AI quietly invades—can cryptography be the last line of defense?

Cryptography-backed proof of humanity may allow us to regain genuine communication and once again understand what real people think and feel.

Written by: Byron Gilliam

Translated by: AididiaoJP, Foresight News

Online surveys are facing a severe "AI bot" crisis. Today, as many as 30% of survey responses come from AI-driven automated programs rather than real humans. These "AI bots," motivated by monetary rewards or aiming to manipulate public opinion and interfere with elections, are polluting data on a large scale, rendering research results in fields like economics, public health, and political science—which rely on such surveys—on the verge of becoming invalid.

In the long run, blockchain-based cryptographic technology (cryptographic proof of humanity) may be the only way to rebuild genuine communication. It can verify that there is a real human on the other end of the network, allowing us to once again hear the voices of "real people" rather than machine simulations.

Sometimes you need to know what real humans think, not simulated human answers.

"We found that the greatest enemy of communication is the illusion of communication."

— William Whyte

The latest online survey research shows that AI bots exhibit a particular preference for "high-risk behaviors."

This is what AI bots claim in questionnaires. But self-reported survey results are not reliable—who knows how these AI bots actually operate in private?

Regardless, this has become a serious problem: AI is making it increasingly difficult for us to understand what people are really doing.

Researchers in economics, public health, consumer behavior, and political science have always relied on surveys to understand human behavior.

Now, as no one answers phone calls, these surveys have moved entirely online.

But it turns out that AI is more enthusiastic about participating in online surveys than humans.

Why are they so active?

Money is the primary motivation.

To address the declining response rate, researchers have started paying internet users to answer questions, with each questionnaire sometimes offering up to $5.

This has led to a new problem: residents of low-income countries impersonating respondents from high-income countries, since $5 is still quite substantial in those regions.

Completing a $5 survey takes 15-20 minutes, and if one keeps searching for paid surveys to make money, the hourly wage is actually not ideal.

But for AI, the entire process takes only a few seconds, and the $5 reward far exceeds the cost of a few seconds of data center computation.

More concerning is that bots may distort polls to interfere with elections or deliberately create information gaps to disrupt geopolitical rivals.

Whatever the motivation, data shows that AI now accounts for 30% of online survey responses, a proportion sufficient to render survey results meaningless.

Even more severe is that 37% of web traffic comes from malicious "attack bots."

With such a massive scale of malicious bots, and CAPTCHA protections already failing, the $120 billions market research industry is facing a serious blow.

Researchers can still adopt other defensive strategies: for example, setting "decoy questions" invisible to humans to screen out AI, or filtering out abnormal responses.

But ultimately, this is a cat-and-mouse game—bots will eventually adapt to new rules.

A more thorough solution is to give up on finding real human respondents: the "artificial society" project directly replaces human survey subjects with trained bionic AI.

By using professionally designed predictive bots to replace malicious (or random) bots, "a collection of AI agents can complete experiments in minutes that would otherwise take months."

This approach is effective for testing new marketing campaigns or predicting how Midwest audiences will accept new potato chip flavors (like pumpkin spice).

But what about major events like presidential elections?

This is a question worth pondering: current phone survey response rates have dropped below 5%, and online questionnaires have become the core method for answering the following questions:

1) Voter intentions

2) Voting motivations

The astonishing accuracy of prediction markets in forecasting elections often leads people to believe that traditional polls are no longer necessary.

But prediction markets themselves still rely on poll data: the French trader who made Polymarket's Trump prediction so accurate did so only after obtaining exclusive poll data.

These polls are, of course, conducted online. At this rate, by 2028, such data may become completely invalid due to bot pollution, and prediction markets will collapse as a result.

More importantly, if polling agencies cannot understand voters' motivations, politicians will find it even harder to truly represent public opinion.

Building a collection of AI agents is unlikely to help—how can they capture the ever-changing thoughts of humans?

AI can only be trained on existing data, while the most critical variables in elections are precisely those new pieces of information that make each election unique.

For major matters like presidential elections, only direct communication with real people is reliable.

Perhaps soon, we will be left with only one last line of defense: blockchain-based cryptographic technology.

The adoption of cryptographic proof of humanity has been disappointing. For example, Worldcoin's iris scanning solution has attracted few users, perhaps because the public doesn't care whether online content is produced by humans or machines.

But survey designers care: sometimes, you must know what real humans think, not simulated human responses.

If online surveys are a form of mass communication, then AI-simulated surveys are merely an illusion of communication.

Cryptography-backed proof of humanity may allow us to regain genuine communication and once again understand what real people think and feel (to the extent that was ever possible).

This technology needs to reach near-universal adoption: conducting surveys only among those who exchange iris scans for tokens clearly lacks broad representativeness.

At the same time, enough trust must be established so that we dare to disclose private behaviors to a sufficiently secure and private system.

But until that day arrives, effective communication between humans may become increasingly difficult.