Can machines learn? This is a question posed most famously in 1950 by British mathematician, Alan Turing. His position is that given enough memory capacity and sufficient programming, machines could imitate the human process of logic and reasoning. Turing is often considered the foundational figure in both modern computer science and the field of artificial intelligence (although that term was coined by later researchers).  

Artificial Intelligence (AI) is a broad term that captures an entire field of computer programs and research. AI uses computer programing to solve problems or perform functions that would normally require human reasoning. One example of this is the recommendations you get from streaming platforms such as Netflix and Spotify. These two companies take what songs or shows that you have enjoyed in the past and gives you the most similar options. For example, did you enjoy watching The Great British Baking Show? Well, you might also like to watch The Great British Baking Show: Holidays or The Great British Baking Show: The Professionals. These AI models work on using pattern recognition to make predictions with probability and statistics. AI seems so much more impressive because it can make these very complicated calculations with an enormous amount of data in moments rather than hours.

Notice the key terms here: AI models can follow logical paths (think if this is true, then do this) and look for general patterns. Does the model know that you like baking shows? Or does it understand that your friend was the one to watch true crime documentaries while visiting you? No. AI models are statistical in nature. It finds patterns in data without true understanding of the data.

Generative AI (or Gen AI) is a class of AI models that takes that predictive power and use it to create text or images rather than simply give a recommendation. When given a prompt, Gen AI models take every bit of information on the internet and create a product that is the mostly statistically likely answer to the prompt. Gen AI takes pattern recognition and uses it to approximate what a human could have produced though logical reasoning.

Gen AI models work through statistical models similar to autocomplete on your iPhone’s messages app. If you type “the”, Apple will give you some suggestions. Those suggestions are just the most common words you use after “the”. Gen AI models like ChatGPT use this same idea, but on a much larger scale to produce pages of text rather than word suggestions. The model has trained on, or learned from, millions of documents, books, and articles then uses math to guess at the most likely text to answer your prompt. What is unique about Gen AI is its ability to act with some degree of autonomy and to adapt to new information.

• Machine Learning: Machine learning is a type of artificial intelligence that allows computer programs to update and adapt to additional information or data. The attribute that makes machine learning unique is that the algorithm updates without human intervention or being coded to perform a certain task.

• Deep Learning: Deep Learning is a subset of machine learning that involves increasing the complexity of both the information used by a model and the output of the model. Deep learning uses neural networks (or multi-layered machine learning algorithms) to classify or categorize large amounts of information. These models were based on structure of the neural networks in the human brain. They attempt to emulate the sheer number of connections made between concepts to better classify and categorize data. 

• Large Language Models (or LLMs): LLMs are computer programs designed to produce human-sounding text. These programs take massive amounts of data from all books, reddit posts, Wikipedia pages, newspaper articles, etc. and finds patterns. We call this stage of the process training. From the patterns found during the training phase, LLMs can predict which text goes with which prompt. Language is about patters and LLMs find those patterns and use them to approximate human-generated text.

• “Hallucinations”: The first thing to understand about Generative AI is that it cannot differentiate between truth and falsehood. Therefore, when the Generative AI algorithm is asked a question for which it does not know the answer, it will just use statistics to give you the most likely answer, even if that answer is fiction. Generative AI models have been known to fabricate citations and even legal cases. It is best to not take anything AI generates at face value and to always verify information (and sources!).