The AI Tortoise and the Human Hare

When I was a kid, my dad used to read to me before bed. One story that stuck was the tortoise and the hare. My five-year-old brain was incredulous. A hare losing to a tortoise? Moving fast is always fastest, and therefore the best outcome, obviously.

I mean, come on. A tortoise beating a rabbit. My kindergarten logic was bulletproof on this one.

Decades later I am watching the same incredulous reaction play out with AI. We are all convinced faster equals better, that having AI do everything for us is obviously the winning strategy. Recent research suggests we might be racing toward a cliff again.

The cognitive cost of speed

A study published in Societies by the Swiss Business School found something kind of alarming: people who rely heavily on AI tools demonstrate significantly weaker critical thinking abilities. The culprit is cognitive offloading, which is when we delegate mental tasks to external tools instead of, you know, actually thinking.

I really do not like it when blogs reference studies but make you go read the study yourself, so the TLDR: the study surveyed 666 participants across different age groups and educational backgrounds in the UK, measuring AI tool usage patterns and critical thinking skills using validated assessments. Results showed a statistically significant negative correlation (p < 0.001) between AI tool reliance and critical thinking performance, mediated by cognitive offloading behaviors.

The pattern holds across ages and education levels. Younger participants showed higher AI dependence and correspondingly lower critical thinking scores. Sam Schechner, a reporter at the Wall Street Journal, echoed this, noticing his brain getting "rusty" after outsourcing French email writing to ChatGPT. "I was surprised to find myself grasping for the right words," he wrote.

The thing that really gets me is that this is not just an academic concern. It is a real-world version of what AI researchers call "specification gaming", when systems, including us, optimize for immediate outputs rather than broader long-term goals. We ask AI to write our emails and it delivers technically correct, soulless prose. We want faster results and we get exactly that, at the cost of our own communication skills. It is the daily-life version of the genie problem researchers worry about at scale. That old tale about getting exactly what you ask for, not what you actually want, plays out every time we prompt for a quick solution without investing any understanding.

If you think I am being dramatic about the stakes, hundreds of tech leaders, including Bill Gates and Sam Altman, recently signed a 22-word statement that deserves serious thought.

Mitigating the risk of Extinction from AI should be a global priority alongside other societal scale risks such as pandemics and nuclear war.

These are the people building this technology, warning us about it. They are thinking about existential risk to humanity. I am thinking about existential risk to our individual cognitive abilities. The underlying pattern is the same: systems optimizing for what we tell them to do, often at the expense of what we actually need.

What I have learned from my own experiments, and probably too many late nights tinkering, comes down to this. Human alone is slow at expected quality. AI alone is fast at below-expected quality. AI plus human is faster than human, slower than pure AI, and better quality than both. The magic happens in that collaboration zone. The catch: if AI does 80% of the work, I cannot engage my brain effectively. It is like editing someone else's half-finished thought, you are missing too much context to contribute anything meaningful. I have found it more productive to let AI handle 70%, then do the remaining 30% myself. I know it sounds like splitting hairs over 10%, but it makes a difference, at least for me.

Mapping your cognitive investment

When I was learning to deploy production-level web apps, the only thing I really knew was that I did not know a lot. Humbling, yes. Helpful, surprisingly so.

So I spent my 30% strategically. I mapped the entire process first, identifying what I knew, what I sort of knew, and most importantly what I had no clue about. Then while building I invested more time in actually learning than in copying AI output. I used AI for all of it, but differently. Instead of asking it to do the work, I asked for analogies, cheat sheets and explanations. Using AI to learn is honestly a ridiculous cheat code. I had effectively built a map of when I wanted to cognitively onboard rather than offload, which is exactly what researchers recommend, using AI to enhance rather than replace cognitive engagement.

The anti-offloading framework

Based on the research and a fair amount of trial and error: front-load your thinking, roughly 20% of effort, by spending serious time understanding what you are trying to achieve before you start with AI, because the research shows higher education levels protect against cognitive decline since educated users approach AI more critically, so write your own brief first. Map your knowledge gaps, about 5% of effort, explicitly identifying what you know versus do not, which stops AI filling gaps you should be learning yourself and turns it into something that explains concepts rather than just executes. Do strategic cognitive onboarding, another 5%, choosing specific moments to dive deep and learn rather than accept output, asking for analogies, mental models and step-by-step breakdowns so AI becomes a teacher, not a replacement. The 70/30 rule ties it together: let AI handle 70% of execution while you keep active engagement in the other 30%, which keeps you in the driver's seat cognitively.

Why this matters beyond productivity

The research reveals a troubling trend, especially for junior professionals. If someone learns in an environment where AI does most of the thinking, they miss developing crucial mental models and problem-solving patterns. They learn to prompt AI but not to think in their domain. That is genuinely a bit terrifying when you sit with it.

The tortoise wins not despite being slow, but because consistently improving your position is what matters most. In an AI-enhanced world that is the difference between sustainable productivity and cognitive atrophy.

So pick one task you regularly hand to AI. This week, try the 70/30 approach, let AI handle execution but spend 30% of your effort understanding and learning, mapping what you know versus do not know first, then using AI as your teacher rather than your replacement. The goal is not to slow down. It is to stay smart while speeding up, which is the only version of fast that the tortoise would recognise.