New Theory: AI's Biggest Flaws Come Down to 6 Missing Skills
A scientist says AI gets tricked and forgets because it lacks six basic survival skills.
WHAT HAPPENED: Jonathan W. Page, a researcher writing on arXiv (a public site where scientists post papers before formal review), published a 34-page paper proposing that adaptation itself has a basic structure. Instead of studying what animals do or how their brains work, he asks a different question: what must any system compute in order to keep surviving? His answer is six "primitives" — Arouse, Orient, Valence, Position, Boundary, and Attune — chosen because they seem necessary for existence, show up in creatures that evolved separately, are preserved across evolution, and can't be broken into smaller pieces.
WHAT THEY ACTUALLY MEAN: Think of them as a survival checklist. Arouse = notice something changed. Orient = figure out where it's coming from. Valence = is it good or bad for me? Position = where am I right now? Boundary = what's me and what's the outside world? Attune = sync up with what's around me. Page argues that fancier abilities like attention, memory, and decision-making are just these six combined in different ways — like how all recipes are built from a small set of cooking techniques. If true, you could compare a bee to a human to a chatbot using the same yardstick.
WHY AI RESEARCHERS MIGHT CARE: Page points at AI's most familiar failures. Chatbots "confabulate" (make up confident nonsense). They fall for "prompt injection" (hidden instructions that hijack them). They get distracted, "reward hack" (game the rules instead of doing the job), and suffer "catastrophic forgetting" (learning something new wipes out old skills). His suggestion: these may not be bugs to patch one at a time, but symptoms of missing computations. If you want an AI that lasts, he says, you need these six, not a copy of a brain. Biology's real gift to AI isn't neurons — it's the functions they evolved to perform.
THE CATCH: This is explicitly a hypothesis, not a result. There's no experiment, no data, and no AI built from it. The six primitives were picked by argument, not measurement, and other researchers may disagree with the list. Papers like this on arXiv haven't passed peer review, so treat it as an interesting map, not a proven destination. Its value today is as a checklist for thinking about why AI systems break — and a reminder that some of their strangest flaws might have fairly ordinary explanations.
- A researcher argues all adapting systems — from bacteria to chatbots — need the same six basic computations to survive and function.
- He suggests AI's well-known problems, like making things up and falling for hidden prompt tricks, come from missing one or more of these six.
- The paper is an untested hypothesis with no experiments behind it, so treat it as a useful idea rather than a proven fix.
Why It Matters
It offers a simple checklist for understanding why AI tools fail you — and hints at what safer AI might require.