Personhood Credentials: Everything to Know About the Proposed ID for the Internet [CNET]

As AI bots get better at impersonating people, a group of researchers is offering a solution to combat online deception.

What if you had to present an ID to use the internet? 

That’s the premise behind a new paper by AI researchers from well-known institutions including OpenAI, Harvard University and Microsoft advocating for personhood credentials, or a new method to prove our humanity online.

The researchers argue increasingly capable AI — which can simulate people in video chats, express human-like experiences and take human-like actions including figuring out how to trick security measures such as CAPTCHAs — lets bad actors orchestrate more effective and nefarious schemes.

“The main animating worry of this paper is we don’t have amazing solutions now, honestly, because bots can impersonate people on the internet,” said Tom Zick, project fellow at the Berkman Klein Center for Internet and Society at Harvard and one of the paper’s authors. “The solutions that we have might not hold up to more advanced AI systems.”

There were more than 3,200 data breaches in the US in 2023, a whopping 72% increase over 2021, according to the nonprofit Identity Theft Resource Center. The majority were cyberattacks. As cybersecurity incidents become more prevalent, researchers say there’s a risk that AI-based activity could take over the internet. And they argue that the best tool to address this challenge while balancing anonymity and trustworthiness for the people involved is giving us an entirely new human credential.

If you’re wondering what these credentials would be like and how they could impact you, here’s what you need to know.

What is a personhood credential?

A personhood credential would be proof that you’re a human being and not a bot. You could then use these credentials to access online platforms and digital services.

“There are two things that AIs can’t do right now,” said Wayne Chang, CEO of digital credential startup SpruceID and another author. “One of them is show up to places in person. The second one is they can’t break cryptography, that we know of.”

Personhood credentials could come in multiple forms, but they will all generate cryptographic proof to verify you. (Cryptographic proof uses cryptography — a method for secure communication via algorithms and secret keys that can encrypt and then decrypt data — to verify the authenticity of data.)

It could be a certificate in your web browser.

It could be linked to your biometric data, like your fingerprint, iris, face or voice.

Or it could be a blockchain-based token like a nonfungible token, or NFT. (Blockchain is a secure database that maintains records, called blocks, that can be used to make data immutable, or unable to be changed. NFTs are digital assets stored on a blockchain.)

Why would we need personhood credentials?

The argument for personhood credentials stems from the need to determine if online content was created by a person or a bot.

Bots have degraded the internet by spreading misinformation, committing fraud and causing services to crash. According to the paper, there is substantial risk that deceptive AI activity could eventually overwhelm the internet as AI helps make this activity more convincing with human-like content, avatars and actions. 

Being able to distinguish a person from a bot would reduce deceptive activity like fake accounts on social media and dating platforms, as well as in online marketplaces where bots buy up products like tickets and sneakers and resell them at higher prices.

“These personhood credentials ensure that only real people create an account and therefore you can increase the integrity of the communities that they’re in,” said Jason Alan Snyder, global chief technology officer at advertising agency Momentum Worldwide.

There’s a training data angle too.

AI and machine learning models rely on data generated by human beings.

“If bots are contributing to that data, it degrades the quality of the models and it leads the models to then make really shitty decisions,” Snyder said. “Personhood credentials can filter out all that bot-generated data and improve AI accuracy, ethics and fairness.”

How would we get them?

The credentials could come from multiple sources.

One is a government office like the DMV or post office, which can verify your identity through government-issued documents and then issue a digital ID or passport.

Another option is your bank, university or workplace — or even theoretically a retailer like Costco, if you’re a member, Snyder said.

Tech companies such as Apple and Google could also issue personhood credentials tied to their existing identity systems, like your Apple ID or your Google account.

“Most authors in the paper would agree that we don’t want to say that [one source like] the post office is going to assert humanity for everyone,” Chang said.

Would we need one personhood credential or multiple?

That’s still unclear.

According to Zick, one personhood credential that applies across the board would be easier than managing multiple credentials. But no one has quite figured out this part yet.

What about privacy and security?

The paper’s authors want a credential that can be private, so you’re not sharing anything about yourself that you don’t want to. But they didn’t find an existing option that toes the line between privacy preservation and personhood verification, Zick said.

It’s another detail that still has to be worked out.

“We’d like to see that you can maybe just use the fact that you bought eggs and then that created this breadcrumb that you can use and it could be privacy preserving,” Chang added. “You can say, ‘I was a human at a grocery store. I’m not going to tell you which one, but I was there,’ and that’s enough for you to post a comment on the Internet.”

Would this be unique to the US or on a global basis?

Bots can originate from any country.

According to software company Netacea, the UK, US, Russia, China and Vietnam are the countries that create and deploy the most bot attacks. Further, IT company Thales Group found that bad bots account for nearly one-third of all internet traffic now.

To distinguish humans from bots in a global context, worldwide adoption of personhood credentials would be best, Snyder said. But that’s a tall order that would require some kind of universal standard, not to mention trust between nations.

“If every country adopted a personhood credential, it would make it much harder for bots to operate across borders because they would lack the necessary credentials to interact with systems and platforms,” he added. “It offers all kinds of inclusivity and fair access, and it gives you a global platform for social media networks and e-commerce and all the rest of the stuff.”

What are the legal and ethical considerations?

Adoption would also require new regulations to ensure personhood credentials respect privacy, security and human rights.

Snyder admitted this is a pipe dream, but said, “It would be really cool if we could get to something that approximated it because it would cut out a lot of this horseshit and it would certainly make AI a lot safer for the world.” 

How far off is this?

Chang and Zick say most of the technology for personhood credentials already exists, but they estimate it will be two to 10 years before we see them in the real world.

The devil’s in the details of implementation, Chang said.

“I think a lot of it is going to be when there’s an actual demand for it,” he added. “We haven’t seen the internet get overrun by AI just yet, so I think it, as many things are, might be a reactive thing.”

Snyder predicts we’ll see early adoption in sectors where secure identity verification is crucial, like finance, health care and the government, in the next one to three years. Closer to 10 years from now, it will become a standard on par with two-factor authentication today.

“It’s not an immediate thing, but it’s something that we think is an immediate problem,” Zick added.

What do critics say?

Not everyone is on board.

Jacob Hoffman-Andrews, senior staff technologist at the nonprofit Electronic Frontier Foundation, called personhood credentials “wildly dystopian,” with the government potentially deciding who can speak online.

“It presents the idea that there will be multiple potential issuers of personhood, but really, in the context of the paper, governments seem like the main target, and governments have historically been very bad at attributing personhood to everybody,” he said.

He called for a better solution to the problem of AI misinformation.

“To some degree, it’s all there in the name,” Hoffman-Andrews said. “Nobody should be the credential of their personhood. Your personhood is an innate feature of your being. It can’t be granted to you by anyone else.”