THE REALITY OF ARTIFICIAL INTELLIGENCE: SOCIAL LEARNING THEORY, CONNECTIVISM, AND HUMAN STANDARDS IN LEARNING VERSUS MACHINE LEARNING

T. Spiess, F. Salcher, T. Dilger

MCI Management Center Innsbruck (AUSTRIA)
Learning is integral to human identity. The ability to learn consciously is what differentiates us from animals. Nowadays technological advancements have helped machines achieve feats that were thought impossible just a few years ago. There are many examples of machines creating new strategies for games that humans have played for centuries. But artificial intelligence has also made big strides in video games like Starcraft II and Dota 2 that were assumed to be understood at a very high level and then completely taken apart by these automated systems. All these examples show that a machine learning algorithm that has only been fed the most basic objectives of a game can, in a very short amount of time, understand and master it to a level that no human has been able to, teaching new ways to play that nobody has thought of before. Computers are also able to create paintings, poems and even newspaper articles that are hard to distinguish from something created by humans. We have reached a point where one might question how far away we are from sci-fi scenarios seen in movies.

The goal of the paper is to leave behind the hype, the “what if” scenarios and figure out how close we actually are to these fantasies and to conclude whether machines have the chance of acquiring the same learning abilities as humans. During this quest we will take a closer look at learning in general by studying the most common learning paradigms and examining Social Cognitive Theory as well as Connectivism. Social Cognitve Theory states that most learning occurs through observation of the behavior of other people and seeing the consequences of their actions. While learning through direct experience is also possible, for highly complex or potentially deadly tasks it is neither effective nor feasible. Connectivism focuses on the idea that learning can happen across peer networks that only exist online, and that knowledge is stored in interconnected nodes. What both these theories fail to address is the idea that a computer can by itself generate behavioral patterns that could be mistaken for human actions. A person could then observe the computer perform this behavior and adapt their actions accordingly.

In order to get a grasp of the actual capabilities of artificial intelligence this paper is first going to look at the definitions of artificial intelligence given by literature. In addition to that interviews with various experts from the field of artificial intelligence are conducted. In total, six experts were interviewed. The interviews lasted between 45 minutes and one hour and were analyzed using the Qualitative Content Analysis by Mayring and by applying the principles of interrater reliability.

The results of this paper point to the conclusion that artificial intelligence in its current state is not able to live up to human standards. The main reason being that the human learning process or any creative process for that matter requires a certain adaption to reality. The paper argues that the current approaches to artificial intelligence do not guarantee that an algorithm has an understanding of the environment it is operating in. Despite that verdict, artificial intelligence has managed to generate real value for many of our daily processes. Especially in decision support systems artificial intelligence is used to switch from rule-based recommendations towards data-driven proposals.