The Future of Distributed AI
Openness promotes innovation, and recent advances in artificial intelligence (AI) have demonstrated its global utility and influence. As computing power increases through resource integration, centralization problems are likely to arise, with entities with superior computing capabilities gaining dominance. This centralization can hinder the pace of innovation. Decentralization and Web3 technologies offer promising alternatives to maintain the openness of AI.
Distributed computing for pre-training and fine-tuning
Crowdsourced computing (CPU + GPU)
Supporting argument: A crowdsourcing model similar to that used by platforms like Airbnb and Uber could be applied to computing, pooling idle computing resources into a marketplace to provide low-cost computing solutions for specific use cases and providing censorship-resistant resources for training models that may face future regulation or bans.
Counterargument: Crowdsourced computing may not achieve the economies of scale required for high-performance tasks, since most high-performance GPUs are not consumer-owned. The concept of distributed computing seems to contradict the principles of high-performance computing.
Distributed Inference
Open source model inference distributed execution
Supporting argument: Open source models are approaching the capabilities of closed source models and are gaining popularity. Centralized services such as HuggingFace or Replicate for model inference raise privacy and censorship issues. Decentralized or distributed vendors can solve these issues.
Counterargument: Local inference, facilitated by dedicated chips that can handle large-scale parametric models, may ultimately dominate. Edge computing offers a solution for privacy and censorship resistance.
On-chain AI agent
On-chain applications leveraging machine learning
Supporting argument: AI agents that require a transaction reconciliation layer can benefit from cryptocurrency payments because they are inherently digital and cannot utilize traditional banking systems. On-chain AI agents can mitigate platform risks, such as sudden changes in the plugin architecture of entities like OpenAI, which can cause service disruptions without warning.
Counterargument: Current AI agents such as BabyAGI and AutoGPT are not yet production-ready. Also, entities that create AI agents can use payment services such as Stripe without relying on cryptocurrencies. The platform risk argument has been used to justify cryptocurrencies in the past, but has not yet been realized.
Data and model sources
Autonomous management and value collection for data and machine learning models
Supporting argument: Data ownership should belong to the users who create the data, not the companies that collect it. Data is a critical resource in the digital age, and the monopolization and inappropriate monetization of it by major technology companies is a serious problem. A more personalized internet requires portable data and models that allow users to transfer data between applications, similar to how they move cryptocurrency wallets between dapps. Blockchain technology can provide a viable solution to the data sourcing challenge, especially in the face of increasing fraud.
Counterargument: Data ownership and privacy concerns may not be a priority for users, as evidenced by the high number of signups on platforms like Facebook and Instagram. Trust in established institutions like OpenAI may overshadow concerns about data ownership.
Token incentive apps (e.g. companion apps)
Crypto Token Reward Concept
Supporting argument: Crypto token incentives are effective in encouraging network growth and behavioral engagement. Many AI-centric applications are expected to adopt this model. The AI companion market represents a significant opportunity and has the potential to be a multi-trillion dollar sector. Past data, such as $130 billion spent on pets in the US in 2022, suggests a robust market for AI companions. AI companion apps have already seen significant engagement, with average session lengths exceeding an hour. Crypto incentive platforms can capture significant market share in this and other AI application sectors.