IP 2 – Artificial Intelligence

Alan Turing was a mathematician and scientist who created the Turing Test to evaluate a machine’s ability to imitate human-like behavior, which influenced how AI systems were designed for the next few decades. He would likely describe intelligence as acquiring new skills to fill the “blank sheet” of the mind (Chollet, 2019, p. 6).

“Alan Turing (1912–1954) is one of the foundational figures in modern computing. He proposed the Turing Test. Turing did not give a simple definition of intelligence. The important insight was we could investigate machine intelligence by looking at observable behavior, rather than agreeing on an abstract definition of ‘thinking’ (OpenAI, 2026a).”

“John McCarthy (1927–2011) helped establish AI as an explicit academic discipline. McCarthy created Lisp, a programming language suited to AI research. He argued that intelligence is not tied to being human or having a brain. What matters is the capacity to accomplish goals in the world using computational processes (OpenAI, 2026d).”

“Herbert A. Simon (1916–2001) was an American scholar. He and Allen Newell developed computer programs intended to simulate human problem solving. He viewed intelligence as the ability of a system to represent information and manipulate those representations in ways that enable it to solve problems and act effectively (OpenAI, 2026c).”

“Marvin Minsky (1927–2016) co-founded the MIT Artificial Intelligence Laboratory with John McCarthy. Minsky built SNARC, an early neural-network machine. His most famous contribution was the “Society of Mind” theory. His approach was: Intelligence emerges when many different processes—reasoning, perception, memory, learning, planning, common sense, language—work together (OpenAI, 2026g).”

“Timnit Gebru has highlighted issues on bias, representation, accountability, and social consequences in AI. An interpretation of her research: Intelligence should not be equated with producing impressive outputs; we should ask what capabilities a system possesses, how those capabilities arise, and what happens when the system is deployed (OpenAI, 2026h).”

Human languages have evolved over time as our needs for expression have expanded, resulting in words constantly being redefined like “viral” or “dab” – or new terms being coined with the prevalence of Internet and meme culture. We are able to “materially change” the contexts of what is said depending on conditions like the content, audience, time, and location (Jones, 2020, p. 26). We also utilize a variety of modes – gestural (ASL), logographic (Chinese), phonetic (English) – while programming languages are “purely performative” in which code literally “does what it says” (Jones, 2020, p. 26).

“The key difference is who—or what—the language is designed for, and how precisely meaning must be specified. In English, “Could you open the window?” is a question about someone’s ability, but in conversation it is a request. A programming language doesn’t allow that kind of contextual interpretation. The computer doesn’t infer that you probably meant something else. Both kinds of language have syntax and semantics. The major difference is that programming-language semantics are designed to be sufficiently formal and unambiguous for machines to execute, whereas natural-language meaning is heavily shaped by human cognition, context, convention, and social interaction (OpenAI, 2026e).”

While intelligence has traditionally been measured in machines through benchmarks on “task-specific skill”, more recent definitions focus on the goal of achieving “generality” by applying existing knowledge to adapt to new, previously unknown tasks (Chollet, 2019, p. 4). Importantly, a facet of human intelligence that cannot be replicated in machines comes in the form of “crystallized abilities” that can only be acquired, developed, and mastered through direct experience (Chollet, 2019, p. 16). As programs are fully hard-coded by engineers, they only represent a “crystallized output” of the “thought process” and do not showcase a machine’s ability to solve problems or generate answers without guidance from its human creators (Chollet, 2019, p. 19).

“The deepest difference is that machine intelligence is engineered, whereas human intelligence is a biological, embodied phenomenon. The comparison is complicated because ‘intelligence’ isn’t a single thing. Consider the word ‘fire.’ An AI can learn relationships involving the word. A human, however, can also have lived through an encounter with fire: seeing flames, feeling heat, smelling smoke, being frightened, remembering a campfire, learning from parents not to touch it, and so forth. Human concepts are embedded in a body, environment, history, and social world. Machine systems primarily acquire their concepts through computational representations derived from data and interaction (OpenAI, 2026b).”

Machines primarily learn by being trained on datasets, which can, in a way, become a limitation for answer generation depending on any biases present – they require human intervention to be taught what acceptable outputs are, while humans have logical reasoning and moral values, to deduce what is ethical. Machines can also “artificially” increase their performance by the incorporation of more priors or training data, which is not proof of generalization (Chollet, 2019, p. 19). While we also possess prior knowledge and experience, it can only be acquired and developed over a natural period of time alongside our ability to generalize and cannot be gamed.

“Machine learning changes a computational model by adjusting parameters from data; human learning changes a living brain through experience, perception, action, memory, and social interaction. Suppose a child learns what a dog is. The child gradually constructs a concept that is considerably richer than a visual category. It becomes connected to sound, movement, emotion, danger, affection, language, social knowledge, and personal memories. A machine-learning system might instead be trained on millions of labeled or otherwise structured examples and adjust its internal parameters so that it becomes increasingly good at recognizing or predicting things associated with ‘dog’ (OpenAI, 2026f).”