Intro to AI:
You're here because you may have heard of ChatGPT, or about actors striking because studios want rights to their AI replicas, or even because you want to know more about how that Roomba found its way into and around your living room. AI generated content and AI powered tools are here, and there is a lot to know. This guide presents the basics, with some links that can help you follow up on your own and resources that you can use to understand the implications of these tools in the classroom.
Let's start with ChatGPT and language models:
ChatGPT is a chatbot used to generate text after a user enters a prompt. This chatbot uses a machine learning based language model to generate that text, having gone through versions developed by the OpenAI company, with the latest version GPT-4 released on March 14, 2023.
Language models are natural language processing applications used to analyze bodies of text to provide a probability distribution over sequences of words. Essentially, they can create strings of probably related words. Based on whatever input, a chatbot is meant to give the "most likely" output. This can be accurate, if the data it was trained on was accurate, or it can be a jumbled mess, as humans can be, if the data it was trained on was a jumbled mess, just as prone to errors and bias as we are. Interestingly, even with accurate training data, these kinds of chatbots are prone to what is sometimes known as "hallucination", essentially a seemingly confident response by an AI that does not seem to be supported by its training data. Which again, can be interesting! Other times, the chatbot's results can be shockingly readable or even quite predictably repetitive. The point is, results vary, and these kinds of tools should be used with care, if we choose to use them at all.
Following up:
There are also ethical questions to consider regarding the sources of training data for these kinds of programs, knowing that biases of input will inform biases in any output. This is obviously a complex problem. Here is just one blog article by Matt Beale that can help expand on the complexities of ethical sourcing of AI training data. If you are interested in ChatGPT's training data in particular, you can read more here.
If you are interested in the history of AI more generally, see the "History of AI" tab on this page. There is also an "In the News" tab with some of the latest updates regarding ChatGPT and other generative AI.
To learn more about the history of AI, the Elements of AI series created by MinnaLearn and the University of Helsinki offers a free online course to help deepen your understanding of how we got to the chatbots, image generators, and applications of AI we have today. The "Introduction to AI" course helps those interested to learn more about what AI is, what is possible (and not possible) with AI, and how it affects our lives -- no complicated math or programming required. It includes six chapters, combining theory with practical exercises and can be completed at your own pace.
So what can AI tools do for us today? What can they not do? Ethan Mollick, a professor at the Wharton School of the University of Pennsylvania, has created a useful comparison table in a July 15, 2023 blog post:
News stories about ChatGPT are being published all the time, especially as ChatGPT itself evolves. Check out these stories for the latest information from a Google news feed:
Here is a feed for AI issues more generally:
Here are a few other curated article selections about developments in AI more generally:
