By Sophia C. ’23

The field of artificial intelligence, or AI, is incredibly vast, spanning over the fields of computer science, psychology, and statistics, to name a few. In more recent developments, as AI algorithms grow more advanced, issues arise circulating the questions of replacing human workers with computers, information and personal data security, as well as the intentions of AI developers. The rapid growth of AI in past years can be attributed to the complex technological machines and systems being developed, as when people first started studying the field, computing was not only expensive but machines were nowhere near as powerful as a basic laptop is today. 

AI development is by no means straightforward, and there is no one specific algorithm or method to use. Popular AI algorithms include support vector machines (SVM), K nearest neighbors (KNN), and linear regression. These are only a few algorithms out of a vast amount, each tailored to being effective in different situations. The goal of each algorithm is to learn how to solve issues on its own in an accurate and efficient manner. By taking in training data, it is able to learn how to do so, most likely using one of the three most popular methods for teaching an algorithm: unsupervised learning, supervised learning, and reinforcement learning. Of course, there are many other learning methods, but these are the most popular.

There are a variety of methods of learning because different algorithms require certain types of learning. For example, a KNN algorithm operates by undergoing supervised learning. In supervised learning, the algorithm takes in data that has already been “labeled” (i.e. able to be categorized by the algorithm) and uses information on what data has what type of classification to then classify unlabeled data. (By contrast, unsupervised learning deals solely with unlabeled data, and reinforcement learning allows the user to provide feedback to the algorithm.) The KNN algorithm undergoes supervised learning by taking in the labeled data, and classifying the unlabeled data by finding the distance between the two pieces of data on a graph. It repeats this process with multiple pieces of unlabeled data, and the ones with the closest distance to the labeled data are put in the same category as the labeled data. While this is a simplified description of the process, it demonstrates how the AI algorithm is able to take in data and learn from it, and use that information to be able to learn on its own. 

AI systems are present in our daily lives- for example, through algorithms that give recommended ads, digital assistants like Siri and Alexa, and, a more recent development, autonomous cars. Autonomous, or self-driving, vehicles rely on AI algorithms to navigate the road and make decisions that a driver would normally face while operating their vehicle. However, this concept can bring problems when thinking about questions such as liability for accidents and how a machine will make the moral choices that a driver makes, such as moving onto the incorrect side of the road to avoid a cyclist. How would a machine be able to think how a human driver would in making these decisions? If an autonomous vehicle were to cause an accident, which party would be blamed for it? The vehicle’s AI utilizes supervised, reinforcement, and unsupervised learning methods to try and answer these questions.

One of the most important uses of these learning methods is having the car learn to be able to identify objects in an image. For example, it would be trained to identify things such as humans and dogs in order to know when to stop for pedestrians. This is one example out of many, as the algorithms in self-driving cars need to learn how to recognize what speed to run at, center itself in a lane, as well as other traffic laws and safety protocols. As the technology rolls out and becomes more integrated and present in daily life, these questions, and many others, still remain as a result of this technology still being unfamiliar.