AI's Next Trick: Faking Identities
And how to bolster Proof-of-Personhood without sacrificing privacy
Since the public launch of ChatGPT a few months ago the world has been gripped with questions around how the new technology will impact humanity.
Lost in the discussions of the technology’s deflationary economic impact (Buzzfeed firings), and how students might use the new tech to flummox their professors (plagiarism claims), is the need to update our intuitions around what AIs are now capable of doing, and what we might expect in the short term future from ChatGPT and other AIs as they continue to advance.
AI systems have already been used to create illusionary personalities on TikTok and Youtube, piloted by their creators. But in the future, the creation of entire groups of fake people will become a common occurrence. Indeed, two years ago a criminal group successfully scammed the Chinese tax authority out of $76 million by compromising the government's facial recognition-based identity verification system. (link)
In a world where AI systems proliferate false personas the ramifications would be dramatic:
Degradation of Trust: When fake personas are allowed to exist, it becomes increasingly difficult for individuals, organizations, and governments to trust who they are interacting with online.
Compromised Security: False personas can be used to carry out malicious activities such as scams, cyberattacks, and identity theft, compromising the security of individuals and organizations.
Damaged Reputation: Fabricated personas can tarnish the reputation of real individuals and organizations by spreading false information and engaging in malicious activities.
Weakened Identity Verification Systems: When AI systems can fabricate fake personas with ease, traditional identity verification systems become less effective, lowering the overall security of digital identities.
The Dilution of the Concept of "Personhood": As the number of fake personas increases, the concept of personhood becomes diluted, making it harder to establish trust and accountability in a digital world.
The intuition we all carry currently is something like ‘reality is the low cost truth provider’. We assume a camera operator need only point and click at a moment of reality to validly stake a flag in that moment’s true existence. By contrast, we assume a Persona creator or image counterfeiter would have to expend sizable resources to counterfeit the same moment of reality.
Yet as AI capabilities expand and become cheaper, that intuition is being undermined. As AIs become better they will decrease the cost to fake an image, a video, a person, a group of people, or even a small network of people - and in doing so begin to shrink the cost between the real and the contrived to a negligible amount.
As this occurs it will erode trust in digital-forward credentialing techniques. Crypto exchange AML/KYC techniques will become more onerous, while in-person communications and credentialing increase in value on a relative basis.
Anticipating this, it’s worth thinking about ways to leverage physical space interactions to help bolster Proof-of-Personhood. Specifically by looking at a technique to expand the cost to fake a Personhood likelihood, while keeping the credentialing costs of a real person low.
Current state of the Proof-of-Personhood:
Last year in the The Network State writer’s cohort 2, Alex LaBossiere’s ‘Proof of Personhood and the Race for Identity’ essay won the first week’s competition:
In addition to showing how advanced AI had become in creating false personas, the essay offered five approaches to creating Sybil-resistant Proof-of-Personhood systems:
In-Person Events: in-person meetups for people to verify each other’s presence
Social Networks: people vouching for each other’s existence creating a personhood likelihood based on network size and density
Online Turing Tests: expanding/improving to build higher quality Turing tests
Strong identities: Trusted third party verification
Crypto Biometrics: a type of strong identity solution using “homomorphic encryption and zero-knowledge proofs to encrypt users’ data such that it never leaves their personal device”(LaBossiere).
But in each of these techniques there are flaws. Again from the article:
It’s important to note that almost all of these solutions are flawed in one way or another. If not by technical feasibility, they aren’t completely airtight, and leave us sure-but-not-totally sure of any given individual’s personhood. In this, we discover degrees of personhood. Since we can never be totally sure one way or another, we end up operating with a confidence interval of someone’s existence and uniqueness. Presumably, there might be a future where a system or network needs a certain level of confidence in you to grant you ownership or governance in it.
Embedded in the selection of these systems is the paramount goal of maintaining privacy for the individual, as the alternatives all slouch towards the Orwellian.
Various erosions of privacy would offer higher degrees of personhood, but would obviously erode the civil liberties and rights that the Western world enjoys.
What’s needed is a way to bolster Proof-of-Personhood networks without eroding privacy. Indeed with the threat of advanced AI systems now upon us, it's crucial to consider new bolstering ideas around Proof-of-Personhood.
Taking inspiration from Physics:
We can draw inspiration from quantum mechanics, which posits that all matter exists as a wave function of possible outcomes, and that it is only an observation of reality that collapses the wave function into a definite form.
Observations often come from direct observation by a conscious observer, but sometimes observations are made through scientific instruments. In these cases, the measuring instrument itself has already undergone wave function collapse which requires that another observer somewhere has already collapsed the wave of the instrument. This leads to a 'Von Neumann Chain,' a concept from the Von Neumann-Wigner interpretation of wave function collapse, which holds that all collapses eventually lead back to a conscious mind, serving as the anchor of the chain.
For ease of calculation, physicists consider a conscious mind as a single entity, but the biological view is more nuanced. It regards the conscious mind as a network of observing neurons rather than a solitary observer.
We aren't here to delve into debates on all this, but to gather insights and move forward. So with that in mind, what’s important here to note, is that a single observer anchoring reality could actually be a compact network of smaller observers with interlocking histories and close proximity.
Physical space attendance as Personhood ‘collapse’:
So let’s imagine that as Proof-of-Personhood networks develop, Physical Spaces will come with their own Proof-of-Space probabilities similar to a Person’s Proof-of-Personhood probability.
A given Physical Space will generally attain a much higher likelihood of the Physical Space being what it purports to be, than a Person will achieve in a Proof-of-Personhood likelihood. This is because that Physical Space has no need of intermittent privacy and can thereby be tied to a constant location through a record of interlocking history, similar to a Blockchain.
As a simplistic example, consider the following: the checkin desk for a co-working space is filmed 24/7 with three different recording devices all with overlapping fields of view, and all of which exist as separate computing devices (not networked so as to increase security). Each device produces a QR code related to the hash of the last few minutes of footage, and the QR code is shown to the other two recording devices and to itself. Each hash (associated to the QR code) is embedded in a given blockchain for reference. The three devices update on asynchronous clocks.
The video footage along with the QR code hashes and the Blockchain record of those hashes create an overlapping and historical record of the space linking it back in time. The video recordings could be made public to bolster the Spaces’ Proof-of-Space likelihood or kept private as needed. The space is semi-public during working hours, but also secured at all times.
Anyone who enters the space is credentialed as being there. A stronger form of which would invite the person to ‘check-in’ using the same tech as an In-Person Event from the above discussion of LaBossiere’s article.
If we map our new Physics inspired architecture onto our dilemma with Proof-of-Personhood then we arrive at a counterintuitive result.
A Person’s need for intermittent privacy puts the Person in a Personhood ‘quantum state’ when not directly observed. This, while a public Physical Space’s agnosticism towards privacy makes it much more suited to have a dense, proximate, and continuous network of overlapping mini observers.
High Proof-of-Space physical spaces are then the ‘conscious minds’ of Proof-of-Personhood networks - not people.
Frontrunning chaos
This architecture gives us what we were looking for: a low cost way to credential people which increases the cost of an AI attempting to counterfeit people to an arbitrarily high amount.
We leverage reality as the low cost truth provider, and to the extent that networks of observers within the space overlap with each other, have hashed links back to their own past, and capture ambient actions in the immediate surroundings, we create a high Proof-of-Space verification technique with a high cost to counterfeit (as any counterfeiting effort would have to both counterfeit the entire history of the space, and find a way to put the space’s history into a Blockchain). Thanks for the inspiration consciousness and quantum mechanics!
The technique as an opt-in attendance model preserves privacy for people, and culturally the technique has clear antecedents: most people don’t flinch at being recorded at a gas station, and at a more expansive level, places like the Connecticut based hedge fund Bridgewater Associates, have been recording their workspace environment for years so as to accurately attribute insights and mistakes. (link)
Additionally, the presence of these higher likelihood entities in a Proof-of-Personhood network would cause a flow-through effect of higher likelihood probabilities throughout the rest of the network (similar to PageRank scores).
Ideally an implementation of this idea would be shepherded by a technologically sophisticated group with a high amount local and global Proof-of-Personhood networking, which could then serve as a scaffolding for less sophisticated groups to plug into. Large tech companies with worldwide offices, larger nation state’s military and embassy locations, and the fledgling Network State all jump to mind as possible candidates for the effort.
The alternative of not finding a sufficiently strong Proof-of-Personhood technique would invite an explosion of false personas, eroding trust in everything but web3/blockchain transactions and in person dealings, as false personas bog down commerce and increase AML/KYC costs in all economic activity.
AIs have just started to catalyze the next steps forward in productivity for humanity. Now comes the new cost of doing business.
Until next time.
-Jack