In my January blog I claimed that “defining exactly what AI is and how it relates to “machine learning”, “big data” and “data science” is trickier than people think”. I promised a blog on this topic, and so here we are. Brace yourself for this one!
In my original blog I defined AI rather succulently as:
“AI is the development of intelligent machines or systems capable of performing some sort of intelligent human-like task or behaviour.”
If it left you confused, thinking it’s a very broad and rather vague definition, then don’t worry, you’re not alone in this.
So many definitions
If you type “what is artificial intelligence?” into Google (using a Google Chrome browser) you should receive around 4,200,000,000 results – a truly mind-boggling number of results. If you follow a few of the top links from your results list, you’ll discover bewildering definitions for artificial intelligence:
“At its simplest form, artificial intelligence is a field, which combines computer science and robust datasets, to enable problem-solving. It also encompasses sub-fields of machine learning and deep learning, which are frequently mentioned in conjunction with artificial intelligence. These disciplines are comprised of AI algorithms which seek to create expert systems which make predictions or classifications based on input data.” – IBM
“Artificial intelligence (AI) is intelligence demonstrated by machines, as opposed to the natural intelligence displayed by animals including humans. Leading AI textbooks define the field as the study of “intelligent agents”: any system that perceives its environment and takes actions that maximize its chance of achieving its goals.” – Wikipedia
“Artificial intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The term may also be applied to any machine that exhibits traits associated with a human mind such as learning and problem-solving.” – Investopedia
It is clear the definitions are very different to one another, with some making a direct link between machine (or artificial) intelligence and the ability of machines to replicate or mimic some human behaviour or intelligence, whilst others are vague and confusing, such as Wikipedia’s definition which refers to “intelligent agents”, whatever those really are.
Lack of an agreed definition
Rather than having a zoo of different definitions for AI, it would be nice if we had a single standard definition for AI that was accepted by academics and practitioners alike. Unfortunately, and at the risk of frustrating you all, I can reveal there actually is no agreed definition of artificial intelligence! In fact, the Council of Europe’s ad-hoc committee on Artificial Intelligence (“CAHAI” – to pronounce it, you have to imagine having a sore throat and you’re loosening the catarrh) reported in December 2020 that:
“To date, there is no single definition of AI accepted by the scientific community. The term, which has become part of everyday language, covers a wide variety of sciences, theories and techniques of which the aim is to have a machine reproduce the cognitive capacities of a human being.” 
In fact, academics, researchers, expert practitioners, as well as several international (and very important sounding) organisations have tried to work on a global and common definition of AI, but so far with fairly dismal results. For instance:
The European Commission’s “High-Level Expert Group on Artificial Intelligence” published in 2018 a 9 page report with the sole purpose of defining artificial intelligence . After 9 pages they define AI to be:“Artificial intelligence (AI) refers to systems designed by humans that, given a complex goal, act in the physical or digital world by perceiving their environment, interpreting the collected structured or unstructured data, reasoning on the knowledge derived from this data and deciding the best action(s) to take (according to pre-defined parameters) to achieve the given goal. AI systems can also be designed to learn to adapt their behaviour by analysing how the environment is affected by their previous actions. As a scientific discipline, AI includes several approaches and techniques, such as machine learning (of which deep learning and reinforcement learning are specific examples), machine reasoning (which includes planning, scheduling, knowledge representation and reasoning, search, and optimization), and robotics (which includes control, perception, sensors and actuators, as well as the integration of all other techniques into cyber-physical systems).”I’m sure you’ll agree with me that this is a pretty catchy definition!
If you think 9 pages is way too long to arrive at a definition for AI, just wait until you try to read the 118-page report (yes, 118 pages!) by the European Commission’s AI Watch Observatory, entitled “Defining Artificial Intelligence 2.0, Towards an operational definition and taxonomy for the AI landscape” which was published in 2021 (It is the follow-up report to their equally riveting 2018 report). Although the report is quite the page turner, I’d literally have to be insane to even attempt summarising such a report here. Suffice to say, despite them reviewing 38 AI policy and institutional reports (including standardisation efforts, national strategies, and international organisations reports), 23 research publications and 3 market reports, all published between 1955 and 2021, the conclusion on page 45 is the dreadful:“The proposed operational definition [of AI] is composed by a concise taxonomy characterising the core domains of the AI research field and transversal topics; and a list of keywords representative of such taxonomy. As AI is a dynamic field, we propose an iterative method that can be updated over time to capture the rapid AI evolution.”If anyone finds this helpful, please reach out to me as we need to talk!
The Organisation for Economic Co-operation and Development (the “OECD”) established a working group which produced the “Recommendation of the Council on Artificial Intelligence” . This is actually an OECD legal instrument, and was the first inter-governmental standard on AI, which was “adopted by the OECD Council at Ministerial level on 22 May 2019”. Despite several well-intentioned aims, including to “…foster innovation and trust in AI…”, the standard doesn’t actually provide a definition of AI. Instead, the framework defines “AI systems”, the “AI system lifecycle”, “AI knowledge”, and “AI actors”. In other words, no help whatsoever.
The challenges of defining intelligence
There are many, many reasons why it is difficult to develop a single succinct definition of AI, and I cannot explain them all in detail here. Instead, I summarise just three reasons:
The field of computer science (including data science and artificial intelligence) is relatively new, at least when compared to traditional academic fields. As such, our field is still going through its adolescent growing pains, with new discoveries and use-cases appearing almost daily. It’s difficult to make concrete definitions of ideas when our own understanding of what can (and cannot) be achieved with AI keeps evolving. Indeed, researchers have recently realised that rather than there being a single type of AI, there may actually be several different types of AI, including: – Artificial narrow intelligence: AI which specialises in only one area; and – Artificial general intelligence: AI which can perform any intellectual task as well as a human can; and – Artificial super intelligence: AI which is much smarter than humans across several fields. I am sure many other variants and definitions of AI will materialise over the coming months and years.
There are deep philosophical debates and uncertainties over what is even meant by “intelligence”. Even “human intelligence” is not very well understood or defined, with there being several different measures to quantify objectively “intelligence”. As a consequence, most definitions of human intelligence in the academic literature are rather vague. If it is difficult to define human intelligence, is it really that surprising we don’t have a single definition for AI?
Even if human intelligence could be properly and consistently defined, unfortunately we humans don’t behave intelligently consistently. We are capable of remarkable intelligence, from every day intelligent activities such as listening, walking, talking, reading, driving vehicles, recognising faces, places, names etc, to academic and artistic intelligence such as unravelling the mysteries of the cosmos, or painting a picture. However, we also act irrationally and irresponsibly, and at times can be phenomenally stupid. Indeed, we often derive some sort of pleasure seeing and hearing about somebody else’s stupidity, even creating entire TV shows on this for our enjoyment.“Two things are infinite, the universe and human stupidity, and I am not yet completely sure about the universe.”– Albert EinsteinOur ability to act sometimes with crass stupidity, whilst a mystery, is perhaps just part of what it means to be human. It raises many philosophical questions when designing AI machines. Do we want to create machines in our true image (i.e. we create them so that on occasion they do the wrong thing) or do we create them to be perfect?
What does AI mean for you?
It seems for now the best we can do is to use “AI” as a blanket or umbrella term to refer to various data science techniques which attempt to emulate or mimic some form of “human intelligence”. When defining AI in this way, according to the CAHAI, an “umbrella term” for AI should be able to achieve a balance between:
“a definition that may be too precise from a technical point of view and might thus be obsolete in the short term, and a definition that is too vague and thus leaves a wide margin of interpretation, potentially resulting in a non-uniform application of the legal framework”.
Overall, not entirely satisfactory, but that is where we are for now. In retrospect, perhaps the definition I used for AI is not so bad after all.
Given the absence of a standardised definition of AI, I feel tempted to run one of those Blue Peter competitions where the public can send in an answer to a question on a postcard and maybe win a prize if their postcard is drawn first out the postbag. In this case, the question is “What is your definition of AI?” Sadly, it’s too impractical for me to organise and run such a competition, and so instead I might put a survey on my LinkedIn account and see what happens.