The AI effect and artificial reasoning
by Miguel Lucas
Does AI reason, or does it only simulate reasoning? For many, if it doesn’t understand what it’s doing, its results can’t be trusted. What if that conclusion were far shakier than it looks?
In 1997, beating a grandmaster at chess was the ultimate proof of intelligence. Deep Blue defeated Kasparov, and the world promptly reclassified it as “brute-force search” — not real intelligence. In 2016, Go was considered impossible for a machine because it demanded intuition. AlphaGo won and was rebranded as “statistical optimization.” In 2024, AI began outperforming experts on bar exams and medical licensing tests. The reaction: “it’s just memorizing patterns.”
The phenomenon has a name. It’s called the “AI effect” 1 and describes how, every time a machine solves a problem once considered exclusively human, that achievement stops being seen as intelligence and gets reclassified as “data processing.” Tesler’s Theorem captures it in a memorable line: “AI is whatever machines haven’t done yet.” A cognitive bias that seeks to preserve human exceptionalism. What was intelligence yesterday is “just an algorithm” today. And tomorrow we’ll move the goalposts again.
The data, meanwhile, tell a different story. On real-world coding benchmarks (SWE-bench), models went from 4.4% effectiveness in 2023 to over 80% in 2025. On PhD-level science questions, from 30% to 94% 2. McKinsey estimates that generative AI could add up to $4.4 trillion annually to global productivity — not by generating images, but by automating tasks that previously required expert human reasoning 3.
And here’s my position, from the trenches of daily use: I don’t care whether AI reasons or simulates reasoning. What I care about is whether what it produces is useful. And I’m thoroughly tired — in the best possible sense — of extracting real value from its outputs. Value that translates into hours saved every day: in reading, comprehension, writing, and programming.
Is this “understanding”? In the biological sense, no. This is a massive predictive model — statistical predictions at an unprecedented scale. But hand it a complex document and ask for an explanation with genuine reflection. What you get is functionally indistinguishable from the reasoning of a highly capable person. At that point, the philosophical question loses relevance next to the economic reality.
Daniel Dennett put it precisely: competence can exist without comprehension 4. Evolution produces extraordinarily complex and functional biological designs without any agent needing to understand the process. No one disputes the result.
While the philosophical debate stays stuck on whether the machine “truly understands,” millions of professionals are already building competitive advantage on what the machine produces. When the revolution finishes unfolding, nobody will ask who was right. They’ll ask who arrived late.