Openness fosters innovation, and recent advances in artificial intelligence (AI) have showcased its global utility and influence. As computing power increases through resource integration, centralization issues are likely to arise, with entities possessing superior computing capabilities gaining dominance. This centralization could hinder the pace of innovation. Decentralization and Web3 technologies offer promising alternatives to maintain the openness of AI.
Decentralized Computing for Pre-Training and Fine-Tuning
Crowdsourced Computing (CPUs + GPUs)
Supporting Argument: The crowdsourcing model, similar to those used by platforms like Airbnb and Uber, could be adapted for computing. This model would aggregate idle computing resources into a marketplace, potentially offering lower-cost computing solutions for specific use cases and providing censorship-resistant resources for training models that may face future regulations or bans.
Opposing Argument: Crowdsourced computing may not achieve the economies of scale necessary for high-performance tasks, as most high-performance GPUs are not consumer-owned. The concept of decentralized computing seems contradictory to high-performance computing principles.
Decentralized Inference
Running Open-Source Model Inference Decentralized
Supporting Argument: Open-source models are approaching the capabilities of closed-source models and gaining traction. Centralized services such as HuggingFace or Replicate for model inference introduce privacy and censorship concerns. Decentralized or distributed vendors could address these issues.
Opposing Argument: Local inference, facilitated by dedicated chips capable of handling large parameter models, may ultimately prevail. Edge computing offers solutions for privacy and resistance to censorship.
On-Chain AI Agents
On-Chain Applications Using Machine Learning
Supporting Argument: AI agents, which require a transaction coordination layer, can benefit from cryptocurrency payments, as they are inherently digital and cannot utilize traditional banking systems. On-chain AI agents mitigate platform risks, such as sudden changes in plugin architectures by entities like OpenAI, which can disrupt services without warning.
Opposing Argument: Current AI agents, such as BabyAGI and AutoGPT, are not yet ready for production. Additionally, entities creating AI agents can use payment services like Stripe without relying on cryptocurrency. The argument regarding platform risk has been previously used to justify crypto, but it has yet to materialize.
Data and Model Sources
Autonomous Management and Value Collection for Data and Machine Learning Models
Supporting Argument: Data ownership should reside with users who generate the data, rather than the companies collecting it. As data is a crucial resource in the digital era, its monopolization by major tech companies and inadequate monetization are significant concerns. A more personalized internet requires portable data and models, allowing users to transfer data across applications similar to moving cryptocurrency wallets between dapps. Blockchain technology may provide a viable solution to data sourcing challenges, particularly in light of increasing fraud.
Opposing Argument: Data ownership and privacy concerns may not be a priority for users, as evidenced by high registration numbers for platforms like Facebook and Instagram. Trust in established entities like OpenAI may overshadow concerns about data ownership.
Token-Incentivized Apps (e.g., Companion Apps)
Envisioning Crypto Token Rewards
Supporting Argument: Crypto token incentives are effective for encouraging network growth and behavioral engagement. Many AI-centric applications are expected to adopt this model. The AI companion market presents significant opportunities, with the potential to become a multi-trillion dollar sector. Historical data, such as the $130 billion spent on pets in the U.S. in 2022, suggests a strong market for AI companions. AI companion apps have already shown significant engagement, with average session lengths exceeding one hour. Crypto-incentivized platforms could capture substantial market share in this and other AI application areas.
