The history of the term superintelligence: where the word comes from

The history of the term superintelligence: where the word comes from

Where the term superintelligence comes from, who shaped the idea, and how to separate old concepts, open hypotheses and sales claims in your company.

•

Want to apply this in your company?

Book a call with Juan →

Superintelligence is the name given to a hypothetical system whose cognitive capacity would exceed human capacity in virtually every area of interest. The word did not come from a product or a laboratory. It gathers decades of discussion by mathematicians, philosophers and authors who asked what would happen if machines began designing machines better than themselves. For a manager, the central point is to separate this older idea from the new claims that appear in commercial proposals.

Key takeaways

  • The idea of a machine that surpasses human intelligence is much older than the everyday use of the word superintelligence.
  • Three contributions are cited often: Alan Turing's speculation about machines, I. J. Good's “ultraintelligent machine” and Vernor Vinge's “technological singularity.”
  • Nick Bostrom's book organized the debate and fixed the term in public vocabulary, with a focus on scenarios and risks.
  • No AI system in use today is recognized as superintelligence, and treating current products as such leads to poor purchasing and risk decisions.
  • In practice, what matters is what the system does, with which data, under which limits and with which measurement.

Before the word: the idea of a machine that surpasses humans

The discussion begins well before the term became common. Alan Turing, one of the founders of computing, addressed in lectures and writings the possibility that machines might come to exceed human intellectual capacities, and that this would require thinking about who would keep control. It was not an engineering project. It was a philosophical speculation about the future of calculating machines, made when computers barely existed.

Another frequently cited milestone is the work of I. J. Good, a mathematician who described an “ultraintelligent machine”: a system capable of far surpassing any intellectual activity of a person, including the activity of designing other machines. If this were possible, each generation of machines could improve the next one faster than the previous one. The reasoning is logical, but it depends on premises that have not been demonstrated, and that is precisely where the debate usually concentrates.

From ultraintelligence to singularity

In the following decades, the idea acquired other names. Vernor Vinge, a mathematician and science fiction author, popularized the expression “technological singularity” to describe a point at which human models of prediction would stop working, because the intelligence involved would surpass our own. The name comes by analogy with physics, where a singularity is the point at which known laws lose validity.

This language had two effects. It helped the discussion move beyond academic circles and, at the same time, mixed technical prediction with narrative. When a manager reads “singularity” in an article, they are almost always facing a thesis about the future, not a report about a product. This distinction seems obvious, but it is the first one lost when the word appears in sales material.

Bostrom and the term's place in the debate

The term gained academic and public use with Nick Bostrom's book, from a philosopher who defined superintelligence as an intellect that greatly exceeds human cognitive performance in virtually all relevant domains. The author also discussed possible paths, such as AI systems designed from scratch, brain emulation, biological enhancement and networks of collective intelligence. Reading the book as an analysis of scenarios and risks helps avoid confusing it with a report on existing systems.

For an executive reading of the argument, see the summary of Nick Bostrom's Superintelligence for managers.

Concept, hypothesis and claim: how to classify

When a sentence about superintelligence appears in a presentation, an article or a proposal, it is worth classifying it before discussing it. A simple way to do this is the following:

  • Established concept: superintelligence is a hypothetical capacity above human level in virtually all relevant areas. There is no accepted measurement that links it to any current product.
  • Research hypothesis: when, by which path and under what control this could happen. There is serious debate, with divergent positions and little certainty.
  • Commercial claim: phrases such as “our agent already operates at a superintelligent level.” Treat these as advertising without evidence until the vendor shows what the system does, on which tasks and with what record.

It is also common to confuse superintelligence with artificial general intelligence. These are different concepts, and that difference is covered in AGI, ASI and AI agents: what your company can actually run.

Illustrative scenario: the proposal with the strong word

Illustrative scenario: the owner of a building materials store receives a proposal from a vendor promising a WhatsApp assistant “with superintelligence” to answer customers, quote products and route orders. Before signing, the manager asks what the vendor actually delivers. A simple checklist helps turn the conversation into something concrete:

  • What exact task does the agent perform, and which real conversations were used to test it?
  • Which data does it access, who can consult it, and where is it stored?
  • What can it not do, for example grant a discount, promise a delivery date or change an order without approval?
  • When does a conversation pass to a person, and how is that handoff recorded?
  • How will performance be measured, by whom, and with which success criteria agreed before launch?
  • How can the agent be switched off or rolled back without stopping service?

If the vendor's answer is a list of adjectives, the problem is not the word. It is the lack of specification.

Governance does not depend on the label

The OECD AI Principles and the NIST AI Risk Management Framework address risk management, transparency, accountability and human oversight without relying on any science fiction term. They are useful precisely because they force the question of what the system does and who answers for its errors. For applications using language models that receive messages from third parties, the OWASP list of risks is a practical starting point, especially for malicious instructions hidden inside texts.

In Brazil, the LGPD still applies to customers' personal data regardless of whether a vendor uses the word superintelligence in its presentation. The relevant question is always the same: which data goes in, for what purpose and under what control.

How we apply the method in practice

The method we use starts by mapping the current process and its baseline, without starting from the term. Then we limit the data and actions the system can touch, pilot a real workflow with records and a success metric agreed beforehand, and finally transfer the operation with a decision to expand, adjust or stop. Along this path, the word superintelligence rarely appears. What appears are the task, the limit and the evidence, which are the elements that allow a safe decision.

Frequently asked questions

Does superintelligence already exist? No system is recognized today as superintelligence. Current models perform unevenly, doing well on some tasks and poorly on others, which is why they need evaluation and supervision in real use.

Who created the term superintelligence? There is no single creator. The idea appears in work by Turing, I. J. Good and Vernor Vinge, with concepts such as the ultraintelligent machine and the singularity, and the expression was consolidated in public debate by Nick Bostrom's book.

Are superintelligence and general AI the same thing? No. General AI refers to systems capable of acting across many different tasks, while superintelligence describes a level of capacity above human level in virtually everything. Confusing the two leads to purchasing decisions based on promises nobody has measured.

Should the topic of superintelligence enter the company's strategy? For day-to-day decisions, the focus should be on the process, the data, the limits and the measurement. The debate about long-term risks is legitimate and worth following, but it does not replace the analysis of a concrete workflow. If you want a shared vocabulary for the leadership team, see the superintelligence glossary for executives.

Sources

superintelligenceAI strategyAI historyAI governancebusiness decisions