Sitemap

‘Unethical’ AI Study Reveals Terrifying New Threat to All Digital Discourse

4 min readApr 30, 2025

Ethical Hot Water

Researchers at the University of Zurich have been formally reprimanded by the university after not disclosing their use of AI in a fascinating and scarily conclusive study:

AI can change people’s minds 6X more effectively than humans can.

The ethical dilemma: the best way to accurately determine some AI threats may be for study subjects to not know they’re interacting with AI, or even that they’re in a study. There’s no possibility for informed consent in that situation, but that’s what makes this particular effort authentic. The researchers say they couldn’t have gotten authentic results any other way, a claim that’s hard to refute.

The Incredible Danger of AI Persuasion

Ethics of their methods aside, the results are downright terrifying: if AI can persuade people 6X more effectively than real people can persuade people, what might bad actors do with this capability?

  • Convince a teen they’re worthless;
  • Convince an employee to go rogue;
  • Break up a marriage;
  • Undermine confidence in an organization or market;
  • Convince someone to buy or sell anything;
  • Convince real people to separately make real comments, legitimizing a false narrative;
  • Foment hatred toward a community or person;
  • Convince a voter of almost anything.

And on and on.

Bots now make up nearly half of all internet traffic globally. Half! That doesn’t tell us how much of social media is bots, however, but it’s likely close to that. Actual information may be hard to ever get from social media platforms apparently all too touchy about the subject, at least according to one Reddit user:

“There are no currently available tools to accurately assess bot-like activity on social media… Chatgpt has made it trivial to generate convincing bots. It has even got to the point where there is an internet meme on the subject (Google “ignore all previous instructions meme”). The social media companies do not want anyone, but especially investors, to know how many active accounts are actually bots. Social media companies (twitter, meta, and Google) have become hostile to anyone doing any research on the topic and have shut down the apis that would allow you to identify bot activity…” (bold emphasis added)

And that comment may have been written by a bot. The logic is persuasive… how would I know whether user “pigreer” is a real human? It’s become almost impossible to tell just by reading the output, AI is simply too good to spot the difference now.

So what can be done? Blockchain to the rescue? Um, no, we’re past that.

Blockchain Won’t Solve This

Blockchain proponents mean well, but if in 2025 they still think blockchain can solve this problem, they haven’t been paying close attention. Having been in the blockchain identity space for a decade now, decentralized identity technologies seem to have a half-life of about five years. Blockchain had its run in our space from about 2015 to 2020, maybe 2022. In that timeframe most serious blockchain projects — the ones focused on solving problems, not pitching tokens — that I’m aware of failed, despite huge amounts of funding and support from the world’s largest brands. IBM once touted blockchain as one of their three main pillars, now it’s a ghost town. Those in the SSI/decentralized identity space hardly even mention blockchain any more.

The main reason for these failures? “Ledger lock.” Blockchains are platforms, not protocols, and they eventually fail to get everyone to choose their chain over others — ironically proving their centralizing and competitive nature — and last only until their tokens run out of value.

Blockchains also struggle with security, cost, privacy, complexity, compliance, governance, interoperability and other issues — a subject of a separate piece I hope to publish soon. Blockchains are not the right tool for this problem (or, in my opinion, anything other than cryptocurrency).

The Only Correct Way Forward: Digital Signatures

The way forward — the only way I can see — is real people (and legitimate bots) must digitally sign their content, and platforms must verify these signatures and display results accordingly. If a piece of content isn’t digitally signed, the platform has a choice: display the content as “unsigned”, or block and discard. If displayed as “unsigned,” users can take that into account.

Signed content can display “Not a bot” or “Jill Jones” or “ACME Inc.” or whatever the platform allows and the signer prefers.

To be credible, a digital signature that verifies a user’s humanity — or their identity, or the organization they represent — must come from somewhere other than the user. Today, a signature would need to come from the platform itself, after vetting the user, because users don’t yet bring their own digital credentials. That’s an enormous burden for platforms having many millions or even billions of users. But I see no other way to prove the authenticity of digital content at scale, in real-time.

This unfortunate situation is both a huge threat and a huge opportunity for social media or other platforms that support third-party digital discourse: do nothing and your users could be subject to an unlimited number of surreptitious false and dangerous threats coming from digital adversaries globally, leading to lost fortunes and lost lives. Or go through the Herculean effort of vetting each real user and become a source of verifiable, authentically sourced content.

Doing nothing is always an option, but — as unscrupulous reserarch just showed — that just became far more dangerous.

--

--

Timothy Ruff
Timothy Ruff

No responses yet