Information Mining: Proof of Useful Work
How to bootstrap a massive dataset to train an Oracle AI
On this day in 2009, the first distributed, peer-to-peer cash, Bitcoin, went live. Bitcoin was based on a proof-of-work system to create digital scarcity and prevent double spending. Today, 14 years later, we’re excited to introduce a new methodology for minting scarce tokens: proof of useful work.
Note: The Information Mining process went live at 11:00 pm GMT on January 3, 2023. If you’d just like to dive in and participate, simply visit http://askluci.tech with a metamask enabled browser, connect, and start asking questions. We also recommend a look at this wiki page for more instructions and tips on how to proceed: https://github.com/cryptohariseldon/Ask-luci/wiki/Question-Quality-Guide
Introducing Proof of Useful work
While excellent at achieving its stated goal, the work performed to mine Bitcoin, namely, brute forcing billions of SHA256 hashes, is not “useful” to humanity in any conventional sense, other than the security it provides to the bitcoin network itself.
Theoretically, the work itself could be something that produces an end product with intrinsic value — such as high quality training data for a machine learning model. It is with this premise that we introduce LUCI’s information mining process to the world.
Information Mining — what is that?
In the simplest terms: IM is the process by which users are rewarded for providing high-quality training data to LUCI.
“Mining” evokes the expereince of putting in extensive labour to recover a scarce and valuable resource. This could be a metal, minerals, or even Bitcoin. Along similar lines, users could also mine for “training data”, to participate in training a large and powerful language model.
The key challenge in building a powerful oracle AI is creating a comprehensive database of question-answer pairs for the AI to learn from. Previous attempts have relied on manual annotation, curation or scraping of existing databases, which are time-consuming and prone to errors. To address this issue, we propose a novel approach called “information mining”, which incentivises users to ask unique and useful questions by rewarding them with Luci Credits.
Information Mining (IM) is the process by which humans are rewarded for transfering their expertise and knowledge to the AI, in annotated question-answer format.
Users participate in the information mining process whenever they ask a unique, specific, and useful question on the LUCI platform. Users can choose to get paid for their efforts in the native currency of the LUCI platform, i.e. LUCI Credits. These credits will be used to avail of premium LUCI AI assistant features in the future — and they can also be traded on decentralised exchanges. The total supply of LUCI credits ever minted will be capped at 100 Billion tokens — and this amount would only get minted if our goal of collecting 100 Billion unique, high-quality question answer pairs is reached.
Since the process of minting new LUCI credits is linked solely to the process of information mining, the token supply will directly correspond to the size of the dataset that LUCI has collected so far — which can be seen as a proxy for the value of the LUCI platform as a whole.
More information on the exact reward mechanisms can be found here: https://github.com/cryptohariseldon/Ask-luci/wiki/Token-Economics
Details about our future monetization plans can be found in our whitepaper.
