Another important distinction is that between ‘strong’ and ‘weak’ AI. While sometimes the term strong AI is used to refer to AGI (Kurzweil 2005a: 260), the term originally derives from the work of the philosopher John Searle (1980), who used it to describe the position of those who believe that an advanced AI would be conscious. Searle critiques this view from the sceptical position of weak AI, which holds that machines can never be conscious. Searle’s famous Chinese Room thought experiment, which hypothesizes a human (or machine) equipped with an encyclopaedic set of rules for translating Chinese into English without being able to speak or understand the language, attempted to prove this position. As Kaplan puts it, ‘strong AI posits that machines do or ultimately will have minds, while weak AI asserts that they merely simulate, rather than duplicate, real intelligence’ (2016: 68). We do not take a definite stance on the question of machine consciousness. The arguments in this book do not depend on machine consciousness being physically or even logically possible, nor on the impossibility of such. On this topic we are functionally agnostic, and, as we argue in Chapter 3, so too is capital.
Actually-existing narrow AI is typically divided into three schools of thought: Good Ol’ Fashioned AI (GOFAI), machine learning (ML), and the situated, embodied and dynamical framework (SED). GOFAI, also known as symbolic AI, was the first approach to AI and remained dominant until the 1980s (Boden 2014: 89). It is an approach that aims to implement high-level cognitive functions, such as logical reasoning, in machines through the manipulation of information encoded in a symbolic language. Such a system creates internal representations of its world or a problem domain in a symbolic language and performs logical manipulations on this representation to think or act. These systems are often constructed out of sets of clearly defined rules. The best examples of GOFAI are so-called ‘expert systems’ or ‘knowledge systems’, which emerged and proliferated in the 1980s. These were intended to capture the knowledge of human experts and make it available to less skilled workers or ignorant managers. Expert systems were used for medical diagnosis, credit scoring and analysis, and business management, but the most famous is IBM’s chess-playing system Deep Blue, which in 1997 defeated the reigning world champion Garry Kasparov. However, GOFAI required vast sets of rules with myriad possible interactions. Solving complex problems in this way necessitates tremendous computational power; for these and other reasons, approaches other than GOFAI were pursued.
introduction: ai-capital
11