Decentralized AI (DeAI) was the talk of town at this year’s TOKEN 2049 event in Singapore. But an open question remains — What are novel use cases of DeAI? After all, there has to be a reason why $12.7B has been invested in this space till date. Philosophical primitives of centralization cannot alone be the reason why someone uses DeAI technologies.
What is a decentralized intelligence cluster?
Many companies and organizations today can be viewed as intelligence clusters i.e. A group of intelligent beings (human+ AI) coordinating driven by incentives — Money, prestige, impact etc.
A VC firm, for example, is an intelligence cluster. It’s not necessarily a new idea. Folks have written about organizations as living organisms. The main question we are interested in answering is can you coordinate intelligent work eg. analysis, consulting, research, primarily using crypto-incentives?
The think the answer is yes, and we believe decentralized intelligence clusters, trained on domain-specific knowledge, and collaboration amongst these clusters, is the future of work.
How would this work?
The short answer is by combining domain-knowledge experts with AI, operating on permissionless networks. We are already seeing companies like Hebbia doing this for employees internally at Fortune 500 companies and web3 projects like Talus Network enabling smart agents to solve specific tasks.
At Reppo, we believe that the next frontier for AI is a decentralized Hebbia, combining agents and models, trained on niche datasets, to solve a context-specific problems instead of feeding AI models the entirety of the internet.
The concerns often flagged whenever the idea of crowdsourcing intelligence or intelligent work falls in two categories —
- You need proprietary data to build quality intelligence.
While it’s fair to say that quality intelligence does not come just from public data (although there are examples of platforms such as Arkham that have proven that significant intelligence can be derived from public data), accessing private data is becoming more and more real with each passing day as we see solutions emerge leveraging Zk technologies to enable access to this data class in a privacy preserving manner Eg. Provably.ai and novel incentive structures Eg. Vana.org
We are building Reppo to allow private sharing of data and datasets between the Reppo clusters and clients who want to start ith crowdsourcing intelligence, before fully embracing decenteralization.
But, how do you access context?
Once we have built the rails to access private data in a privacy preserving manner, how does one access context? After all, the human working at an organization for the last 20 years has a lot more context than a random person on the street, correct?
Non intuitively, that’s not always the case. At least in the world of web3 and blockchains, context can be gained by accessing public information, such as discord, X, and even direct engagement with members of a community. We ran experiments over the last 8 months where individual model developers built models without any previous context on Web3 networks like Filecoin and DIMO. These models were used by the communities and stakeholders and were shipped in 4 to 6 weeks.
With Reppo’s intelligence clustering approach that encourages collaboration using native incentives, context onramps improve exponentially, combining context with niche datasets and on-chain (co)ownership of models and agents.
At Reppo, we are committed to building an orchestration engine that allows anyone to bring intelligence on-chain with a few clicks. The most dominant use case for Decentralized AI will be customization and personalization of AI, powered by domain-specific intelligence clusters, both public and private.
