Bots, deepfakes, and how to tell who's human online (ft. Ben Horowitz and Alex Blania)

The core problem discussed is "Proof of Human," which addresses the increasing difficulty of distinguishing humans from AI agents online. This challenge is amplified by the rapid advancement of AI, particularly with the advent of models like ChatGPT and the rise of sophisticated bots used for psyops and propaganda. The speaker, Alex, from Proof of Human, explains that the current situation is just a glimpse of what's to come, with AI capabilities expected to become superhuman and understanding humans far better than humans understand AI. The concept of "Proof of Human" aims to verify that an individual interacting online is indeed a unique human, ideally with a single, controlled account. This contrasts with the current state where platforms like X (formerly Twitter) struggle to combat millions of bots daily, often operated by a single human. A new distinction is introduced: "agent on behalf of a human," where an AI acts with explicit permission and rights granted by a unique human owner of an account. This means the platform still belongs to a verified human, even if an AI performs actions on their behalf. Several approaches to proving humanness were considered and largely discarded. The "web of trust" model, where users attest to knowing each other, was rejected because AI can easily mimic online behavior and create false attestations. Relying on government IDs was also dismissed due to privacy concerns, the potential for government overreach impacting free speech, and the global nature of the internet, where individual government systems are insufficient. Biometrics emerged as a potential solution, but with a critical distinction. Standard biometrics like Face ID or fingerprints perform one-to-one authentication (verifying a person against their own stored data). Proof of Human requires one-to-N authentication, meaning a new individual must be proven unique against all previous individuals in a network. This exponential problem requires a high degree of mathematical entropy, making simple biometrics like faces or fingerprints insufficient beyond tens of millions of users. Iris scans are considered to have enough entropy for uniqueness. A significant challenge with biometrics is replay attacks. To address this, Proof of Human separates verification (initial registration) from authentication (ongoing checks). For verification, a custom hardware device called an "orb" is used. It employs multiple sensors to prevent display-based attacks and captures iris data. This data is then processed using multi-party computation and zero-knowledge proofs to ensure privacy. The iris code is split into pieces sent to multiple computers, so no single entity possesses the complete data. A zero-knowledge proof allows a user to prove their uniqueness to a platform without revealing their identity to the platform or Proof of Human. Re-authentication poses another hurdle. While new iPhones can store signed biometric data for one-to-one checks, older or less secure devices are vulnerable to deepfakes. This suggests a hybrid approach where newer devices might handle re-authentication, while older ones might require more frequent visits to an "orb." The normalization of biometrics is expected as AR/VR systems, like Apple's Vision Pro, incorporate iris scanning. This will make using such technology more commonplace and less intrusive. The privacy aspect of Proof of Human is emphasized, stating that despite using biometrics, anonymity and extreme privacy are preserved through advanced cryptographic techniques. The need for Proof of Human extends beyond combating bots on social media. Other critical applications include: * **Dating Apps:** Ensuring users are interacting with genuine humans, not bots. Tinder is already using a verification badge. * **Video Conferencing:** Preventing deepfake impersonations during important calls, especially for high-value interactions like financial discussions. * **Gaming:** Guaranteeing players are competing against other humans, not AI that could offer superhuman performance. * **Content Platforms (e.g., YouTube):** Differentiating between human-created and AI-generated content, and verifying if audiences are real humans or bots, which impacts advertising revenue and creator authenticity. * **Creator Economy Platforms (e.g., Patreon, Substack):** Maintaining the personal connection between creators and their supporters, ensuring support goes to actual people. The speaker highlights the rapid escalation of the AI threat, stating that current issues are less than 1% of what's expected in the next one to two years. AI is becoming exceptionally good at understanding and manipulating human psychology, making it a powerful tool for sophisticated psyops. Proof of Human is currently focused on scaling its operations, particularly in the U.S. This involves three key pillars: platform integration, device distribution, and user adoption. They have 18 million verified users and 40 million total in the app. The immediate goal is to increase orb distribution, aiming for accessibility within 15 minutes for most people in the U.S. This requires significant partnerships with large retailers or even mobile orb delivery services ("Orb on Demand"). The initial reception to Proof of Human was skepticism, with many viewing the orb technology as futuristic or even outlandish. However, recent advancements in AI, particularly around "Claude bots" and "Moldbook," have shifted perception, making the problem urgent. The focus has moved from market risk to execution challenges, such as deploying devices and normalizing their use. The project also offers alternative verification methods like "Face Check" (using face biometrics with multi-party computation for anonymity) and government ID verification with NFC chips, acknowledging these are temporary solutions or have adoption challenges due to stigma. Looking ahead, the speaker believes governments will need to implement cryptographically strong identification systems to manage public finances, voting, and social programs, especially in an AI-driven world prone to massive fraud and impersonation. Proof of Human is seen as a crucial piece of this larger infrastructure upgrade necessary for democracy's survival. The project anticipates significant platform integrations in the near future, though initial rollout will be geographically focused to normalize the concept. The ultimate goal is for Proof of Human to become a ubiquitous and essential tool for navigating an increasingly AI-influenced digital world.