Is critical thinking the human kryptonite for Superintelligence?

On 29 September 2026, President Trump signed Executive Order 14434, "Inaugurating the Era of Super Intelligence". Federal agencies must now use the term "superintelligence", or "SI", wherever they used to write "artificial intelligence". Nine days later, the President added on his social media (Truth Social) that the White House considers anyone still using the old term "THE ENEMY" (A/N: full capitalisation is verbatim). Washington has given artificial intelligence its super hero's cape.

Let's look under the cape. Section 3 of the order clearly specifies that superintelligence carries the same meaning as "artificial intelligence" under the existing federal law, the same systems, with the same definitions, and same capabilities. Nothing became more intelligent on 29 September (clearly!), the costume is the only thing that changed. A new legal definition is due by late November (Section 3.b), meaning that the label preceded the substance. As a scientist I recognize the suspicious sequence: conclusion first, evidence pending.

To be a little more precise and elaborate on that topic, the costume is not that innocent. In the technical literature, the term superintelligence refers to a hypothetical (and awaited by many) AI system that greatly outperforms humans in virtually every cognitive domain. Renaming current models "super-" is therefore a claim. In meta-research, the branch of research that studies how researchers do research, we have a word to describe such presentation of results in a more favourable way than the data support: spin. Spin in abstracts inflates the perceived benefits of treatments or impact to the study. 

I have seen what lies beneath the cape, and I am still studying it. In a study published in 2025, I tested the ability of free online AI detectors to distinguish western blots generated by ChatGPT from authentic blots. The generative model produced realistic blots on request, largely identified by detectors.  The safeguards I recommended were unglamorous: human screening of suspicious articles first, raw images shared, every processing step documented. More recently, I wrote a blog post pointing at the even increased capabilities of general purpose LLM to create western blot images. Now, I am working on other projects on related topics.

Superman's famous weakness was a green rock from his lost planet. Superintelligence as a political brand, has one too, a simple request: "show me the data" (which can also be formulated "show me your sources"), which no executive order can ban (yet, but let's keep such scenarios for dystopian oeuvres. Hopefully). Critical thinking does not harm the technology. What critical thinking defeats is the superlative, the assumption that anything labelled "super" needs no verification. Under the cape, it reveals Clark Kent: capable, hard-working, but fallible.

Words shape trust, and trust determines how much checking we do. Beyond the quirk, branding sceptics as "the enemy" has consequences. Agencies that approve drugs, fund biomedical research, assess environmental health risks, or perform financial audits will now call their algorithms "superintelligence" in official documents. Science is grounded on the working assumption that no claim is above verification, whether it comes from a pharma company, a dataset, a machine, or a president. Washington may keep the cape on its letterhead, the rest of us should keep looking carefully underneath.

References

Executive Order 14434 of 29 September 2026, "Inaugurating the Era of Super Intelligence." Federal Register 91:63129 (2 October 2026). https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/

Trump DJ. Post on Truth Social, 8 October 2026 (as reported by AFP).

Bostrom N. Superintelligence: Paths, Dangers, Strategies. Oxford University Press; 2014. https://global.oup.com/academic/product/superintelligence-9780199678112?cc=ch&lang=en&

Boutron I, Dutton S, Ravaud P, Altman DG. Reporting and interpretation of randomized controlled trials with statistically nonsignificant results for primary outcomes. JAMA 2010;303(20):2058–2064. https://jamanetwork.com/journals/jama/fullarticle/185952

Gosselin R-D. AI detectors are poor western blot classifiers: a study of accuracy and predictive values. PeerJ 2025;13:e18988. doi:10.7717/peerj.18988 https://peerj.com/articles/18988/

My recent blog post on AI in academic fraud: https://databio.ch/news-and-blog/generative-ai-changes-the-game-of-scientific-fraud

Banner created by Gemini, text drafted (bullet points) with Claude and written by (super intelligent) RDG.