Artificial intelligence and machine learning landscape [chapter]

2020 Artificial Intelligence in Management  
Artificial intelligence (AI) is the central focus of this book -it is therefore only fitting that we begin our deliberations by reflecting on the term itself. The first word denotes something fairly easy to define: artificial simply means man-made. Naturally, any attempt to delve on specifics such as e.g. form, physical location or architecture of such a creation would significantly complicate things. As will be shown in the subsequent chapters of this book, at present, intelligent systems are
more » ... asically computer programmes running on devices utilising silicon chips. They can operate 'autonomously', take advantage of remote intelligent services (e.g. in the Cloud Computing model) or engage with other objects to establish networks. These seemingly trivial observations are in fact anything but. These days, highly advanced operations can already be performed by quantum computers (inherently different from contemporary integrated circuitry), data can be stored in DNA codes (rather than on digital memory chips) and calculations can be performed with the use of protein structures (and one does not mean animal brains). As we can see, there is currently more than one way to potentially approach the issue of artificiality -anyone interested in the topic could do far worse than to read through Bostrom's compelling analysis of the same (2014). A concept far more difficult to put a finger on is intelligence. There have been many definitions over the years, mostly in the field of psychology. It has been a long-standing dream to create structures capable of assisting or replacing people in solving various day-to-day problems. One of humans' 'competitive advantages' over animals is the ability to use tools -objects enhancing people's innate potential, e.g. strength or speed. In time, such tools evolved into machines -still, their contributions continued to be limited to physical activities. The development of higher cognitive functions, in particular the ability to formulate complex goals, adopt and implement strategies and plan actions necessary to that end, inspired people to search for methods of enhanc-Andrzej Wodecki -9781839104954 Downloaded from Elgar Online at 11/29/2020 08:28:10PM via free access Artificial intelligence in management 2 ing their abilities also in those specific areas (i.e. no longer exclusively physical but also mental). A sort of a 'prototype' in this context came in the form of mathematical theorems and algorithms which (due to their abstract nature) helped in categorising and correlating certain seemingly diverse problems and subsequently fairly easily solving them (e.g. the laws of geometry applied in the construction of the Tunnel of Eupalinos in ancient Greece). The next natural step was to 'implement' those originally abstract algorithms in physical devices: initially simple (e.g. the abacus) and then increasingly complex (e.g. Babbage's difference engine) (cf. Davis 1949). As algorithms continued to be developed, so did the range of problems to which they could be applied, which on the one hand created a demand for increasingly advanced computing machines, and on the other generated entirely new problems. There were certain rather narrow areas (e.g. arithmetic) where machines fairly quickly proved themselves to be more capable than humans. However, as computers began to become more popular, in particular more available to a wider group of programmers, the spectrum of problems that could possibly be solved by man-made machines grew exponentiallywhich soon inspired people to reflect on the machines' intelligence. Before committing to a definition of intelligence, let us first take a somewhat closer look at the problems one may reasonably expect intelligent machines to solve. To that end, we will use the slightly adapted classification first proposed by Russell et al. in their fundamental textbook on AI (2010). Let us begin by characterising the environment wherein our machine might operate. As the given task is being performed, the same can remain unchanged (static as e.g. in the game of chess) or evolve -i.e. be dynamic. 1 Moreover, the next consecutive state of the environment may depend solely on the actions of the machine (in which case the environment is deterministic) or be influenced by factors independent of the agent (a stochastic environment). The environment can also be discrete (described by variables with a finite set of possible values) or continuous. The latter distinction is particularly important from the computing perspective: in a world of infinite choices, the system must have the capacity for generalisation. Finally, our machine (often described as the agent) may act within the given environment either independently or be accompanied by other 'beings' (be it machines, animals or humans) -in which case we can talk of a multi-agent environment. The next distinction is related to knowledge about the environment. The agent may know the rules and laws applicable in the given context (e.g. in board games) or not know them (as is often the case in real life). Another aspect of knowledge about the environment is access to information about its state: the agent may either have full knowledge of the environment's state, only partial knowledge thereof or no knowledge whatsoever (be blind).
doi:10.4337/9781839104954.00005 fatcat:5j25xnrqizetxkeyekaboa6pki