In our modern digitized society, many buzzwords like Artificial Intelligence (AI) and ChatGPT dominate news headlines. These topics accompany concerns regarding academic integrity and the security of various jobs, from legal workers to market research analysts. In my own life, I constantly see the use of machine learning models to aid with essay writing or even generating art. Although some people may find these applications fascinating, others find them scary and worrisome. Therefore, it is important for us to be informed about the benefits and disadvantages of machine learning and cybersecurity.

Machine learning is a subset of artifical intelligence and computer science. Generally, it deals with training computer models to perform various tasks on unknown data, such as learning how to identify objects and categorize images. The typical mechanism behind these computer models is a neural network. A neural network, which is modeled to the human neuron, contains many layers of artificial “neurons”, which take several inputs into a “linear function” and output a number to the next layer, and so on. Eventually, when this output reaches the last layer of the network, it categorizes or identifies the original input using true/false statements. In more metaphorical terms, the process of machine learning can be likened to a game of telephone where the machine is trying to relay quantitative information to output a qualitative representation.

Besides the jargon, machine learning is particularly useful in performing tedious categorization tasks and social media recommendation algorithms; however, machine learning also has its downsides. A neural network, like ChatGPT, is actually powered by over a trillion parameters that define a large function (called the “cost function”). These numbers are usually only known by a select group of executives and founders, and misuse of these models could put a vast majority of workers’ jobs at risk. Therefore, there is a need for greater transparency in the world of machine learning research, so that AI and machine learning may be used for the betterment of society rather than for only a handful of greedy individuals. Fortunately, some AI companies like Anthropic already pursue this way of thinking, which provides some hope for the future of our society — if others follow suit.

Secondly, cybersecurity is a current concern due to the rise in the need for digital privacy for both users and companies alike. From a more mathematical perspective, cybersecurity and cryptography are also powered by large datasets of numbers (prime numbers to be specific), similar to the aforementioned machine learning mechanisms. One long-held concept related to these prime numbers are Carmichael numbers, which behave very similarly to prime numbers but are actually composite numbers formed from multiplying two prime numbers together. Though a more rigorous definition exists, common examples of Carmichael numbers include 561, 1105, and 1729. The relation between Carmichael numbers and cybersecurity arises when they are mistakenly used in these large datasets of prime numbers. If Carmichael numbers are used in place of a prime number, they present harmful dangers for the cybersecurity of websites, email encryptions, etc. Thus, we see a crucial need for increased research on these types of numbers. Although we often jump to advocacy or stringent legislation as effective solutions for cybersecurity, we must also consider and address the theoretical weaknesses of this ubiquitous issue.

This article was edited by Grace Hur.