The ability of artificial intelligence is rapidly approaching that of humans, and in many subdivisions, it has surpassed humans. Although the recent rise of large language models (LLMs) suggests that these models are becoming more versatile and seem more "generally intelligent", they feel more intelligent because they perfectly learn the main interface for human interaction: language. While it's unclear whether this breakthrough will lead to artificial general intelligence (AGI) in the short term, we now have models trained to perfectly mimic high-powered humans in digital interactions. Some might say that the age of AI has already begun.
More and more people are able to use increasingly powerful models in uncontrollable ways. A steady stream of image-generating models and software to generate deepfakes are open source, and Meta's LLaMA language model has been leaked and runs on a laptop.
A notable short-term consequence is that, until recently, we believed that text-based information was sufficient to prove human identity, known as the Turing test. Bots are common on social media platforms and have been involved in manipulation for some time. However, in most cases they are indistinguishable from real human users. Today, modern AI has either passed the Turing test, or is very close to it. This will make it impossible to determine human identity by intelligence alone in the future. Additionally, recent impersonations using deepfakes have demonstrated that video-based authentication is becoming increasingly unreliable. As a result, there is no reliable way to verify a human's identity online.
However, proof of personhood in the digital realm will be a vital and possibly inevitable tool, especially in this new chapter of human history. While there are various approaches to Proof of Personhood (PoP), it is critical that this vital infrastructure prioritizes privacy, autonomy, inclusivity, and decentralization for the benefit and protection of individuals.
In response to this challenge, the Worldcoin project has launched an open identity protocol called World ID. Advanced generative AI makes it possible to digitally generate data that cannot prove a person's human identity. Therefore, the focus needs to shift to verifying human identities through real-world proofs. Tools for Humanity — a contribution to the Worldcoin project — supported the design of a custom, open-source hardware device for assessing human identity based on unique physical characteristics (i.e., biometrics). This state-of-the-art device uses multiple neural networks to verify biometrics and uniqueness without storing any image data, issuing an AI-safe PoP credential on World ID. While there are many ways to do proof of personality (different applications have different requirements), the current state of the technology suggests that this method is the only scalable, fraud-proof, and inclusive global PoP establishment mechanism.
Through PoP credentials, the World ID protocol gives everyone the ability to prove their human identity online without the need for a third party. The protocol leverages zero-knowledge proofs for maximum privacy and is ultimately governed by the people through World ID itself. Today, preliminary versions of hardware devices, mobile clients, and deployment mechanisms have been implemented, which will gradually become decentralized. Applications can interact with proofs on the protocol through a recently launched SDK. The protocol itself is permissionless and aims to eventually support diverse credential types that can be attested by anyone. World ID will be compatible with verified credential standards, allowing the diversity of personal social interactions to be represented (soul-bound tokens, cross-social data, etc.).
To date, more than 1.4 million people have participated in the World ID program. If successful, World ID will become the largest network of real humans on the Internet, open to all as a public good.
Reasons for introducing personality testimonials
Advanced generative AI introduces two mechanisms to improve online fairness, social interaction, and trust: (1) Limit the number of accounts each individual can create to protect against impersonation attacks. This is especially important for enabling digital and decentralized governance, equitable distribution of scarce resources like Universal Basic Income (UBI), social benefits, and subsidies; and (2) preventing the dissemination of AI-generated content that is nearly identical to human-created content Indistinguishable, spreading disinformation by deceit or scale.
While there is no good solution, human identity proofs can solve both challenges.
Authenticating accounts through the PoP provides natural rate limiting, essentially eliminating impersonation attacks. Naturally, people can use their credentials to authenticate bots, but on a very limited scale. For example, creating 1000 bot accounts requires finding 1000 human users willing to continually verify their authenticity.
Distinguishing between human-created and AI-generated content is more difficult. It’s worth noting that AI-generated or co-authored content isn’t necessarily undesirable, and in fact may be the opposite. It only becomes a problem when it is used to spread disinformation at scale in a credible manner.
Basically, smart checks will no longer be effective human discriminators. PoP enables users to choose to only interact with authenticated accounts or verified content. Similar to how users can filter various content types on social media (such as "follow" and "for you" pages), PoP helps filter content or accounts that are confirmed to be human. Also, it can implement a reputation system to make people reluctant to spread untrue information, whether it is generated by AI or not. This can also help prevent the behavior of personal verification bots.
Ultimately, PoPs can be viewed as the basic building blocks of digital identity. Among them, the first layer, the human identity proof, establishes a person's human identity and uniqueness. The second layer, digital identity authentication, ensures that only the rightful owner of the identity can use it. This layer answers the question "Who are you?" Finally, digital authentication, the third layer, focuses on answering the question "Who are you?"
world of personality
After building a global network of authentic and unique human identities, we will be able to improve every aspect of society.
Establishing a Fair Distribution of Wealth in the Age of AI: As AI advances, the fair distribution of value captured and created through UBI will play an increasingly important role in countering the centralization of economic power. In order to ensure that each individual is only registered once and ensure a fair distribution, a global personality proof protocol is needed.
Advanced Spam Filter: By processing only messages verified by a human, Proof of Personality lays the foundation for an advanced spam filter, cleaning up Twitter timelines and eliminating CAPTCHA or browser DDoS protection for a smooth user experience.
Reputation System: Proof of Personality enables reputation scoring by effectively preventing the creation of multiple accounts. This enables global, frictionless and collateral-free lending, especially beneficial for those without access to the traditional financial system.
Governance: Digital collective decision-making faces significant challenges. Web3 projects generally rely on token-based governance (one token, one vote), which may exclude some individuals and disproportionately support those with greater economic power. As of today, only a handful of projects have explored actual democratic voting, such as Optimism's Citizens House. A reliable, Sybil-resistant mechanism simplifies the implementation of democratic governance models based on unique personal identities, such as "one person, one vote", and expands the design space for new structures based on unique personal identities, such as quadratic voting. The Worldcoin project utilizes proof of personality to make the protocol truly governed by the people, ensuring that the project benefits everyone. While the exact governance structure such as direct voting or elected representatives requires in-depth consideration, this represents a paradigm shift that enables true decentralization. Another particularly important application is AI. To ensure that the benefits of AI are shared by all and not restricted to a privileged few, it is critical to engage inclusive participation in its governance.
Authentication: Biometric-based authentication can be part of the solution to the growing problem of digital identity theft, which can have serious consequences for affected individuals. In 2021 alone, data breaches affected 300 million people.
Fair distribution of scarce resources: Through the use of proofs of personality, key elements of modern society, including subsidies and social benefits, can be made more equitable. This is especially important in developing countries where social welfare programs face resource capture problems - false identities are used to obtain more resources than individuals are entitled to. In 2021, India will save $5 billion on subsidy schemes and reduce fraud by implementing a biometric-based system. A decentralized proof-of-personality protocol could extend similar benefits to any project or organization globally, allowing users to better share value and align with incentives.
Looking back, it's hard to believe that there was never an easy, privacy-preserving way to authenticate on the Internet. This article illustrates the need for custom biometric devices by exploring various approaches to establishing proof of personhood and digital identity verification.
Design requirements for proof of personality
The security requirements for the PoP mechanism depend on the specific application and potential fraud risks. In addition, some applications may choose to accept multiple PoP mechanisms for registration to increase coverage, as long as they can accept users to register multiple times with different PoP credentials. However, for high-stakes use cases like subsidies, AirDrop for global fairness, and UBI, a single, highly secure, and inclusive mechanism is needed to prevent multiple registrations. Depending on the context, voting and reputation scoring may also require stricter security measures.
When evaluating approaches to building a global PoP, there are several important considerations:
Privacy: Above all, privacy cannot be compromised in the name of convenience. All interactions should be anonymous by default and support multiple profiles that are not publicly linked to other platforms.
Autonomy: Users must always be in control of their accounts, their data, and how it is shared.
Fraud Prevention: Mechanisms must be able to prevent duplicate registrations. An unreliable mechanism that allows multiple registrations would severely limit the design space of possible applications, eliminating trust in use cases such as democratic governance, reputation systems, and fair distribution of scarce resources (e.g. UBI, authority subsidies, etc.).
Inclusiveness and Scalability: A global PoP should be inclusive of all. This means that the mechanism should be able to distinguish between billions of people. There should be a viable path to implementation at a global scale, and people should be able to participate regardless of nationality, race, gender or economic status.
Decentralization: PoPs are infrastructure and should not be controlled by a single entity to maximize resilience and integrity.
Continuity: Once proof of identity is granted, it must be ensured that it is difficult to sell or steal, but easy to recover. While the primary need for recovery is self-explanatory, both recovery and authentication help prevent the sale of PoP credentials and ensure that only the rightful owner can use them for authentication purposes. Despite these precautions, it is important to recognize that they do not fully protect against collusion or other attempts to circumvent the one-person-one-proof principle. To address these challenges, innovative thinking is required in mechanism design and attribution of social relations.
Feasible Personality Proof Mechanism
There are different mechanisms for establishing a global PoP. The table below compares the different approaches and their effectiveness in addressing the above requirements.
online account
The easiest attempt to establish a PoP at scale involves using existing accounts such as email, phone numbers, and social media. However, this approach fails because a person can have multiple accounts on each platform. Also, the famous CAPTCHA (often used to protect against bots) is ineffective here, since anyone can pass multiple CAPTCHAs. Even the latest implementations, with almost all major vendors moving from "marking traffic lights" to so-called silent CAPTCHAs (e.g. reCaptcha v3), which essentially rely on internal reputation systems, are limited.
In summary, current approaches cannot deduplicate existing online accounts (i.e. ensure individuals can only register once), such as account activity analysis, lack the necessary fraud prevention capabilities, and cannot defend against attacks with significant incentives, such as against established financial services operations As evidenced by the large-scale attacks.
Official Identity Verification (KYC)
In online services, proof of identity (usually a passport or driver's license) is usually required in order to comply with Know Your Customer (KYC) regulations. In theory, this could be used to deduplicate individuals globally, but it is problematic in practice for several reasons.
KYC services are not globally inclusive; more than 50% of the global population does not have a digitally verifiable ID. In addition, it is difficult to construct a privacy-protected KYC verification method. When using a KYC provider, sensitive data needs to be shared. This can be solved by using zkKYC and NFC-readable ID cards. The relevant data can be read by the user's mobile phone and verified locally after being signed by the issuing authority. Unique human identities can be proven by submitting hashes based on user ID information without revealing any personal information. The main disadvantage of this approach is that the incidence of NFC-readable ID cards is much lower than that of ordinary ID cards.
Without an NFC-readable ID card, authentication can be vulnerable to fraud, especially in emerging markets. ID cards are issued by states and authorities with no global verification or accountability. Many verification services (such as KYC providers) rely on data accumulated by credit bureaus over time and are thus out of date, unable to verify their authenticity with the issuing authority (ie authority) as there is usually no API available. Fake IDs, along with the real data used to create them, are readily available on the black market. Furthermore, due to its centralized nature, corruption at the level of the issuing and validating organizations cannot be eliminated.
Even if the authenticity of the data provided can be verified, establishing global uniqueness between different types of identity documents is still not easy: fuzzy matches between documents of the same person are extremely error-prone. This is due to the variability of personal information such as addresses and the low entropy captured in personal information. Similar issues arise as people issue new identification documents over time, with new document numbers and (possibly) personal information. These challenges lead to substantial error rates, including false positives and rejected users. Ultimately, under the current infrastructure, it is not possible to pass KYC verification to launch a global PoP due to lack of inclusiveness and fraud resistance.
Web of Trust
The basic idea of a "web of trust" is to verify identity claims in a decentralized manner.
For example, in the classic web of trust employed by PGP, users meet (via identity documents) at "key signing parties" to prove that keys are controlled by their purported owners. More recently, projects like Proof of Humanity are building webs of trust for Web3. These networks allow for decentralized verification using facial photos and video chats, avoiding the requirement of face-to-face.
However, because these systems rely heavily on individuals, they are vulnerable to human error and Sybil attacks. Requiring users to stake funds for each new user they verify or verify increases security. Doing so, however, increases friction, as users are penalized for making mistakes, making them reluctant to verify others. Additionally, this reduces inclusivity, as not everyone may be willing or able to lock up funds. There are also privacy-related issues (such as posting facial images or videos) and the risk of fraud using deepfakes, and these mechanisms fail to meet some of the above design requirements.
social graph analysis
The idea of social graph analysis is to use information about relationships (or lack thereof) between different people to infer which users are real.
For example, it can be inferred from the relationship network that users with more than 5 friends are more likely to be real users. Of course, this is an oversimplified inference rule, and projects and concepts in the field such as EigenTrust, Bright ID, and soulboundtokens (SBTs) propose more complex rules. Note that SBTs were not designed as a proof-of-person mechanism, but rather for applications that require proving relationships rather than unique human identities. However, they are sometimes mentioned in this context, so a discussion related to them makes sense.
The underlying observation of all these mechanisms is that social ties constitute unique human identifiers if it is difficult for a person to create another profile with sufficiently different ties. If creating additional relationships is difficult enough, each user will only be able to maintain one profile with rich social relationships, which can serve as the user's PoP. A key challenge with this approach is that the required relationships are slow to build globally, especially when relying on parties such as employers and universities. It's not clear in the first place how easy it will be to convince institutions to participate, especially in the early days when the value of these systems is still minimal. Furthermore, it seems inevitable that, in the near future, AI (possibly assisted by humans to obtain "real world" credentials for multiple different accounts) will be able to construct such profiles at scale. Ultimately, these approaches require abandoning the notion of a unique human being and accepting the possibility that some people may have multiple accounts and appear in the system as separate unique identities.
Therefore, although valuable for many applications, social graph analysis methods do not meet the fraud resistance requirements of the PoP mentioned above.
biometric technology
None of the above systems can effectively verify uniqueness on a global scale. The only mechanism that satisfies all PoP requirements, including reliably distinguishing crowd sizes in untrusted environments, is biometrics. In fact, Indian authorities have demonstrated the effectiveness of biometrics by implementing the Aadhaar system to deduplicate registration in social welfare schemes, saving $5 billion in annual fraud costs. Importantly, biometric systems can be implemented in a highly privacy-preserving manner, as no images need to be saved, and the verification system can even be decentralized.
biometric pattern
Different systems have different requirements. Verifying that a user is the rightful owner of a phone via FaceID is very different from verifying the uniqueness of billions of people. The main differences are accuracy and fraud prevention. With FaceID, the biometric is actually used as a passcode, and the phone makes a single 1:1 comparison to determine if the user is who they claim to be. Establishing global uniqueness is much more difficult. Biometrics must be compared 1:N with (eventually) billions of previously registered users. If the system is not accurate enough, more and more users will be falsely rejected.
Due to the error rate, the inclusiveness of the system is mainly affected by the biometric characteristics. Iris biometrics outperform other biometric modalities and have achieved false match rates of 1.2x10^14 (one false match in a trillion) two decades ago, even without recent advances in AI. This is orders of magnitude more accurate than current state-of-the-art facial recognition technology. Furthermore, the structure of the iris exhibits remarkably stable characteristics in time.
Also, the iris is difficult to modify. Modifying a fingerprint by cutting is easy, whereas precise imaging can be difficult because the ridges and valleys of the fingerprint can wear down due to manual labor. Furthermore, using all ten fingerprints for deduplication or combining different biometric patterns is vulnerable to combinatorial attacks, where different existing identities are combined to create new identities (e.g., by combining fingerprints from different people). In theory, DNA sequencing can provide high enough accuracy, but DNA reveals a lot of additional private information about the user (at least to the party running the sequencing). Furthermore, from a cost perspective, it is difficult to scale and to achieve reliable liveness detection measures. In contrast, facial biometrics have better liveness detection capabilities than DNA sequencing. But facial recognition is far less accurate than iris biometrics. This will lead to more and more false matches as the number of registered users increases. Even under the best conditions, the false rejection rate for a legitimate new user increases to double digits on a global scale of billions of people, preventing the system from being inclusive. However, even iris biometrics aren't perfect -- there's always a small margin of error. Whether a system can be built across the human spectrum that rejects no one is an open research question, but there are reasons to believe it is achievable.
verify hardware
In terms of biometric verification itself, the fastest and most scalable route will be the use of smartphones. However, there are two key challenges with this approach. First, smartphone cameras are insufficient for iris biometrics due to the low resolution of the iris over the entire iris region, which reduces accuracy. Additionally, imaging in the visible spectrum can create specular reflections on the lens covering the iris, and the low reflectivity of brown eyes introduces noise.
Second, the achievable level of security is very low. For PoP, the important part is not identification (i.e. "is someone who they claim to be?"), but proving that someone is not registered (i.e. "is this person already registered?"). Successfully attacking a PoP system does not require an attacker to impersonate an existing individual, a challenging requirement that needs to be addressed, requiring unlocking someone's phone. It just needs the attacker to look different from everyone who has registered so far. Smartphones lack multi-angle and multispectral cameras and active lighting for liveness detection to detect so-called presentation attacks (i.e. spoofing attempts) with high confidence. A widely viewed video demonstrates a seemingly effective way to spoof Samsung's iris recognition, illustrating just how simple such an attack can be without capable hardware.
Additionally, we need to set up a trusted execution environment to ensure registration starts on legitimate devices (not emulators). While some smartphones contain dedicated hardware to perform these operations (for example, the Secure Enclave on the iPhone or the Titan M chip on the Pixel), most smartphones around the world do not have the hardware necessary to authenticate the execution environment. Without these security features, basically no security is provided, and spoofing of image capture as well as registration requests is easy for a capable attacker. This will allow anyone to generate any number of synthetic registrations. Therefore, custom hardware needs to be used.
Recovery and Certification
In addition to the initial enrollment process (deduplication step), biometrics enable continuity. Even in a decentralized system, recovery mechanisms can be designed so that individuals can easily regain access to their PoP through their biometrics. In addition, biometrics can serve as proof of ownership, a concept that is often encountered in everyday life: when verifying someone's identity, the auditor not only checks the authenticity of the ID, but also confirms that the photo of the person presenting it is associated with the ID The photos above match. Similar to Face ID, performing local facial recognition on the user's phone can be used to authenticate the user, ensuring that only the rightful owner of the PoP certificate can use it for authentication. By implementing local zero-knowledge proofs on the user device, using signed image data from a custom biometric device, secure and seamless proof of ownership can be achieved, extending the security of trusted hardware to the user's mobile phone.
Make personality testimonials a reality
Consistent with the rationale presented in this blog post, the Worldcoin project built a proof-of-personality mechanism based on a custom hardware device, using iris biometrics, as this is the key to ensuring inclusivity (i.e. everyone can register regardless of their location or background) ) and the only way to prevent fraud and promote fairness for all participants. Hardware devices can issue AI-safe personality certificates. The issuance of certificates is privacy-preserving because no images need to be saved. Using this certificate does not reveal personal information because the protocol employs zero-knowledge proofs. See the dedicated privacy blog post for more details. Future development of the community can achieve continuity by allowing users to authenticate themselves, ensuring that only they can use their certificates. Furthermore, relying only on a biometric recovery process without any memory makes it possible for credentials not to be lost.
Proof of Personality certificates form the basis of the World ID protocol, an open, permissionless identity protocol that enables individuals to prove claims about themselves (i.e., certificates issued by anyone) in a self-sovereign manner. The protocol also plans to support verified credentials, and its dependencies will be fully decentralized.
Furthermore, the Worldcoin project proposes an initial implementation of a mobile client for the protocol and a deployment mechanism through independent operators. Both aspects will be elaborated in an upcoming blog post.
Using Proof of Personhood, Worldcoin is launching a global identity and financial network as a public utility that empowers individuals regardless of nationality or background, accelerating the transition to a future that welcomes and benefits everyone on the planet. The identity layer will enable humans to distinguish other humans and advanced AI online and empower individuals and enable organizations for a global digital identity. Used in conjunction with a financial layer to distribute wealth and build a global non-state UBI infrastructure for AI financing. The latter may come sooner than expected given the current progress and it will be necessary to build the infrastructure when needed.
So far, more than 1.4 million people have participated in the first small-scale phase of the protocol's founding phase. The project will soon transition to the next phase.
