Article
Amidst artificial intelligence hype, how many hypers have answered the question, 'what
Originally posted on LinkedIn on August 22, 2025.
Amidst artificial intelligence hype, how many hypers have answered the question, ‘what exactly is “intelligence”?’ Pop-philosopher and Silicon Valley icon, Nick Bostrom, certainly hasn’t, in his decade-old book I review in the article linked below. We keep arguing about AI, AGI and “superintelligence” as if “intelligence” were a settled term. It isn’t. Everyone’s waving at an intuition while measuring something else entirely. (Usually “how well it does this one benchmark after being spoon-fed the internet”.) We often conflate: 🏎️🧠 Speed with smarts (fast ≠ intelligent; a jet-powered potato is still a potato). 💾🏆 Memorisation with mastery (regurgitation isn’t reason). 🎩🌐 Narrow tricks with generality (party piece ≠ mind). 📊🌍 IQ-ish scores with capability in the world (life is not a multiple-choice exam). A workable definition (v1.0) Intelligence is the capacity to achieve a broad range of goals in novel, uncertain environments, using limited data and limited resources, by forming, testing and revising models of the world, individually and with others. Plain English: can it figure things out it hasn’t seen before, under pressure, without wasting time, data or electricity, and can it collaborate? How to tell if something deserves the “intelligent” label- Ask whether it shows: 🌐 Generality – performs across different kinds of tasks, not just more of the same. 🔄 Transfer & abstraction – reuses what it learnt there to solve problems here. 💧 Sample efficiency – learns from small hints, not a firehose. 🛡️ Robustness – copes when the world shifts under its feet (distribution shift, anyone?). ♟️ Causality & planning – reasons about why things happen and chooses what to do next. 🛠️ Self-correction – notices and fixes its own mistakes without being handheld. 🤝 Social competence – coordinates with humans (and other agents) without becoming a menace. 🔋 Resource frugality – does the above without burning a small power station. If your system falls over outside its training data, congratulating it for “intelligence” is like calling a sat-nav a cartographer. Why this matters Safety & governance: You can’t regulate what you can’t define. Benchmarks aren’t a constitution. R&D focus: Rewarding generality and sample efficiency changes what labs build. Human capital: Recognising social, causal and adaptive skills stops us treating people like slow servers. A pocket test Before saying “intelligent”, try: “Would I trust this to handle a new situation, with tight time/energy/data budgets, in a way that improves after its first try, and play nicely with others while doing it?” If not, perhaps it’s powerful, impressive, useful… just not intelligent. I’ve offered Definition v1.0. Tear it apart, improve it, or replace it; but let’s stop arguing about super-X before we’ve agreed what X is. #AI #AGI #Intelligence #MachineLearning #CognitiveScience #Ethics