AI effect
Based on Wikipedia: AI effect
In 1997, when IBM's Deep Blue defeated world chess champion Garry Kasparov in a decisive match, the global reaction was not a unified celebration of machine intelligence, but rather a rush to diminish it. Within days of the victory, the narrative shifted from "a computer has learned to think" to "a computer simply calculated faster." Critics and observers alike began to argue that Deep Blue did not possess true understanding or creativity; it had merely executed a brute-force search through millions of possibilities. The victory, once hailed as the pinnacle of artificial intelligence, was instantly reclassified by many as mere computation. This moment encapsulates a recurring paradox in the history of technology: the very moment AI succeeds, it ceases to be considered intelligence at all.
This phenomenon is known as the AI effect. It is not merely a semantic quirk; it is a fundamental cognitive and sociotechnical pattern that defines how humanity perceives machine capability. As machines acquire new skills—solving complex equations, recognizing faces, or composing poetry—the definition of "intelligence" shifts to exclude those newly achieved feats. What was once the holy grail of artificial cognition becomes, in the eyes of the public and often the experts, just a routine engineering problem. It is as if the goalposts of human uniqueness are constantly moved backward, ensuring that we never have to concede ground to our creations.
The Moving Goalposts
The mechanism behind this effect is elegantly summarized by Tesler's theorem, a maxim often attributed to Larry Tesler but widely circulated in AI circles: "AI is whatever hasn't been done yet." This definition creates an impossible standard for artificial intelligence. Under this logic, any task that remains unsolved is the domain of true intelligence, while any task that has been solved is merely algorithmic computation.
Historian Pamela McCorduck documented this recurring feature in her 2004 book, Machines Who Think. She observed a consistent pattern: once an AI system successfully solves a problem, the academic and public consensus immediately reclassifies the achievement. The solution is absorbed into other domains—mathematics, statistics, engineering—and stripped of its "intelligent" label. The logic is circular but deeply ingrained in our cultural psyche. If a machine can do it, it isn't thinking; if it were truly thinking, a machine wouldn't be able to do it yet.
Researcher Rodney Brooks, a prominent figure in robotics, captured this sentiment in 2002 with a sharp observation: "Once systems are understood, they are often regarded as 'just computation'." The moment we understand the mechanism behind a capability, the magic evaporates. We no longer see a mind; we see gears turning. This is not simply a matter of technical pedantry; it reflects a deep-seated need to preserve a special role for humans in the universe.
The Paradox of Success
The AI effect creates an "odd paradox," as McCorduck termed it. Successful AI systems are assimilated into the background of daily life, leaving researchers with no new problems that still carry the label of "AI." This forces the field to constantly chase the horizon. When a problem is solved, the definition of intelligence retreats further, pushing the boundary of what counts as "real" thinking to more complex or abstract domains.
Consider the trajectory of game-playing algorithms. In the mid-20th century, computers playing games like checkers and chess were viewed with awe. These tasks required strategy, foresight, and a form of logic that seemed distinctly human. Early systems capable of these feats were hailed as demonstrations of machine intelligence. But as these systems improved and their inner workings became transparent to computer scientists, the narrative changed. The victory was no longer seen as a triumph of intellect but as a triumph of processing speed and memory.
The defeat of Garry Kasparov by Deep Blue in 1997 is the quintessential example of this reclassification. While the event was a watershed moment for AI research, the immediate critical response focused on the limitations of the machine's "mind." Critics argued that Deep Blue relied on brute-force methods—checking every possible move rather than understanding the game in the way a human does. By defining intelligence as something that requires intuitive understanding or "true" comprehension rather than calculation, skeptics could maintain that the machine had not actually won. It had just computed faster than any human ever could.
From Core AI to Routine Engineering
This pattern of reclassification extends far beyond board games. Technologies that were once considered the core problems of artificial intelligence have gradually faded into the background as standard engineering solutions. Optical character recognition (OCR), for instance, was a holy grail in the early days of AI research. The ability to take an image of text and convert it into editable data seemed like a profound cognitive task, requiring pattern recognition akin to human vision.
Today, OCR is ubiquitous. It powers everything from digitizing historical documents to scanning receipts on smartphones. Yet, almost no one describes it as "artificial intelligence" anymore. It is treated as a solved problem, a routine utility. Similarly, speech recognition was once the cutting edge of AI research. Systems that could understand spoken commands were viewed as miracles of cognitive science. Now, with voice assistants integrated into cars, homes, and phones, these capabilities are often dismissed as mere statistical modeling or signal processing.
Michael Swaine reported in 2007 that advances in AI are frequently presented under different banners once they become practical. Marvin Minsky, a founding father of the field, noted that successful AI innovations often evolve into separate disciplines. When a technology works reliably and is widely deployed, it loses its "AI" label. Nick Bostrom echoed this in 2006, observing that widely adopted technologies are rarely labeled as AI by their users or developers.
This shift has profound implications for how we view progress. If every success story is immediately rebranded as "just math," the public perception of AI advancement can become distorted. We see a field where nothing ever quite works, because everything that does work is quietly removed from the category of AI. The visible face of the industry becomes dominated by unsolved problems and theoretical challenges, while the solved problems—the ones actually changing the world—disappear into the infrastructure of daily life.
The Cognitive Bias of Human Uniqueness
At its heart, the AI effect is often interpreted as a form of cognitive bias. It suggests that human expectations of intelligence are not static; they shift dynamically based on what machines can achieve. When a machine fails to do something we consider intelligent, it confirms our belief in the uniqueness of human cognition. But when a machine succeeds, we redefine the task to ensure our superiority remains intact.
Michael Kearns suggested that this behavior stems from a desire to preserve a special role for humans in the cosmos. If a machine can paint like Van Gogh, play chess like Kasparov, and write code like a senior engineer, what is left that makes us unique? By reclassifying these achievements as "computation" rather than "intelligence," we maintain a psychological buffer. We tell ourselves that while machines are good at processing, they lack the spark of true understanding, consciousness, or creativity.
This pattern is not unique to human-machine interaction; similar dynamics have been observed in studies of animal cognition. When an animal displays complex problem-solving skills, researchers often initially attribute it to higher intelligence. However, as the mechanisms behind the behavior are studied and understood, explanations often shift toward simpler, more mechanical processes like conditioning or instinct. The AI effect is simply a more acute version of this phenomenon applied to our own creations.
Herbert A. Simon, another titan in the field, noted that artificial intelligence has always provoked strong emotional reactions. This is because it strikes at the core of our self-definition. To admit that a machine can think is to admit that thinking might be something less than magical, something that can be replicated and engineered. The resistance to this idea manifests as the AI effect—a constant redefinition of terms to keep human cognition in a separate, elevated category.
The Commercial Reality vs. Public Perception
The 21st century has brought a curious contradiction to the AI effect. On one hand, the term "AI" has become ubiquitous in public discourse and marketing. Companies are eager to attach the label to their products, from washing machines to stock trading algorithms, hoping to convey innovation and sophistication. This widespread commercial use has given the field greater visibility than ever before.
On the other hand, as these technologies become embedded in everyday applications, they reinforce the very trend of normalization that McCorduck described. A 2016 survey of artificial intelligence noted that AI systems are increasingly invisible because they are everywhere. They work so well and are so integrated into our lives that we stop seeing them as "AI" and start seeing them as just another part of the digital landscape.
This duality creates a confusing narrative for the public. We hear about "Artificial General Intelligence" (AGI) and existential risks in news headlines, while simultaneously using tools powered by state-of-the-art deep learning to navigate traffic or filter spam without a second thought. The gap between the hype of unsolved problems and the reality of solved ones widens with every passing year.
In recent years, the rise of large language models and generative AI has reignited the debate. These systems can write convincing essays, generate photorealistic images, and hold fluid conversations. Yet, even as they demonstrate capabilities that were once thought to be the exclusive domain of human creativity, a chorus of critics immediately steps in to describe them as "statistical parrots" or "mechanical predictors." The argument follows the same script: if it works, it's just pattern matching; it doesn't understand.
This interpretation emphasizes how technologies are reclassified over time as they become widespread and commercially successful. It is a cycle of discovery, implementation, and demystification. But unlike other scientific fields where success builds on previous knowledge without changing the definition of the field itself, AI seems to define itself by its failures. The "AI effect" ensures that the field is always chasing the next horizon, never quite arriving at the destination it seeks to describe.
Philosophical and Historical Context
The phenomenon has deep roots in the philosophical history of mind and machine. Early AI systems were often described using models involving human computation, such as human-assisted Turing machines, when problems could not be fully formalized. The assumption was that if a computer could mimic a human process step-by-step, it was performing intelligence. As computers became more powerful, they began to solve problems in ways that bore no resemblance to human thought processes—such as Deep Blue's brute-force calculation or modern neural networks' opaque decision-making.
When the method diverges from the human experience, the label of "intelligence" is often revoked. This leads to questions about functionalism (the philosophy of mind) and whether intelligence is defined by the internal process or the external output. If a machine produces an intelligent result, does it matter how it got there? The AI effect suggests that for many, the answer is yes; if the mechanism isn't human-like, it's not intelligence.
Philosophers argue that this reclassification might reflect genuine conceptual distinctions rather than mere bias. Perhaps there is a fundamental difference between simulating intelligence and possessing it. However, the timing of these shifts—the immediate demotion of AI to "computation" upon success—strongly suggests that social and psychological factors play a significant role.
During periods known as AI winters, when progress stalled and funding dried up, researchers sometimes avoided the term "artificial intelligence" entirely due to its negative connotations. They might rebrand their work as "cognitive science," "machine learning," or "expert systems." This was a pragmatic response to the AI effect: if you can't solve the problem, don't call it AI, because calling it AI invites the definition of intelligence that you haven't met yet.
In the modern era, while the term is back in vogue, the underlying dynamic remains. The "God of the gaps" fallacy applies here as well: we attribute "intelligence" only to what we cannot explain or replicate. As our knowledge expands and fills those gaps, the concept of machine intelligence shrinks to fit the remaining unknowns.
The Future of the Definition
As we move deeper into an era where AI systems are capable of complex reasoning, creative generation, and autonomous decision-making, the AI effect will likely continue to shape the discourse. We may see new categories emerge to distinguish between "narrow" intelligence, which is just computation, and some elusive form of "general" or "conscious" intelligence that remains out of reach.
The challenge for society is to recognize this pattern for what it is: a defensive reaction rather than an objective assessment of capability. If we accept that intelligence is a spectrum and that machines can exhibit intelligent behavior without necessarily possessing human consciousness, we might be able to engage with these technologies more honestly. We can appreciate the sophistication of a system that recognizes your face or translates a language in real-time as a form of intelligence, even if it differs from our own.
The history of AI is not just a history of algorithms and hardware; it is a history of human self-perception. Every time a machine masters a task, we are forced to ask what that means for us. The AI effect is the shield we raise in response, insisting that the victory belongs to the engineer, not the machine. But as the capabilities grow more profound, the shield becomes harder to hold up.
Ultimately, the "AI effect" serves as a reminder that technology does not exist in a vacuum. It interacts with our deepest beliefs about who we are and what makes us special. Whether we view these advancements as mere computation or genuine intelligence may depend less on the silicon chips themselves and more on our willingness to expand our definition of thought beyond human limits. The paradox remains: the more AI succeeds, the less it seems like AI. And perhaps that is because true success means blending so seamlessly into our world that we no longer recognize it as something separate from ourselves.
"AI is whatever hasn't been done yet." — Tesler's Theorem
This simple maxim captures the eternal chase of the field. It is a promise of the future and an explanation for the present. As long as we define intelligence by what machines cannot do, we will never have to admit that they can think. But in doing so, we risk missing the profound reality that thinking itself may be more mechanical, and less magical, than we ever dared to imagine.