My first love was self-sovereign distributed data, where each person owns and controls their data, hosting it wherever they choose and permissioning it under their own terms. But I got lost in the complexity of building a robust distributed identity infrastructure. How can you give permission to someone if you can’t 'name' them in a way that is verifiable and resistant to subversion? There's no point in saying "only John can access this" if Tom can show up and convincingly say, "I'm John."
This issue isn’t theoretical—many modern digital problems stem from weak identity foundations. Take email, for example. SMTP, the core protocol, evolved without a strong sense of identity. Had we designed email with a robust identity layer—and maybe a little reputation—spam might have been less rampant. Instead, we've had to patch identity onto email systems, mostly at the DNS layer. Could better choices early on have changed the landscape of digital trust?
As we enter the era of AI and Personal AI, this challenge resurfaces. We will increasingly rely on agents to interact, assist, and even make decisions on our behalf. But how can we trust these agents? How do we know they are who they claim to be, and whose interests they truly serve? When I ask my AI how to unwind after a long day, it might suggest a refreshing Diet Coke. But is that suggestion rooted in understanding my preferences, or is it influenced by unseen commercial incentives?
Recognition and Identity in AI
In the animal world, intelligence is often measured by the ability to recognize oneself and others. The mirror test is a classic example—when an animal identifies itself in a reflection, it demonstrates a form of self-awareness. Similarly, recognizing specific others—distinguishing one individual from another—marks advanced cognitive development.
AI, in contrast, remains limited in this capacity. While AI excels at pattern recognition, it lacks the ability to form a persistent sense of identity, either of itself or others. This limitation restricts its ability to build trust and context in interactions. Without a foundation for recognizing specific entities, AI systems risk becoming tools of confusion or exploitation.
Embedding Identity Systems into AI
One solution is to deeply embed identity frameworks into AI architectures from the outset. Decentralized Identifiers (DIDs), Verifiable Credentials (VCs), and similar systems could provide AI with a structured way to "recognize" and differentiate entities.
Persistent Identity Chains: AI could track verifiable chains of identity, ensuring that when it reports information—like "Brad says buy this stock"—it can verify that it truly came from the Brad you trust.
Verification of Origin: By leveraging cryptographically verifiable credentials, AI can ensure that information hasn’t been tampered with and originates from a trusted source.
Reputation Frameworks: Identity systems could incorporate reputation mechanisms, helping AI prioritize information from sources that consistently meet a trust threshold.
Chain of Custody: AI could provide transparency on how information was received and processed, ensuring that its recommendations are based on data with verifiable origins.
The Path to Trusted AI
Trustworthy AI isn’t about making machines socially aware; it’s about ensuring that humans can trust the chain of custody behind AI-generated insights. When AI states that "Brad recommends this action," it should be able to prove that the recommendation came from the right "Brad"—the person you trust, not an imposter or manipulated data source.
The real question is: How do we create systems where AI is not just technically accurate but verifiably trustworthy? In an era where decisions increasingly rely on AI advice, embedding identity systems at the core isn’t just beneficial—it’s fundamental.
