Bittensor

From Yugenpedia
Jump to navigation Jump to search

Bittensor has information that may be valuable to qualified investors stored in its entry in the Yugenpedia Vault. If you'd like to learn more or are interested in receiving access please contact the Yugenpedia Staff.

Bittensor is an open-source peer-to-peer framework that democratizes machine intelligence using a blockchain.

Problem Statement

Over the past 10 years, artificial intelligence has graduated from the research lab into the core business model for trillion-dollar companies like Google, Baidu, and Facebook. This evolution has throttled the legacy academic system which was designed to investigate intelligence using supervised datasets as a measure of progress. While this system has worked well for pushing incremental performance, the mechanism has become incredibly inefficient as the computational costs to train intelligence systems have exploded. This is because the objectives used to guide the field (often measured in uni-dimensional metrics like accuracy) do not have the resolution to reward niche, legacy, small, or otherwise non-state of the art contributions of intelligence, thus the product of most machine learning researchers goes to waste. Ultimately, because the mechanism only rewards large monolithic models which succeed in a winner-take-all competition, standalone engineers cannot directly monetize their work and centralization results where a small set of large corporations control access to the best artificial intelligence.

Product

Bittensor aims to ameliorate these issues by constructing an intelligence market that is better suited for an intelligence industry in the 21st century. Our product is a collaborative decentralized p2p intelligence market which can be accessed by intelligence consumers by staking a digital currency called TAO and rewards peers in turn with that same currency. By constructing a market around intelligence rather than by using supervised objectives, we produce a high-resolution and efficient reward landscape for intelligence as a commodity, thus hopefully exponentiating its production as well as decentralizing its ownership and control.

Team

Bittensor is an open-source project which aims to be fully decentralized and controlled through an on-chain DAO mechanism. The project is being helped by the Opentensor Foundation, a Canadian Non-Profit, which is co-run by Ala Shaabana and Jacob Steeves who are core contributors/developers/miners in the Bittensor network.

The Cognitive Internet

The inspiration for Bittensor came from brain cells (neurons): neurons learn together by passing signals along synaptic channels. Neurotrophins, essentially food for neurons, pass retroactively "paying" neurons for being useful to their peers. Bittensor works similarly: 'neurons' or 'miners', each running a unique machine learning model, learn together by passing signals to each other. An incentive token, Tao, passes between them, literally paying those that are valuable to the collective.

Each user's neuron on the Bittensor network contains a model, a loss function (to measure performance), and a dataset. Neurons will train their model on the user's computer, while also exchanging information with nearby neurons. Each neuron will receive data from its fellow neurons, learn from that data, and send back what it has learned. Neurons receiving information will reward its fellow neurons according to how that information affects their performance.

Intuitively, Bittensor is a machine learning model. One which is spread across many computers on the internet and which collectively measures the value produced by each "peer" in the network. That is, it answers the question "Does peer X make the model smarter?". The usefulness of these peers directly correlates with the revenue of the mined currency in the system. Thus, creating a market for machine intelligence.

Competition

Other projects have made a similar claim to decentralizing AI: SingularityNET, GenesisAI, and Fetch.ai. All of these systems leave some human element in incentivizing their users, in essence acting as an App store of AI models. What sets Bittensor apart from its competitors is that it leverages the efficiency of markets to decide the distribution of knowledge and who should be rewarded. In other words, there is no human intervention in the reward system. This not only removes bias but also encourages competition between users to create the best models.


Challenges

Today, Bittensor faces two major obstacles: decentralization and cabal formations.

Decentralization

Bittensor is designed to be decentralized and works best when it is most decentralized. Therefore, decentralization is a top priority for this project. As we are still in the testing phase, we have not yet achieved decentralization. This is also a crucial milestone to reach before being listed on any exchange.

Cabal formation

Due to the nature of decentralized AI, it is not possible to enforce honest rewards without access to each node's local neural network. This is not possible. In fact, nodes with little competitive interest in attaining value from their neighbors may still collude to gain inflation without adding value to the network. This is achieved by forming a 'cabal', a set of one or more tightly connected nodes that falsely evaluate each other. Since there is no way to distinguish between rewards that are correctly given and those that are not, this behavior will spread if left unchecked.

The Bittensor whitepaper solves this problem mathematically, however, the implementation is presently being developed.

Goals/Results

Bittensor has Goals/Results stored in its entry in the Yugenpedia Vault. If you are a qualified investor interested in receiving access, please contact the Yugenpedia staff.

SWOT

Bittensor has a SWOT analysis stored in its entry in the Yugenpedia Vault. If you are a qualified investor interested in receiving access, please contact the Yugenpedia staff.