Defining Artificial Intelligence in Modern Technology

What exactly is artificial intelligence? It sounds like a simple question, but the more you dig into it, the more you realize how philosophically loaded it actually is. Scholars, engineers, and philosophers have been wrestling with this question since at least the 1950s – and no single definition has won the debate. The reason is straightforward: AI is not just a technical concept. It sits at the crossroads of computer science, cognitive psychology, logic, and philosophy, and each of these disciplines defines “intelligence” differently. Understanding AI properly means understanding the full range of perspectives on what it means for a machine to think, reason, or act.

The foundational question: can machines think? 🔗

Before defining AI, we need to ask what AI is even trying to achieve. According to the Internet Encyclopedia of Philosophy, the central philosophical question at the heart of AI is not just a technical one – it is the question Alan Turing famously posed: “Can a machine think?” What makes this profound is that “intelligence” and “thought” are not scientifically settled concepts. In European intellectual and legal traditions, moral standing has always depended not just on outward behavior, but on inner states of mind. So when a computer appears to perform intellectual tasks, the immediate question is: is it actually thinking, or just simulating thought?

This tension between appearance and genuine intelligence runs through every serious attempt to define AI. The discipline has been described, at its broadest, as the attempt to build machines that – in appropriate contexts – appear to think or act like persons. As the Stanford Encyclopedia of Philosophy notes, this goal alone ensures that AI remains of enormous interest to philosophers, who continue to debate whether such goals are even attainable.

Early scholarly definitions of AI 🔗

The founding figures of AI each offered their own definition, and these early formulations still shape how the field is understood today. The IEP records that John McCarthy characterized AI as the effort to discover and implement computational means to make machines behave in ways that would be called intelligent if a human were doing the same. Marvin Minsky defined it similarly – getting machines to do things that would require intelligence if done by humans. Both definitions are deliberately behavior-focused, but they sidestep the deeper philosophical question of whether a machine performing intelligent acts is genuinely intelligent.

This sidestepping is intentional. It reflects a pragmatic strand in AI research that says: define AI by what it produces, not by what it “feels” internally. But critics have long argued this leaves the most important question unanswered. As philosopher John Searle argued in 1980 with his famous Chinese Room thought experiment, even a machine that passes all behavioral tests for intelligence may not actually understand anything – it may simply be manipulating symbols according to rules, with no real comprehension behind it.

Strong AI vs. weak AI 🔗

One of the most important conceptual distinctions in any definition of AI is the line between strong AI and weak AI, a distinction introduced by Searle. Weak AI accepts the prospect of machines that act intelligently, without claiming that they genuinely think. Strong AI goes further – it claims that an appropriately programmed computer truly has a mind, in the full and literal sense of the word. As summarized in the Cambridge Handbook on AI, strong AI holds that the computer really is a mind, not merely a simulation of one.

This distinction matters enormously for how we approach AI ethics, law, and development. If AI is only ever weak – a sophisticated tool – then questions of machine rights, consciousness, and moral responsibility look very different than if strong AI is possible. Current philosophical research frames this as the core question: is AI merely a complex mechanism imitating human thinking, or is it capable of genuine consciousness and self-awareness?

The four approaches to defining AI 🔗

The most systematic and widely used framework for categorizing AI definitions comes from Stuart Russell and Peter Norvig in their landmark textbook Artificial Intelligence: A Modern Approach. Now in its fourth edition, this book is considered the standard text in the field, used at over 1,500 universities worldwide. Russell and Norvig observed that definitions of AI cluster around two key distinctions: thinking vs. acting, and human-like behavior vs. rational behavior. Crossing these two distinctions produces four distinct approaches to what AI is and what it should do.

1. Thinking humanly: the cognitive modeling approach 🔗

This approach defines AI as the attempt to replicate human cognitive processes in a machine. According to OpenLearn, a very strong thread throughout AI history has been the idea that modeling human thought processes could enable us to reproduce such processes computationally. This approach is closely tied to cognitive science – an interdisciplinary field spanning psychology, computer science, philosophy, linguistics, and anthropology – all trying to understand how human minds actually work.

The challenge is obvious: to program a machine to think like a human, you first need a precise, scientifically verified account of how humans think. That account does not fully exist. Cognitive science has made progress, but the human mind is extraordinarily complex. As noted in Springer’s AI overview, research in this school seeks to reproduce the processes, representations, and results of human thinking – not just the outputs, but the internal mechanics of thought itself.

2. Acting humanly: the Turing test approach 🔗

This approach shifts the focus from internal cognition to external behavior. Rather than asking how a machine thinks, it asks whether a machine behaves indistinguishably from a human. This is essentially the approach behind Alan Turing’s famous test. As OpenLearn explains, Turing proposed that if a human observer communicating via text with both a machine and a human cannot tell which is which, the machine has passed the threshold for intelligence. Passing this test would require, at minimum, natural language processing, knowledge representation, automated reasoning, and machine learning – effectively covering the core capabilities AI has pursued since its inception.

The “acting humanly” standard has been influential, but also criticized. Harvard Law’s Petrie-Flom Center points out that if we lower the bar to acting like a human in just one respect, almost any software qualifies – including a calculator that performs arithmetic better than most people. The test is simultaneously too broad and too difficult to pass in its full, demanding form.

3. Thinking rationally: the laws of thought approach 🔗

Rather than modeling human cognition – which is often biased, inconsistent, and emotional – this approach aims for something more ideal: logical, correct reasoning. The goal here is to formulate laws of thought using systems of symbols derived from mathematical logic, and to build machines capable of reasoning from premises to correct conclusions. This tradition is deeply connected to classical logic, dating back to Aristotle’s syllogistic reasoning, and formalized in modern symbolic logic.

The appeal is clear: logic is precise and checkable. If a machine reasons according to valid logical rules, we can verify that its conclusions follow from its inputs. However, a key limitation is that not all intelligence is reducible to formal logical rules. A great deal of human knowledge is informal, context-dependent, or expressed in probabilistic rather than absolute terms. Translating the messy complexity of the real world into clean logical propositions is itself a massive unsolved problem in AI.

4. Acting rationally: the rational agent approach 🔗

This is the approach that Russell and Norvig themselves favor, and it is the dominant paradigm in modern AI research. A rational agent is one that perceives its environment and takes actions to achieve the best possible outcome – or, in conditions of uncertainty, the best expected outcome. As defined by Russell and Norvig, a rational agent is one that acts so as to achieve the best outcome or the best expected outcome when there is uncertainty.

This approach is more general and more tractable than the others. It does not require replicating human cognition exactly, nor does it insist on passing behavioral tests designed around human norms. As OpenLearn explains, acting rationally can include thinking rationally as one component, but it covers much more – everything an agent might need to do to successfully navigate its environment and achieve its goals. Many of the capabilities involved in the Turing test, such as language understanding and learning, are also subsumed within this broader framework.

Why no single definition wins 🔗

Each of the four approaches captures something real about what we mean by intelligence, and each has blind spots. Human-centered approaches (thinking and acting humanly) are intuitive and grounded in concrete performance, but human intelligence is itself imperfect, biased, and not fully understood. Rationalist approaches (thinking and acting rationally) are more mathematically rigorous, but rationality alone does not capture everything we associate with intelligence – creativity, empathy, common sense, and adaptability all go beyond optimal logical inference.

As philosopher Tobias Rees has argued, what makes AI a genuinely philosophical event is that AI systems challenge the formerly clear distinction between humans and machines, and between living and non-living things. MIT Sloan Management Review puts it bluntly: philosophy increasingly determines how AI systems reason, create, and innovate – shaping everything from what counts as knowledge, to how AI represents reality. The choice of which definition of AI you adopt is not just academic; it determines what you build, how you evaluate it, and what ethical responsibilities you accept.

From definition to domain: what AI actually does 🔗

Regardless of which definitional framework one uses, there is broad agreement on what AI actually involves as a field of study and practice. Across approaches, AI fundamentally concerns the study, design, and building of intelligent agents that can perceive their environment, store and use knowledge, reason from that knowledge, and act to achieve goals. Whether the benchmark is human performance, logical validity, or rational optimality, these core capabilities – perception, representation, reasoning, and action – remain central.

The debate over definitions also reflects a deeper question about the relationship between human intelligence and machine intelligence. The philosophy of artificial intelligence – a branch of both the philosophy of mind and the philosophy of computer science – asks whether human and machine intelligence are the same kind of thing, whether machines can truly have consciousness, and whether intelligence is fundamentally computational. These questions are not resolved, and they continue to shape every new wave of AI development, from expert systems to deep learning to large language models.

What do you think? If a machine can consistently outperform humans on reasoning and problem-solving tasks, does the question of whether it is truly thinking still matter – philosophically or practically? And does the approach we use to define AI (human-centered vs. rationalist) ultimately determine the kind of future we are building with these technologies?

References

1. https://iep.utm.edu/artificial-intelligence/
2. https://plato.stanford.edu/entries/artificial-intelligence/
3. https://www.apu.apus.edu/area-of-study/arts-and-humanities/resources/exploring-the-connection-of-philosophy-and-artificial-intelligence/
4. https://philarchive.org/archive/MLLPOA
5. https://www.eu-scientists.com/index.php/fag/article/view/117
6. https://en.wikipedia.org/wiki/Artificial_Intelligence:_A_Modern_Approach
7. https://www.open.edu/openlearn/mod/oucontent/view.php?id=116249&section=2.4
8. https://link.springer.com/chapter/10.1007/978-3-030-51110-4_2
9. https://petrieflom.law.harvard.edu/2018/11/30/the-tricky-task-of-defining-ai-in-the-law/
10. https://www.noemamag.com/why-ai-is-a-philosophical-rupture/
11. https://sloanreview.mit.edu/article/philosophy-eats-ai/
12. https://en.wikipedia.org/wiki/Philosophy_of_artificial_intelligence