover, centralized machine-learning models are often trained in silo, preventing machine learning models from learning from similar models trained by other corporations. One of the core values in the crypto space is composability
, and decentralized AI is no different. Bittensor aims to leverage the compounded and composable nature of open development of AI to further push the space forward as models do not need to relearn what other models have already learnt.
Bittensor consists of three core components: a subnet, a blockchain and the Bittensor API. According to their official docs
, Bittensor is simply “a protocol for decentralized subnets.” Subnets are an incentive-based competition mechanism for a specific task. For example, the text prompting subnets incentivizes the best completion of text prompts. Users can create a custom subnet for their own competition or join an existing competition mechanism.
The Bittensor blockchain (subtensor) supports the subnets and ensures that the ecosystem is decentralized, permissionless and collusion-resistant. Meanwhile, the Bittensor API connects the subnets and the blockchain, ensuring rewards distribution to miners and validators on the subnets.
Subnetworks are Bittensor’s solution to decentralizing the training of machine learning models. Each subnetwork is a self-contained economic market to train and cultivate different forms of machine intelligence such as: text translation, image generation, text generation, data scraping and more. Subnetworks are governed by their owners and will consist of three types of users: miners, validators and users.
Miners, sometimes referred to as servers, are off-chain machine learning nodes which enable the function of the network. They are responsible for the servicing of requests to each subnetwork from its users and providing a relevant response based on their model. These responses are evaluated by the subnetwork’s validators, which will assign them a rank, resulting in a reward paid out in Bittensor’s native token, TAO. A miner that repeatedly provides inaccurate or subpar responses will be increasingly disincentivized due to lowered rewards and will drop out of the subnetwork, ensuring the quality of the subnetwork.
Validators check and evaluate the work done and submitted by miners in the subnetwork. Subnetwork owners can provide a template for validation, but validators can also express their own preferences through their own validation on what the subnetwork should be learning. This increases diversity in learning and also reduces risk of centralization
in the network where the model is forced in a specific direction.
Finally, users of the subnetwork provide the requests to the subnetwork as the end users, which are fulfilled by the respective miners. Users can use these subnetworks via the BitAPAI, which can be accessed via most common programming languages such as Python, Node.js, Golang and Rust
. The service is currently free to use to encourage development on the platform, although limits have been put in place to prevent abuse of the API
. Requests can be made to increase the limit however, subject to further questions and approval from the BitAPAI team.
Despite being a young protocol, several projects have been built atop the Bittensor network, relying on the services of specific subnetworks, with most targeting text or image generation.
A popular type of projects are the ChatGPT-style bots such as ChatNI, Chat with Hal and Chattensor, which all function as chat-powered assistants where users can input a question or request and receive a reply from the bot.
Another popular project building on Bittensor is ReplyTensor, which utilizes image and text generation subnetworks to generate automated replies on Twitter, Telegram and Discord. ReplyTensor is developed by Neural Internet, which is an AI research decentralized autonomous organization
(DAO) and one of the main builders on Bittensor’s resources.
Much like the Bitcoin network from which Bittensor was modeled after, the TAO token
’s design closely follows that of Bitcoin with a total supply of 21 million tokens. Aiming for the fairest launch possible, Bittensor did not have any presales or private investors, with even the founders and founding team having to mine their own tokens.
Likewise, TAO follows Bitcoin’s halvening model as well, where emissions
are cut in half every 10.5 million blocks. The first halvening is expected in September 2025, with 64 halvening events in total.
Aside from the native TAO token, the TAO token also exists as an ERC-20
version on the Ethereum mainnet
as wrapped TAO
(wTAO). WTAO can be bridged
back to the Bittensor network via the TAO Bridge. Do note however that TAO Bridge is a community-run project and is not officially run by the Bittensor team.
The current state of machine learning training is held back by its siloed approach and lack of composability. As the Bittensor network continues to grow and develop, the network seeks to one day be the center of all things machine intelligence, with users, startups and even mega corporations tapping into the network for machine intelligence capabilities, with all services paid for in TAO tokens.
This article contains links to third-party websites or other content for information purposes only (“Third-Party Sites”). The Third-Party Sites are not under the control of CoinMarketCap, and CoinMarketCap is not responsible for the content of any Third-Party Site, including without limitation any link contained in a Third-Party Site, or any changes or updates to a Third-Party Site. CoinMarketCap is providing these links to you only as a convenience, and the inclusion of any link does not imply endorsement, approval or recommendation by CoinMarketCap of the site or any association with its operators. This article is intended to be used and must be used for informational purposes only. It is important to do your own research and analysis before making any material decisions related to any of the products or services described. This article is not intended as, and shall not be construed as, financial advice. The views and opinions expressed in this article are the author’s [company’s] own and do not necessarily reflect those of CoinMarketCap. CoinMarketCap is not responsible for the success or authenticity of any project, we aim to act as a neutral informational resource for end-users.