What is AI? An explanation by the 4 different schools of thought
Artificial Intelligence (AI) is probably one of the most controversial topics of the 21st century. The last time I searched Google News for the keyword “AI” I came across a lot of scary headlines like “Fake-News-Generating AI Deemed Too Dangerous for Public Release”, or “Creepy AI generates endless fake faces” or “Should we be worried about ‘killer’ AI robots?” and “Should we be worried about killer AI robots?” When I discuss AI with friends or my family, I find that our common idea of AI differs greatly or that there is simply no realistic idea other than science fiction experiences.
That’s why I’ve made it my goal to objectively explain various AI topics in a short series of blog posts, so that you can understand what’s behind the term AI and better understand how it works. I will share with you my findings from the course on Artificial Intelligence at the TU Munich and the probably best known work on AI, “Artificial Intelligence” by Stuart Russel and Peter Norvig, which is over a thousand pages long, and illustrate them with my own infographics. I will also send each blog post to my 94 year old grandpa and discuss with him, whether he understands what I am trying to explain you.
In this first article I will introduce you to the different definitions of AI, the six main areas of AI, and the well-known Turing test. Enjoy!
What is Artificial Intelligence?
In fact, there are many definitions of “Artificial Intelligence” or short AI. If you are looking for artificial intelligence on Wikipedia, you will find (among other) these two definitions:
“Artificial intelligence (AI) […] is intelligence demonstrated by machines as opposed to the natural intelligence of humans and other animals.” [1]
“Die Fähigkeit eines Systems, externe Daten korrekt zu interpretieren, aus diesen Daten zu lernen und diese zu nutzen, um bestimmte Ziele und Aufgaben durch flexible Anpassung zu erreichen.”[2]
Russel and Norvig approach the definition of the term by distinguishing between the following four “schools of thought”:
On the one hand, a distinction is made between human-like behavior and rational behavior, and on the other hand between reasoning and acting. This results in different approaches on how AI systems are build, which can differ fundamentally from each other. The left definitions measure their success to the extent that they can imitate human behavior, while the right ones try to achieve their goal as rationally as possible. According to Russel and Norvig an AI system is rational if it simply does the “right thing”, given what it knows.
1. Systems that think humanly
The basic idea is the following: If we understand how the human brain works, we can simulate or rebuild it. Through psychological experiments, introspection and brain imaging we can try to gain insights about the mechanisms and patterns in the human mind. Mostly cognitive scientists follow this approach, but also psychologists and neuroscientists contribute on that field.
The interdisciplinary field of cognitive science brings together computer models from AI and experimental techniques from psychology to construct precise and testable theories of the human mind.
This led to definitions of AI like:
“[The automation of] activities that we associate with human thinking, activities such as decision-making, problem solv- ing, learning . . .” (Bellman, 1978)
2. Systems that act humanly
One of the most popular approaches to the definition of AI is the Turing Test, proposed by Alan Turing (1950). It explains whether a machine acts humanly. Imagine a large box in which a man or a machine could hide. You can exchange written messages via a letter slot with the “system” behind the box. Now your task is to write questions and messages to the secret system in the box and determine if it is a human or an AI. The AI would pass the test if the human questioner cannot tell whether he is communicating with a machine or a human being. According to the definition of Turing, a computer that is capable of passing this test, possesses an artificial intelligence.
According to Russel and Norvig an AI would need the following capabilities to process the messages and return answers:
- Natural Language Processing: in order to understand and use language to communicate,
- Knowledge Representation: so that the AI can store what it knows and hears,
- Automated Reasoning: which means the AI is able to use the stored information and derive knowledge from it,
- Machine Learning: in order to learn from past input and actions, so the AI can adapt to a changing environment and detect patterns.
Moreover the Total Turing test additionally requires the human interrogator to see the AI system through video and give it mechanical tasks. In order to pass this far more complicated challenge, the AI would need:
- Computer Vision: the ability to perceive objects in the environment,
- Robotics: in order to fulfil mechanical tasks.
This illustrative definition approach is still relevant today and led to the creation of the six main disciplines in the field of AI till today.
One definition of AI that fits into the acting humanly school of thought is the following:
“The study of how to make computers do things at which, at the moment, people are better.” (Rich and Knight, 1991)
3. Systems that think rationally
This approach is mainly based on logic and has its routes in the Greece philosophy. A logic-based AI system uses a set out of rules, so called syllogisms, which it uses to draw conclusions. For example: “Socrates is a man; all men are mortal; therefore, Socrates is mortal.” But this school of thought comes with its limits in practice. First, it is difficult to know all the rules in a complex world from which to draw logical conclusions. Second, the translation of perceived informal knowledge into logical rules is not as easy, as one might think, especially if one cannot be certain if the knowledge is true.
4. Systems that act rationally
The last school of thought is the most modern approach in AI. It tries to define AI, with the concept of so called rational agents. An agent is just something that acts and a rational agent is one that acts so as to achieve the best outcome or, when there is uncertainty, the best expected outcome (Russel and Norvig, 2016). This approach tends to be better than the previous schools of thought, because it is more inclusive and general, since it allows logical inference to be used as a conclusion drawing method but also opens the door for a variety of other mechanisms to achieve rationality. Moreover in the field of Computer Science the approaches based on human acting and thinking tend to be more difficult to implement, because of their complexity and lack of being well defined. On the other hand the standard of rationality can be expressed easier with mathematics and hence be implemented better. Consequently a modern definition of AI can be expressed as:
“Computational Intelligence is the study of the design of intelligent agents.” (Pooleet al., 1998)
In the next blog post I will explain more about the concept of intelligent agents. If you liked this blog post, please leave me a clap, share it and get in touch with me on medium, LinkedIn or tendex.net.
About the author: Otto Lang has studied Information Systems at TUM and is the founder of Tendex, an IT solutions company based in Munich. Tendex focuses on building web based software solutions for clients, as well as stand alone software products like instalics.com.
Sources:
»Grundlagen der Künstlichen Intelligenz« [foundations of the artificial intelligence] (IN2062) at the Technical University of Munich lectured by Professor Dr. Matthias Althoff,
[1] https://en.wikipedia.org/wiki/Artificial_intelligence,
[2] “Siri, Siri in my hand, who is the fairest of them all? On the Interpretations, Illustrations and Implications of Artificial Intelligence” by Andreas Kaplan and Michael Haenlein,
“Artificial Intelligence - A modern approach” by Stuart J. Russell and Peter Norvig
