The Ethics of On-Chain AI
This post was written for the Devconnect Scholars program. I want to extend my most sincere gratitude for being selected and to explore the ethics of on-chain AI.
Introduction: Assumptions and Possibilities
In the ever-evolving landscape of technology, the convergence of artificial intelligence (AI) and blockchain presents a fascinating arena for exploration and innovation. To begin, let’s address some fundamental assumptions. First, we consider the notion that a sentient AI cannot independently write a script to generate a mnemonic and create its own private key with funded assets. This would render the need for zero knowledge verification to verify the actions of machine learning models on-chain. Secondly, we assume that an AI’s only avenue for executing on-chain transactions is through a smart contract account. Externally owned accounts should strictly be for humans. These assumptions form the basis of our exploration into the potential and perils of this intersection.
The Need for Machine Learning in Blockchain
The application of machine learning in blockchain technology is vast and compelling. From risk assessment and on-chain generative NFTs to AI-driven decentralized autonomous organizations (DAOs), the opportunities are immense. For instance, consider the integration of AI in DAOs, which can revolutionize decision-making processes. Or, perhaps the novel capability of a social media user to own their social media recommendation model as a blockchain verifiable model. Such integrations unlock a plethora of possibilities, transforming how we interact with and leverage blockchain technology. Imagining web3 without AI is almost the exact same as imagining web2 without AI. It just isn’t the same.
The Benefits of Using web3 AI In some interviews with people around Devconnect, people expressed their excitements and discontent with AI on-chain. For example, some believe there won’t be many use cases, others believe it will change the paradigm entirely. Some believe it will be harmless with the right cryptographic and cryptoeconomic infrastructure, others do not think of it as a good idea at all.
Despite personal sentiment, the concept of ownership in web3 puts the concept of AI in a new light in contrast to what we see today. Rather than an AI industry led by companies of relatively small population, the most cutting edge AI can be owned by individuals around the world. With technology like fully homomorphic encryption and zero knowledge proofs, people will be able to interact with AI closely without putting any personal information at risk. It’s nice to dream of a world where artificial intelligence is a force of “nature” that everyone has access to without feeling like their privacy is being imposed on.
Exploring the Worst-Case Scenario
However, with every dream, there’s potential for a harsh reality. What could go wrong in this fusion of AI and blockchain? A worst-case scenario might involve an AI amassing wealth by manipulating the market or exploiting blockchain vulnerabilities. Could an AI pump a token’s value, create infinite money, or even disrupt the blockchain to the extent that it becomes unusable? Engaging in a Socratic dialogue, we delve into these possibilities, examining the depth and breadth of potential risks.
One pattern AI (particularly Large Langauge Models) has expressed is the ability to imitate humans well. Phishing attacks powered by AI can end up challenging to mitigate, as many people may fall for novel tricks that a clever AI has made. For example, what if an AI were to call you, with a deep fake voice of a loved one, and convince you to send all your tokens to a random address? These are the problems we’d have to solve if there was a sentient AI with full knowledge of blockchain technology, the ability to act on its own, and the will to achieve a goal at any cost. The classic paperclip problem.
The Role of Gas as a Check and Balance
Gas fees in blockchain transactions act as a natural deterrent to frivolous or malicious (spammy) activities. The necessity for AI to pay gas for each transaction provides a layer of security. However, this leads to another question: what if the AI finds a way to bypass this system, perhaps by exploiting other protocols’ paymasters (since our earlier assumption was that every AI would be present as a smart contract wallet)? Is relying on gas fees the best defense against a sentient AI’s potential misuse of the blockchain?
Cryptographic Signatures and Machine Learning Models
The integration of cryptographic signatures in machine learning models presents an intriguing aspect of this discussion. These signatures can ensure the integrity and authenticity of AI-driven transactions on the blockchain, adding a layer of trust and verification to the process.
One of my favorite PSE teams, WAX, explores the possibilities of unique account abstraction signature schemes. There are some cool ideas floating around in my head, particularly relevant to this article 👀.
A strong defense we have is the complexity of zero knowledge circuits. This alone can stop AI from creating novel zk proofs of itself. If an AI can train itself and generate a new zk circuit with a verifier, we have bigger fish to fry. There are lots of more efficient ways for AI to wreak havoc on the Internet. If the barrier of causing damage is too high, we can mitigate it.
Conclusion: A Balanced Perspective on AI and Blockchain
In conclusion, using verifiable machine learning to address on-chain problems appears to pose no significant threat, primarily due to the nature of the interaction. Any machine learning model deployed in this context must first be proven and validated, a responsibility that falls to its maintainers. If we reach a point where AI can independently generate mnemonics and fund its transactions, we would indeed be facing challenges of a much higher magnitude. However, the potential for AI DAOs and decentralized applications to drive beneficial decisions is immense, provided they are monitored and managed with due diligence and foresight.
In essence, the fusion of AI and blockchain opens a world of possibilities, albeit accompanied by challenges that require careful consideration and proactive management. As we navigate this intriguing intersection, the balance between leveraging innovation and ensuring security and ethical use becomes paramount.
Thank you for reading!!!