This document is a TL;DR for the more in-depth technical paper which was posted on the Opentensor Twitter page on Jan 9th.
Intro
Opening up the governance of Bittensor to the widest possible group of individuals, whilst simultaneously aligning Bittensor towards producing the highest quality of machine intelligence, is no small feat. Indeed it is only possible at the intersection where good governmental choices, and self-interested choices, are one-and-the-same for TAO holders.
The BIT1 proposal, colloquially termed Dynamic TAO, is one path through the eye of this needle. It undermines the centralization of control in Bittensor’s root network, whilst replacing it with aligned market participants.
Background
Bittensor is currently composed of 32 self-contained incentive mechanisms (subnets), which produce various decentralized digital commodities i.e. data scraping, storage, model training, and ML inference. All of these subnets are accessible to TAO holders, and interconnected within the ecosystem that is Bittensor.
As to date, Bittensor calculates the distribution of TAO such that each block is determined via the ‘Root Network’: a voting pool composed of Bittensor’s top 64 validators (measured by the sum of TAO that has been delegated to them). The more Root Network emissions a subnet receives, the more commodity it can extract, whilst the subnets which do not receive sufficient emission are pruned out.
The distribution of TAO is essential to how Bittensor runs: for whom, for what, to riches or to ruins. The BIT1 proposal is designed to improve the previous mechanism. Replacing the centralized root network determination with one in which all TAO holders vote directly — increasing or decreasing the proportion of emission a subnet receives, based on the distributed estimation of how valuable its commodity is, or will be, to the ecosystem.
Why?
““The dispersion and imperfection of all knowledge are two of the basic facts from which the social sciences have to start” (Hayek 1952a, p. 50).”
There are multiple reasons for the The BIT1 proposal. However, its primary raison d’être is that it translates the calculation of TAO’s inflation distribution away from the subjective preferences of the small set of root network validators and re-assigns the responsibility to a more efficient market mediated by all TAO holders.
This proposal is essential because the Root Network is, in its current form, centralized around a small number of groups. The pareto (powerlaw) distribution of their vote makes this process centralized in a way that cannot be simply solved by down-scaling i.e. quadratic voting schemes which merely push participants to split their accounts or requires KYC sybil resistance.
Centralized institutions can be trusted to make decisions based on narrow self-interest, cronyism, or apathy. All of which will lead, and have already lead, to the misallocation and misdirection of Bittensor’s economic resources and value proposition. It is unreasonable to presume that the root network’s validators will perform due diligence on each subnet without direct consequences of economic incentives and penalties.
Among other reasons listed at the bottom of this article, The BIT1 proposal opts to disperse this decision-making process and bring the market into its computation.
How does it work?
The BIT1 proposal’s core addition to the Bittensor protocol is the demarcation between TAO that is held on the balance of a wallet (i.e your coldkey) and Dynamic TAO that is held on that wallet’s staking accounts ( i.e. the hotkey). The demarcation is made by introducing an intermediary pool function, which translates TAO to-and-from its Dynamic TAO equivalent.
What makes this proposal ‘dynamic’ is that the amount of staked balance attained during a staking operation, and the amount of TAO received during an un-staking operation, dynamically changes based on the current exchange rate of the intermediate pool. The exchange rate of the pool is determined by the reserves of TAO and the Dynamic Token in the pool, the quantities of which modulate when participants unstake or stake — either removing TAO and depositing Dynamic TAO or vice versa.
The main argument made by the proposal is to use the relative prices for each subnet, in order to calculate how much TAO should be distributed towards each mechanism per block. We are therefore replacing the root network with the price determinations created by individual TAO holders, choosing to invest (stake) or divest (unstake) in each subnet.
What makes this proposal uniquely Bittensor, and why it has been called the ‘subnet token proposal’, is the way in which Dynamic TAO is created. This is because each Dynamic token has its own emission schedule, with 1 token minted every block into its pool, and 1 token simultaneously created and distributed to the miners and validators within the subnet. TAO emission only interacts with this token by being added into the pool, such that the Dynamic token for each subnet represents a varying portion of that emission, but has its own supply and Bitcoin emission schedule.
Validators use their weight in the Dynamic token to reach consensus. Rewards for each subnet are payed out in that same token. The supply and demand curves for each token determine that pool’s resting price. Where all of the pools balance to a 1/(n_subnets) price without intervention. This forces subnets to attract demand for their digital commodity that they create. Otherwise, they cannot attain TAO inflation, and cannot liquidate their Dynamic tokens. This shields TAO inflation from ineffective subnets, only emitting TAO into the pools that add value to the Bittensor ecosystem.
Dynamic TAO is non-fungible with other currencies, and only exchangeable in TAO. Each Dynamic Token only represents a varying quantity of TAO and does not dilute TAO. Nor does it reduce TAO’s demand, since all of the demand for Dynamic Stake in each subnet is itself subsumed by the demand for TAO (as the intermediary).
Delegation in the Dynamic TAO system remains similar to its previous design with the exception that TAO holders now stake a single Dynamic token directly to a delegate on that subnet. Nominators choose where their stake weight should go, directing delegates into the “best” subnets. Delegation rewards are always made in terms of the subnet Dynamic stake.
Summary:
The proposal to remove the root network centralization, and thus undercut the the domination by TAO whales, is critical. Dynamic TAO directly decentralizes the computation of the flow of emissions, offloading the responsibility to the market, whilst introducing consequences for bad speculation, lethargy, apathy, and self-interest. The proposal instead adds a new layer to Bittensor, where any holder is actively involved in the vote for which subnets get rewarded, how much, and when.
As a byproduct, this brings more pressure to subnet owners to perform well. The whole market will be watching them and they must impress the market, rather than a few entities. The increased pressure comes with potentials for outsized gains for hard work. We expect Bittensor subnet development to accelerate, as the market is constantly watching the code and products. At any moment, any holder can reward or penalize a subnet team by moving their stake elsewhere.
Those that claim Bittensor should focus more on AI and less on crypto, or vice versa, misunderstand the union of both. Bittensor is the leader in Crypto-Dynamics, where incentives meets AI. Dynamic TAO is both shield and sword for this mission.
Happy mining 🚀
@const
Properties of Dynamic TAO
- Dynamic Tokens are non-fungible, they cannot be transferred, they can only be unstaked into TAO.
- Each Dynamic Token begins with an initial supply of 32x the lock cost paid by the subnet creator.
- Dynamic Tokens are emitted by the chain, with 50% of the supply being injected into the Pool, and 50% used to incentivize miners/validators within the subnet.
- The Dynamic Token associated with the subnet recieves 50% of the consensus weight in that subnet. The remaining 50% is given to global TAO economic weight.
Properties of Dynamic TAO Pools
- There is only 1 pool associated with each subnet. It is run automatically, and individuals cannot create their own or inject liquidity into them.
- Each pool contains liquidity for both TAO and the subnet’s specific Dynamic Token (i.e. Alpha).
- The pool uses the uniswap protocol to maintain a constant multiple of each token, and to change the price based on the ratio of these reserves.
- The TAO portion of the pool is the sum of TAO that has been staked into this subnet plus what has been injected via TAO inflation.
- The Dynamic token portion of the pool is the sum injected by that subnet’s coinbase (minted every block) plus that which has been unstaked by participants in that subnet (miners/validators)
How does this change Delegation?
- Delegation is now made in terms of each Dynamic tokens, i.e. you hold a dynamic token as reserve with each delegate
- Delegation rewards are always made in terms of the subnet Dynamic Token which is reflected in your staking balance, and not converted into TAO
- Nominators receive the Dynamic Token from every subnet which their delegate validates on.
