Technology Instruction: AI

Artificial Intelligence, or AI, is a term used to describe the process of computers mimicking human intelligence. They do this by analyzing tons of data at once and presenting it in a way that is approachable to humans. The computers are not actually "thinking" as you may imagine a human doing, as they still don't have human brains. Rather, they are looking at huge datasets that would take humans a long time to comb through, and drawing conclusions based on the available data.

*Please note that the field of AI is rapidly expanding, so this is not an exhaustive list of the concepts surrounding AI or the various AI products.*

Algorithms- Sets of rules that tell a computer what to do with the data it receives. You may have heard of "the algorithm" on social media, which is the social media site analyzing your behavior (likes, comments, shares) and predicting what content you will continue to find engaging. It also chooses what content to hide from you.

Deep Learning- Deep learning is a type of machine learning (see definition below) where three of more layers of a neural network process data on a more complex level. Neural networks are programs set up to make decisions similar to how humans do- by passing data between different nodes (similar to the firing of human brain neurons) and weighing actions against each other. Deep learning models can complete the same tasks that machine learning models can, but are typically better at them because they can organize how data relates in a hierarchy to other data.

Generative AI- Usually referred to as gen AI, these are AI programs that are able to stitch together patterns learned from training data to create completely new content such as text, images, and pictures. The Google Say What You See game is an example of gen AI, where you give keywords to the computer and it uses those to create a new image. Gen AI has to learn from existing content, which is why there is much debate about it in creative fields. If a gen AI is reliable enough at replicating a specific style, creators fear it could make them irrelevant as the tasks previously assigned to them could be given to an AI instead.

Hallucinations- This term is used when an AI gives an answer that is factually incorrect or may not exist at all. AI chatbots present all information as if it correct, so it can be difficult to tell without some fact checking if that is the case. Hallucination can look like citing journal articles or studies that don't actually exist, talking about people that aren't real and attributing actions to them, and talking about events that never happened- among other things. 

Large Language Models (LLMs)- LLMs are designed to interpret and generate text like a human would. A common example of this is ChatGPT. When you input text into an LLM, the response it gives you is essentially a long string of predictions. The model uses the preceding word to guess what the next word should be, and keeps doing this until it comes to the natural end of a sentence. This is where hallucinations can come into play. The LLM is not checking that the words it is predicting to complete the sentence are actually accurate, so while it is worded like it could be, it may not be.

Machine Learning- AI by default does not "know" anything, so it is up to humans to feed it data to interpret. When an AI has enough data of a certain type, it can use that to make predictions about future events. Once the AI has enough base data, it can start recognizing patterns without being explicitly programmed to recognize them. An example of this is the spam folder in your email. When you mark an email as spam, the AI can learn what the typical markers of spam messages are (suspicious links, certain email addresses, etc) and start sending similar messages to Spam without anything telling it to. Marking something as incorrectly sent to Spam also adds to the data the AI can use.

Natural Language Processing (NLP)- Chatbots are trained to recognize human languages and the syntax that humans use to communicate. What this means for you, is that you can communicate with an LLM as if you were having a conversation with a friend. You can ask questions or give suggestions in the same way you would to a human, and the LLM will respond in a way that makes sense to humans.

Vibe Coding- LLMs can generate computer code as well as standard sentences. Vibe coding is a term used to describe the act of telling an LLM to create the code for something, rather than a person having to write out the code. The word vibe is used as a sort of slang, indicating that you are just "going with the vibes" of the code the AI gives you. Vibe coding can be a handy tool for those who have no coding experience to start making projects. However, LLMs frequently create coding errors, which you may not know how to fix if you don't understand the code the AI is writing.