Decentralized Information Systems and the Future of AI
Introduction
The FTC’s recent probe into OpenAI and the introduction of the new European AI Act emphasize the urgent need for accountable information tracking systems for use in AI model training and in managing AI outputs. Decentralized information systems offer a solution to address these issues. While decentralized money the first and most well-known decentralized information system use case, often steals the decentralization spotlight, the value and potential of decentralized information systems extend far beyond the realm of money. Beyond finance, decentralized information will have significant impact on most sectors of society and it is decentralized information, not money that may have the most profound impact on our societies in the long run. In no field is this more pronounced currently than the field of AI, where decentralized information systems could have immense significance. This blog post explores their importance and the challenges that must be overcome for a successful AI and decentralized information ecosystem to thrive.
The Power of Decentralized Information Systems
While the association between decentralized information systems and cryptocurrencies is well-established, their role in the future of AI is often overlooked. Decentralized information systems offer a range of advantages and opportunities for the AI landscape. Cryptocurrencies will for example play a crucial role in facilitating autonomous machine-to-machine micro-transactions needed in the AI economy. However, the most significant short-term impact of decentralized information systems on AI lies in two key areas: upstream AI training and downstream inference tracking.
Upstream AI Training: Controlling Information Flow
The quality and diversity of data used for AI training significantly impacts the performance and capabilities of AI models. Decentralized information systems can play a vital role in controlling and managing the data used for AI training. They can enable AI developers to regulate and track which datasets have been used in AI training. By empowering individuals to contribute and share their data securely decentralized systems can also create more comprehensive and representative datasets, ultimately leading to improved AI models. Furthermore, by decentralizing access to data, individuals and organizations can maintain greater control over their information, enabling them to selectively contribute to training sets while preserving privacy and security. This empowers data owners to decide how their information is used, leading to more responsible and ethical AI development. Decentralized Information Networks (DINs) can also enable high levels of upstream auditability and proof of compliance for AI systems aiding developers and regulators in ensuring AI systems are following regulations such as GDPR and the new EU AI Act.
Downstream Inference Tracking: Machine vs. Human Deductions
In the AI landscape, distinguishing between machine-generated inferences and human deductions is also vital for transparency, accountability, and trust. Decentralized information systems can facilitate this distinction by providing tamper-proof records of the information used throughout the AI supply chain. By leveraging decentralized information technologies, AI systems can ensure that their inferences are traceable, auditable, and verifiable. This fosters confidence in their results and helps prevent biases and hallucinations from propagating through AI ecosystems. Cryptography can also be used to regulate access to inferences on a rules and systems basis that can be automatically and immutably enforced by DINs allowing for full transparency in AI system design even in sensitive areas such as law enforcement, employment and education. Inference tracking can also be used to ascertain when and how AI has been used and what percentage of a decision process is AI versus human driven again enabling greater trust.
Challenges and Opportunities
Despite the immense potential of decentralized information systems in making AI development safer and more accountable there are significant challenges that must be addressed to foster a meaningful decentralized information ecosystem to support AI.
Data as an Asset: Rethinking Metrics
Data, as an asset class, is still not fully understood by today’s data scientists and information systems engineers. Traditional metrics that rely on data size alone do not adequately capture the value and significance of different types of modern data. For instance, a video of a cat dancing on a table might occupy several megabytes, while a social security number or a private key could be just a few bytes. Clearly, the latter contains far more value and sensitivity. To create an effective decentralized information ecosystem, we must develop new metrics that capture the diverse attributes and nuances of different types of data.
Designing for Non-Financial Transactions
The current design of decentralized information systems primarily caters to financial transactions, rendering them ill-suited for non-financial information assessment. The value of money lies in its numerical representation, allowing for straightforward comparisons and calculations. I.e. money is a one-dimensional information asset whose primary metric is quantity. To unlock the full potential of a decentralized information ecosystem, we need to expand the complexity of information assets and develop systems that can manage more complex types of information assets.
When we shift our focus to personal data for example and, particularly identity-related data, we encounter a multifaceted landscape that transcends a one-dimensional quantitative measure. Personal data encompasses a wide range of attributes that are crucial for determining its relevance, including sensitivity, veracity and provenance for example. Unlike money, personal data possesses multiple dimensions of significance beyond size that must be considered within a holistic decentralized information system.
Understanding and incorporating ways to measure these additional dimensions of data within a decentralized information systems is crucial for building AI-ready information assets. By adding sensitivity, veracity, provenance, and contextual relevance into the powerful capabilities of decentralized information systems, they can enhance privacy, trust, and data quality. This, in turn, empowers individuals to exercise greater control over their data, facilitates reliable AI training, and enables the development of AI applications that are compliant with data protection regulation while providing valuable insights.
Conclusion
Decentralized information systems hold immense potential in shaping the future of AI. By empowering individuals and organizations to control data flow for AI training and enabling transparent accountability in inference tracking, these systems enhance privacy, trust, and data quality. Overcoming challenges related to data metrics and expanding the capabilities of decentralized systems will unleash their transformative power. Together, decentralized information systems and AI will pave the way for an ethical, responsible, and legally compliant future, driving innovation and societal progress.
Dele Atanda is founder of metaMe, the world’s first self-sovereign AI and ‘clean data’ operating system and chairs The Internet.Foundation, an NGO dedicated to advancing the ethical use of data in commerce. Dele has led innovations for Fortune 100 companies like IBM, BAT and Diageo that have become gold standards for engagement within their sectors.
Dele’s critically acclaimed best seller, The Digitterian Tsunami: Web 3.0 and the Rise of the N.E.O Citizen, published in 2013 established him as a thought leader on web 3 and his award winning crypto-media franchise, metaKnyts on the battle between artificial intelligence and intelligence amplification for the soul of the metaVerse, is world’s first CryptoComic franchise.
