You Already Have AI — Wozniak Said So at Graduation
Apple co-founder Steve Wozniak told graduates they already have AI — actual intelligence. Here's what that pun means in practice for Sri Lankan engineers and students.
Steve Wozniak's graduation speech pun — telling students they already possess "AI," meaning actual intelligence — drew a cheer from the crowd, according to Business Insider. It is easy to read it as a clever one-liner from a tech legend and scroll past. It is harder, and more useful, to ask why a room full of graduates cheered that particular line in 2026.
The anxiety underneath the applause
Students graduating into this labour market are not applauding an abstract philosophical point. They are applauding because many of them have genuinely wondered, at some point in the last two years, whether their degree still leads anywhere. AI tools can write code, essays, legal briefs, design specs, and test suites. The tools are improving fast. And if you are a Computer Science graduate watching that happen, the question is uncomfortable.
Wozniak's line works as reassurance. But it only works if you parse what "actual intelligence" means in practice, because "you're smart, don't worry" is not an actionable answer. The more specific version of his argument is something like: automated tools are good at producing outputs given clear inputs. Humans are good at knowing which inputs matter, whether the output is correct, and what question to ask next. That distinction is doing a lot of work.
Where AI runs out and people start
Language models are trained on text. They know what has been written down. They are weaker on what has not — local regulations buried in government PDFs, professional norms that travel by word of mouth, domain-specific edge cases that have never been Stack Overflow'd.
Here is a concrete example. Sri Lanka's APIT (the withholding tax on employment income) has specific brackets, reliefs, and an exemption threshold that changed in the 2025/26 assessment year. If you ask a generic AI assistant to calculate APIT for a given salary, you may get plausible-looking output that is subtly or significantly wrong. The model has no reliable way to know whether it has current brackets or a stale version. You, as someone working and living here, know to verify against the IRD's published tables. That is actual intelligence. The [Sri Lanka [Income Tax](https://induwara.lk/tools/tax-calculator-global) Calculator](https://induwara.lk/tools/sri-lanka-tax-calculator) on this site cites the IRD directly for exactly this reason: the calculation is only useful if the rates are current and verified by a person who knows where to look.
This scales across almost any local domain. Sinhala and Tamil language edge cases, NIC number formats, local court holidays, merit-based school admission cutoffs, bank branch settlement calendars — the knowledge needed to build something genuinely useful for people here is underrepresented in global training data. The person who holds that knowledge and can direct a tool toward it is doing something the tool cannot do on its own.
Actual intelligence is about evaluation, not just creation
There is a more fundamental point here than local knowledge. The thing that makes an engineer useful in a world with capable AI tools is not raw generation speed — models have that covered. It is the ability to evaluate outputs, spot failures, and catch the confident-sounding wrong answer.
An AI assistant writing code has no stake in whether the code is correct. It has no memory of the last bug this class of error introduced. It does not know that a previous version of this same function caused a production incident last quarter. The engineer who runs that output through her own mental model, tests it against known edge cases, and asks "but what happens when this value is null?" is providing something the tool cannot supply itself.
This is Wozniak's point made operational: actual intelligence is what you use to evaluate what the tool generates. Generation is the easy half now. Evaluation is where judgement lives.
What this means for you
If you are an engineering student or developer working with local systems, three things follow from this framing.
Invest in domain knowledge, not just tooling. The person who knows how EPF contributions stack, or exactly how a working-days calculator should treat mercantile versus government holiday calendars, can direct an AI toward a correct solution. The person who only knows how to prompt has no way to verify the result.
Do not mistake fluency for correctness. AI outputs are often fluent. They read well. That is precisely what makes errors harder to catch. The skill is learning to pause on outputs that sound right and check whether they actually are — especially in areas where you know the ground truth.
Build things that require local knowledge. The domain where AI assistance is weakest is also where a human builder adds the most value: local financial tools, regional language support, compliance-specific calculators, systems that have to be verified against government sources that are Sinhala-only or not indexed online at all. That is the gap where actual intelligence still clearly outruns the artificial kind.
The cheer at Wozniak's graduation speech was not just for the wordplay. It was for the reminder that the skills worth building are the ones hardest to replicate automatically. That list is shorter than it used to be — but it is not empty. Know your domain. Verify the output. Own the question.