One type of topic that is increasingly being set at many tournaments is artificial intelligence (AI). Due to AI being such a complex and novel topic (both in reality and in debating), lots of debaters don’t manage to go much beyond slippery slope arguments implying some type of malevolent impending doom and struggle to make nuanced arguments. Examples where AI is involved, include route-planning, Alexa/Siri, and spam filtering of your emails.
This article is meant to serve as the first of a two-part guide to debating AI, going through the basics of artificial intelligence, starting with machine learning (ML) and ending with deep learning (DL) this week. Next week, I will talk about examples of applications of AI, ML, and DL in the real world (and why they are used). Conclusively, Gigi Gil will put the knowledge to use in the context of debating by discussing a debate motion on artificial intelligence in the system of criminal law. Any additions or questions? Let me know what you think in the comments!
Artificial intelligence refers to the field involved in enabling machines to make intelligent decisions in a similar way to humans and preferably even more intelligently. In short, AI is about modelling human intelligence, often using user-provided data as input. It can be very simple (e.g. a simple check whether the answer that you gave to a quiz is correct) or highly complex (e.g. an algorithm that is able to make art indistinguishable from that made by humans). The Venn diagram to the right explains it well.
For AI systems, one may build specific rules to decide the response, e.g. if the student’s answer to question 5 is A, then it is correct; otherwise, it is wrong. Machine learning (ML) is defined as a subfield of AI in which the system is not given explicit rules on how to complete a certain task. For instance, suppose that we want to predict whether a person has COVID-19. An ML model is given some input data, perhaps the medical history of several patients and whether they have had COVID-19 or not, which is called the training data. The model will then try to find patterns between the appearance of COVID-19 and (some of) these medical variables, but it is not explicitly told by the person creating the algorithm how to find these patterns.
I will not delve deeper into the types of machine learning algorithms that there are except for two examples. Most academics that have done any kind of statistics will have heard of the first example of machine learning: the linear regression (explanation video by Statquest). I was honestly surprised when I first heard that such an ordinary concept is a part of something so “fancy” as machine learning. If you think about it, it makes sense: you feed in the data, define the formula that you want to estimate, and some fancy mathematics figures out what the best parameters (“betas”) are to fit that data.
A more complex ML model is the decision tree (explanation video by Statquest).
This decision tree above was built using ML techniques and tries to predict if someone applied for a loan or not. At each split, it determines which variable it should split on for the best results (with, again, some fancy mathematics). Apparently, the best variable to split on at the start is to ask if someone is over the age of 30. Let’s say they are, so we follow the right-hand side. Then, it is apparently the most optimal to ask if that person has less than 2 children. If this person does not, the decision tree predicts that this person did not get a loan.
Machine learning is also often referred to as a “black box”. That means that we often do not exactly know how the system came to a certain decision. While the two examples mentioned above are straightforward to interpret, it is indeed difficult to interpret the decision process from many other ML models, such as neural networks / deep learning (explanation video by Statquest). The goal of this section is to simply explain the most widely applied black-box models: neural networks and deep learning. In the next article, we will talk about where and why these black-box models are nevertheless applied.
The human brain consists of neurons, which are cells within the nervous system that transmit information to other nerve cells, muscle, or gland cells. These neurons are connected with so-called synapses. The original idea behind neural networks was to try to emulate this biological structure of the human brain with all kinds of complex mathematical calculations.
Basically, deep learning is just a very complex neural network algorithm. Nonetheless, neural networks and deep learning are simply very exciting (both in academics and to businesses) because they can produce models with extremely accurate results, even though they are very difficult, or sometimes nearly impossible to interpret.
In the next article, I will discuss real-world applications of AI, ML, and DL. I hope that this guide was clear to follow and gave you an insight into what AI actually is. If you have any questions, additions, or other remarks, please put them in a comment or drop me a message.