What are AI Tools?

To put it simply, AI Tools are programs which can produce content for users based on their inputs (usually called "prompts"). The content they can produce ranges from essays, to code snippets, to political speeches, digital animation, data analysis, and more. Indeed, AI Tools are, in many ways, the equivalent of 3D printers in that they help us skip many of the established "middle steps" most of us are used to in various processes. As an example, a tool like ChatGPT is meant to take away the burdens of research, notetaking, and compiling that information into "our own words" and simply produce an essay based on our subject of interest. Whether or not ChatGPT is successful at that task or not is something we'll save for later, but it's worth noting first that ChatGPT isn't just any sort of AI Tool, it's what's known as a Generative AI Tool.

What is a Generative AI Tool?

Generative AI Tools are similar to any AI Tool in terms of what they're asked to produce, but how they do so is very different. The core difference is that a Generative AI Tool produces new, original content rather than being limited to analysis and compilation. If you ask a traditional AI to give you a animated short film from 1928, it might grab and show you Steamboat Willie. If you ask a Generative AI to do the same, it will make an entirely new animated short film for you, never seen before, in the styles popular during 1928.

Generative AI is the type of AI Tool we will be referring to for the rest of this page when we say AI Tools, even though the term can be used more broadly. We do this not just for brevity's sake, but because many of the linked resources and articles do the same thing, so consistent terminology wins out over specificity.

What are some examples of AI Tools?

  • ChatGPT - Arguably the reason you're reading this page. ChatGPT generates essay-style response to prompts, and can even produce basic code in Javascript, Python, and other coding languages. It is what's known as a "chat bot" because it accepts prompts and replies in a format similar to online chatrooms.
  • Jasper - AI Art and writing generator. Its website boasts that you can "generate months of social media content in minutes" and in 26 languages too.
  • LaMDA - Google's AI Chat bot.
  • MidJourney - An art generator which generates art based on written text. One of several AI Art Generators currently involved in yet ANOTHER lawsuit, this time with Disney.
  • OpenAI Codex - Code generator. Takes a written description in natural language and converts it into one of 12 coding languages.
  • Stable Diffusion - Art Generator. One of the first, though certianly nowhere close to the last.

How do they work?

AI Tools work in a similar way to the human brain, although at a much faster rate, and with many of the expected pitfalls of doing so. While the exact specifics of how each AI Tool works are well-kept secrets, the general process begins with the AI being "fed" as many examples as it can - usually thousands, or even hundreds of thousands of examples - of what it is expected to produce. From there, it analyzes the examples, breaking them down into smaller parts to figure out how each example compares with the others, and categorizing those results, usually with the help of a human to help correct it at first. For example, an AI Tool designed to write screenplays might break down the screenplays it has collected by scene, then each scene by paragraph, and then each paragraph by line. It would then sort each line of text into categories that have been identified for it like "dialogue" and "camera directions" and "setting information." This is where things get a little hazy, but at this point the AI then analyzes the patterns within the different texts, based on scene, paragraph, and line, identifying commonalities between them, assigning percentages to the odds of "events" (e.g. four long paragraphs in a row, or a scene less than two lines long), compiling keywords, and so on. Finally, a user inputs a prompt, and the AI takes the various breakdowns it has come up with to produce original content based on what results it finds best fits that prompt. The more AI Tools are used, and the more data they are given, the more-refined they typically become, better able to produce exactly what a user wants with less-specific prompting.

In theory, and many even find in practice, this saves both time and money, reduces strain, increases efficiency, and produces quality content with no more effort than it takes to type a few words. However, the streets aren't quite paved with gold.

The Problems With AI

Because AI like ChatGPT are not what's known as General AI - AI which can actually think and problem-solve like human beings - they can only do what they are specifically programmed to do, despite what people and companies might use them for. In the case of ChatGPT, it's programmed to produce text based on analysis of word probabilities. It doesn't care if the words it produces are factually accurate, if the sources it cites are made up, or if the answer it gives doesn't actually make sense. It has no capacity to care, only to produce text based on analysis of word probabilities. Yes, they can produce text at a rapid rate, and what they produce may, in fact, be factually accurate, with proper sources, and make sense, but it has no ability to analyze any of these three factors. So, when someone uses them for research purposes, or to put together arguments for their side of a debate, the AI will give them the text they ask for, but not necessarily the text they need. Newer models are being trained constantly to try and work around these issues, but how does one program fact checking without first fact checking all of its input sources? What about new information that comes to light from sources not in its input? These questions are only the beginning.

There are other concerns from ethical to environmental, from exaggerated to absurd, and from totally innocuous to globally significant. We do not have time to get into all of them now, but we hope some of the resources below will help you get a better idea of the current picture.

Why are we talking about AI Tools?

Right now, the big conversation around AI Tools at MSUB is their effect on higher education. Are AI Tools the future? Are they plagiarism? How do we use them in the classroom? How do we stop students from using them at home? What are they actually capable of? What do I do if AI makes me obsolete?

There are a lot of questions right now, and three times as many answers from all across the educational spectrum. As they are now, it's safe to say that no one's going to lose their job over this anytime soon. AI Tools aren't sophisticated-enough to be error-free, and so they are still easy to detect, often to the untrained eye. Though this won't always be the case, many are looking towards the future, and what AI Tools might mean for instructors and students alike. We at the CTL are working to get you an in-depth analysis of the situation soon, but for now here is a list of resources we've found which help to dissect, explain, and expand on the start we have here. Check back soon for more up-to-date information!