How Blockchain can help Public AI
We all have heard of AI problems. AI Fairness, accountability, transparency, and ethics (AI FATE) are four general issues in AI systems. Fairness refers to the principle and practice of ensuring that AI systems operate equitably, without creating unfair biases. Transparency refers to the quality of making AI systems clear and understandable to various stakeholders. AI ethics is concerned with ensuring that the development, deployment, and use of AI technologies align with moral values, ethical principles, and societal norms. Finally, accountability refers to the principle that AI systems developers should be responsible for ensuring that these systems function ethically, transparently, and without bias.
1- Public AI
With two-thirds of AI cloud computing resources belonging to Microsoft, Google, and Amazon, we should not expect AI to serve public interests and comply with AI FATE. For the greater good of society, it’s essential to have robust public AI to offset the influence of corporate AI, along with more robust democratic institutions to oversee all AI operations.
Achieving public AI requires a proper AI governance that enforces AI FATE standards. While Brookings recently offered a centralized government-oriented solution, I’d like to highlight blockchain as a potential decentralized solution, which can help mitigating some of the AI FATE issues.
2- How Blockchain Can Help?
To outline potential areas blockchain can help public AI governance, we need to understand key steps involved in building an AI system. Data selection, data gathering, pro-FATE model alignment, and competing interests’ prioritization are key decisions that need constant oversight. More particularly, an AI governance framework must be continuously involved in deciding which data to use for training (e.g. public or private data), how to collect training data (e.g. with or without user consent), how and to whom an AI system should be aligned to receive pro-FATE feedback, and whose interest to prioritize when addressing AI harms during implementation.
Blockchain technology is a decentralized digital ledger that records transactions across multiple computers in such a way that the registered transactions cannot be altered retroactively. This technology is the backbone of cryptocurrencies like Bitcoin, but its applications extend far beyond digital currencies.
Transparency, consensus mechanism, security, and immutability are key features of blockchain that makes it a potential framework for public AI governance, by tracking the process of designing and training AI systems.
Blockchain technology can be employed to effectively monitor, verify, and document the data on which an algorithm was trained, including details about when and by whom it was trained, as well as the measures taken to vet and validate that data. This brings transparency to AI training.
The consensus mechanism guarantees that a piece of data (and its collection method) is utilized in training only when there is a consensus among stakeholders. By defining proper FATE tests and version control capability (as proposed by Casper Labs), the consensus mechanism ensures that any new model version failing to meet FATE criteria will be suspended. In such instances, the system reverts to the previously approved model version until the detected issues are fully addressed and the new version is re-evaluated for compliance. Remarkably, the selection of these FATE tests can also be recorded on the blockchain, ensuring further transparency and accountability.
Blockchain uses cryptographic methods to secure data. During the consensus phase, if a private piece of data is verified, blockchain’s security measures ensure that the data is encrypted, thereby maintaining the privacy of the data owner. Finally, the immutability feature of blockchain ensures that once data has been recorded onto a blockchain, it cannot be altered or deleted without extraordinary effort, if at all.
3- Blockchain Is Not A Silver Bullet for AI
Implementing blockchain technology in the pursuit of public AI presents notable challenges and is not a silver bullet for all AI-related issues. First, although the approach is decentralized, it needs intermediaries equipped with relevant expertise and team members to initiate, as well as incentives for stakeholders to sustain engagement. Second, blockchain on its own cannot fix issues of fairness in AI. Although it can verify the implementation of various fairness measures, only a limited number of these measures can be addressed through technical solutions. Lastly, numerous companies have yet to recognize AI governance as a comprehensive framework overseeing their AI development processes.
Coinfident is a Swiss-based startup providing analytics and security tools for crypto traders. This article is for information purposes only and represents neither investment advice nor an investment analysis or an invitation to buy or sell financial instruments. Specifically, the document does not serve as a substitute for individual investment or other advice.