Despite the growth and development within the Artificial Intelligence (AI) field in recent years, the term AI itself has been one of the hardest to define. This has led to a situation where any reference to the term AI raises its own concerns with reference to the scope and understanding within that context. Often the term AI can take a fluid and open ended form, while others forming a far more rigid and fleshed out understanding of the term in that specific situation. Additionally, there are also situations where in a single sentence the term and its meaning may change. Interestingly, this does not automatically mean that the term has been incorrectly used or applied. However, in the context of policy development and legal frameworks it creates the additional difficulty of identifying rights and liabilities. Therefore, to understand AI it needs to be interpreted beyond just an academic understanding of the term. Broadly speaking the term can be understood in three specific contexts – theoretical, subject specific, and applicational. By understanding AI within each of these scopes it would give a more holistic and broad-based understanding of the technology. By no means can the term be considered exhaustive for all purposes.
First, having a theoretical discussion of AI is arguably critical in any interpretation of the term. The academic and theoretical discussions surrounding the development of AI have largely been the driving forces for its interpretation for the majority of its discussion and debate. One of the most famous references to AI and its development is through the Turing Test, formulated by the mathematician Alan Turing. During his lifetime he neither referred to the technology as AI, nor his conception as a test. Known as the imitation game, it was a method through which a computing system could imitate a human to such an extent that the response of the machine would be nearly indistinguishable from that of a human. This was penned in his famous 1950 article “Computing Machinery and Intelligence”. It focused on the ability of computing systems to “think”. According to this test if a human player who was communicating with a computer and a human behind a terminal could not distinguish between the two, based on his questioning then it could be said that the computing system has the ability to think. Later referred to as the Standard Turing Test, it received multiple revisions and criticisms through the years; including but not limited to setting quantitative standards to measure at what point a computer can be considered to think, setting fixed responses to the computing system that can be reproduced under given situations, and submitting questions in languages that the human or the computer are incapable of understanding. Interestingly, it represents a significant landmark in the development of an understanding of AI, because it lays down what is known as the human standard of intelligence. The ability to think and replicate human interaction is in itself considered intelligent behaviour and to this date is accepted as one standard of intelligence.
This is followed closely by the work of John McCarthy. He is often credited with framing the term “Artificial Intelligence”. During the 1956 Dartmouth AI Summer Research Project, McCarthy laid down his conception of “intelligent functions” with reference to computing. Instead of outright providing a definition for intelligence, McCarthy analysed intelligence in light of intelligent functioning. He gave the example of a computer algorithm that could offer intelligence suggestions on driving routes based on information and calculations. The ability to intelligently process this information and provide outputs was according to him and intelligent operation of computing.
With time this conception has developed, leading to the development of two concepts – Artificial General Intelligence (AGI) and Artificial Specific Intelligence (ASI). Both of them fall within the ambit of AI. However, AGI can be understood as the ability for a computing system to be able to subsume all facets of intelligence and not only mimic human intelligence but replicate and advance beyond it in totality. On the other hand, ASI refers to the ability of computing systems to be able to perform specific tasks intelligently. Deep blue is a system that many consider to be one of the first iterations of ASI, as it had the ability to play chess and beat some of the best players in the world. This was roughly based off of the works of John von Neumann.
David Poole looks at computing intelligence in terms of the changing environment and the ability to adapt to and environments based on learnt experiences and physical limitations. The ability to act intelligent is predicated on the ability to exercise autonomy, whether it is in the physical or digital environment. To be able to quickly learn and apply oneself, in this case a computing system, could be considered as intelligence. Many scholars have gone on to attempt to discuss and interpret the scope of intelligence. Nils J Nilsson takes a similar approach, but simplifies the end to a far greater extent. According to him any activity that involves the application of intelligence can be referred to as AI. However, what varies is the degree of intelligence expressed.
Although the extent and breadth of the discussion relating to the development and understanding of AI and intelligence can be extended, this broadly match the trajectory of discussions relating to AI at an academic and theoretical level. Independent of this the development of technology that has come to be referred to as AI and its usage have led to a different line of questioning of how to address, define, and regulate technology.
Secondly, the understanding of AI in the context of a subject domain has increasingly gained traction in the world. As opposed to the understanding of AI as a theoretical construct, by looking at it from a particular domain we can understand whether a particular action is intelligent or not based on its application within that context. For example, the ability of an autonomous vehicle to be able to calculate its position to reach a particular destination, while avoiding obstacles, following traffic rules, and ensuring safe transport, can together be considered intelligence in the context of the ability to drive. Such a conception can be extended to any domain, with each activity being understood with reference to this understanding of AI.
Thirdly, an applicational understanding of AI takes into account the application of computing technology as a means to an end, as opposed to an end in itself. Using this approach if a system has the ability to perform an intelligent task that involves the application of intelligence as a means to achieve an end, can be considered an applicational understanding of AI. For example, the ability of a computing system to recognise images based on previously learnt data is an example of applicational understanding of AI. In itself it may or may not have use, but can be applied in a manner that would be useful.