Mental Models of Artificial Intelligence

As the popular saying goes “There is nothing new under the sun” and Artificial Intelligence is no exception to that. The term artificial intelligence is so old that if you ask your grandparents about artificial intelligence (assuming they were one of the few predominantly white males studying mathematics and statistics at the time) they would probably tell you how they had a class project in artificial intelligence and how it is amazing it is being utilized today. This is even more so if you ask your parents as well (also presuming they were in the predominantly white male field of computer science and engineering).

One of the major reasons Artificial Intelligence is widely being discussed today is because the computers that are available now are more advanced than what was available in previous generations. Back then there were about 4 kilobytes worth of storage in a computer which is not even up to the size of a photo downloaded online (that is on average 200 kilobytes!). We can now do more on a phone than was physically possible when artificial intelligence was first introduced. With people having more access to better computers, artificial intelligence which was theorized years ago is now being developed more widely.

This article is not a history of artificial intelligence though but addresses why even with the prevalence of this technology in our everyday life, more people have less of an understanding of what, where, and how the underlying technology is used.

AI use increases but users’ understanding decreases

Understanding Artificial Intelligence is not limited to those who have no background in computer science but ranges from people who have little to no knowledge of computation to experts who develop the technology. This is because most common machine learning algorithms that are producing amazing results such as Deep Learning are difficult to understand because these algorithms are hidden algorithms. This is what is popularly referred to as a black box.

Understanding Artificial Intelligence is not limited to those who have no background in computer science but ranges from people who have little to no knowledge of computation to experts who develop the technology. This is because most common machine learning algorithms that are producing amazing results such as Deep Learning are difficult to understand because these algorithms are hidden algorithms. This is what is popularly referred to as a black box.

A black box is something where the underlying processes that produce an output are not known to users. A black box algorithm produces outputs without any explanation. A good example of a black box algorithm is when your parents/teachers tell you to do a task and when you ask why you are told “because I told you so”. Or if you were to get a grade back from a test and ask why, you are told “that is the level of work you produced”. The similarities between these two conversations are that at the end of the day you are told something but you do not understand why the parent/teacher told you that thing.

Research shows that many of the users and product owners of artificial intelligence algorithms do not have an understanding of how they work [1]. There is a lot of work in Explainable AI, an approach where AI’s recommendations are accompanied by explanations or rationales, to fix this issue by creating better frameworks and heuristics for how these algorithms work [2] but it is often focused on academia which can be difficult to decipher because the language used to describe how these algorithms work are not commonly used in everyday life and accessibility for these research articles can have limited access [3].

Why we need to know about AI

Artificial intelligence today is important to understand because it is being used widely in our society for most industries and technologies we interact with. Artificial intelligence can be found in healthcare as diagnostics tools, social media for recommendation systems, facial recognition for security systems, stock market in predicting the outcome of a companies stock, autonomous vehicles in driving, and even less known examples such as agriculture for crop yield detection, mortgage underwriting for interest rate determination, the judicial system in determining sentencing duration, vaccine development for faster testing, and the list goes on.

The technology has the potential to revolutionize how we interact and use services and products which is why some companies are investing in the space. As there is a large potential for good there is also a large potential for harm. For example, an AI algorithm that determines the interest rate of mortgage loans can either allow someone to get a fair loan interest rate or an unfair interest rate. This can determine how much per month a person pays on their mortgage which also has effects on how much a month a person has for disposable income. As you can imagine, this could cascade to further problems of having access to other resources.

The problem is the social injustices that exist in the world in terms of discrimination or bias are not eliminated when it comes to Artificial intelligence technology. On the contrary, some may even say they are exaggerated when it comes to AI [4].

Because Artificial Intelligence is being used mostly in hidden and often misunderstood ways there exists a need to build better mental models for Artificial Intelligence.

Mental Model for Artificial Intelligence?

What is being studied in the Explainable AI community is how there exist two subsets of polarized views on the use of AI technology. There is a subgroup of people who overtly trust AI and hardly question the results of an AI algorithm and there is another subgroup who actively avoids using AI technology because of its disadvantages [2]. These groups are what I refer to as polarized users. I propose that these users should not exist or be rare, especially in future iterations of AI technology where it will continue to be ingrained in our everyday lives. A good way to combat these two polarized groups is information, which is where mental models can be key.

A mental model can be defined as knowledge representations of technological systems, which are generated through interaction with the respective system [5]. It is how you think about the world around you in an abstract sense. For example, a mental model of how to switch between channels of the TV will be pointing the remote control in a certain way allowing signals from the button you press to communicate between the “mind” of the TV which enables the channels to be flipped. As you can probably tell from this example, I am not an expert at controls signals or electronics of a television but I created this mental model so I know if the channel doesn’t change, there could be something wrong with the remote position, remote power, etc.

Whether or not we know it, most people create mental models for how the world works around them and they need not be accurate but sufficient to recognize a pattern or diagnose an issue. Mental models are not necessarily science but art and vary with each person based on experience, socioeconomic factors, etc. This article, therefore, does not give a rigid definition for what a mental model of artificial intelligence is but allows the reader to form their own personal mental models of AI by providing the basics of how AI works.

Continue Reading the Next Blog to learn more about Artificial Intelligence