What is Artificial Intelligence?

Why Defining AI Is Harder Than It Sounds
Sheikh, Prins and Schrijvers (2023) are clear: “Defining AI is not easy” and there is “no generally accepted definition”. This is not just an academic problem. It affects real decisions. When people say, “we are using artificial intelligence”, they may be referring to very different things.

Definitions can fail at both extremes.

• Too broad. If Artificial Intelligence is defined simply as “algorithms”, the term becomes almost meaningless. Algorithms existed long before Artificial Intelligence and are used in many everyday tools. Sheikh and colleagues note that such a definition would include a pocket calculator or even a cookbook. That does not help leaders make informed decisions.
• Too strict. At the other end, Artificial Intelligence can be defined as computers fully imitating human intelligence. However, Sheikh et al. explain that this risks “defining the phenomenon out of existence”, because many current systems would not meet that standard.

For professionals, this matters because definitions shape scope. Are you discussing a specific tool designed for a defined task, or a broad claim about human-like intelligence? Clarity begins with choosing and stating your definition.

Two Useful Ways to Define AI And Why Context Matters
Abbass (2021), proposes guiding definitions to help clarify what should and should not be considered Artificial Intelligence. He offers the following definitions:

• Definition 1. “Artificial Intelligence is the automation of cognition.” (Abbass, 2021).
• Definition 2. AI is “social and cognitive phenomena” enabling machines to integrate socially, perform competitive tasks requiring cognitive processes, and communicate by exchanging high-information messages and shorter representations (Abbass, 2021).

Abbass also warns that no definition for AI will be error-free, sufficiently universal, or concisely unambiguous. (Abbass, 2021). For leaders, the lesson is practical: you can adopt a definition that suits your purpose, but you should acknowledge limits and avoid pretending your definition settles everything.

A Practical Definition Used in Policy
After reviewing multiple definitions, (Sheikh, Prins and Schrijvers, 2023) adopt what they describe as an “open definition”, based on guidance from the European Commission’s expert group on Artificial Intelligence: “Systems that display intelligent behaviour by analysing their environment and taking actions, with some degree of autonomy, to achieve specific goals.”

This definition is helpful because it is neither too broad nor too strict. It distinguishes Artificial Intelligence from general digital technology, while remaining flexible enough to include future developments.

However, the authors also explain that even this definition has limits. Phrases such as “some degree of autonomy” can be vague. They also show that some task-based definitions could technically apply to a thermostat, even though most people would not consider a thermostat to be Artificial Intelligence.

The message is not that definitions fail, but that they require careful use.

What People Often Mean When They Say “Artificial Intelligence”
Much of the recent progress in Artificial Intelligence relates to systems that learn patterns from data rather than simply following fixed rules; these systems are often described as “self-learning algorithms” that can recognise patterns in data (Duuren and Pous, 2020).

They also note that many people who talk about Artificial Intelligence today are referring specifically to these pattern-learning systems.

(Jaboob, Durrah and Chakir, 2024) describe Artificial Intelligence as an interdisciplinary field that combines computer science, mathematics, and cognitive psychology. They highlight areas such as systems that learn from data, systems that analyse and work with human language, systems that process images, and robotics. They also emphasise that Artificial Intelligence raises challenges, including bias and ethical concerns.

For professionals, this provides two grounding questions:

• What specific type of system are we using?
• What limitations or risks must we consider?

Narrow Artificial Intelligence and the Idea of General Intelligence
Sheikh, Prins and Schrijvers (2023) make an important distinction. Most current applications fall under what is often called “narrow” or “weak” Artificial Intelligence. These systems focus on specific capabilities, such as recognising images or processing speech.

This is contrasted with “artificial general intelligence”, which would involve understanding and simulating the full range of human intellectual skills. They note that most experts believe this is at least several decades away, if it is achieved at all.

For professionals, this distinction reduces noise. Most decisions today relate to focused systems designed for specific tasks, not machines that replicate human intelligence.