When generative AI came rolling out of development lab obscurity last year, most of us were surprised—even shocked. AI has always been a path to streamline production and distribution. In the past, it had come to replace blue-collar jobs. When I first tried it out in August, I wrote in this column, “Higher Ed, Meet GPT-3: We Will Never Be the Same!” I was astounded at the speed and cogency of GPT in answering complex questions with multiple paragraphs in the blink of an eye. Half a year later, it is time for an update as I prepare to travel to the UPCEA Annual Conference in Washington, D.C., where AI will be high among the topics of conversation.
First, know that this technology is nuanced—it is not just OpenAI’s ChatGPT. Generative AI covers a wide range of applications and tools. I now keep shortcuts to three different versions of generative AI on my laptop and phone home screen. I use all three daily. There are many hosts for the basic technology using various versions of knowledge bases and configured to present results with multiple features. Yes, I sprang for the $20 a month for ChatGPT Plus. It is ready for me on a click’s notice. Yet another app I have found useful every day is Perplexity. I am most taken with the auto-embedded citations of sources in the response, much like we do in research papers. This is most useful for deeper digging into topics. Its knowledge base is not delimited to 2021 as is ChatGPT. A somewhat flashier app is You.com. This is also free as of now and has the benefit of the added “/imagine” extension that opens an image generator. Refining prompt inputs to the text-to-image generator is worthwhile to have copyright free images designed to your specifications.
A fourth application that holds great potential to those of us in higher ed is ChatPDF! It is what you might imagine, a tool that allows you to load a PDF of up to 120 pages in length. You can then apply the now-familiar ChatGPT analysis approach to the document itself. Ask for a summary. Dig into specifics. This will be a useful tool for reviewing research and efficiently understanding complex rulings and other legal documents.
In these examples, the tools are mostly using GPT 3.5 with some advanced features promised to come in the full release of GPT 4.0, expected in the near future. OpenAI CEO Sam Altman hints at the soon-to-come multimodal GPT ahead of the 4.0 release, moving smoothly among text, images and other modes but not yet video, which he says will come later. Not to be forgotten is the Google product Bard, which launched to a stumbling start by giving a wrong answer in the very public reveal of the LaMBDA (Language Model for Dialogue Applications)–powered application. Now in restricted beta release, it is expected to provide a credible competition to ChatGPT.
Much has been made of the possibility of cheating by students using generative AI. Yet it seems that faculty are much more likely to use the products to enhance their teaching and research than students are to use it to cheat.
It is clear that generative AI has enormous potential in higher education. It auto-updates its knowledge base far more frequently and deeply than faculty possibly can. Yes, ChatGPT is delimited by a 2021 cutoff of knowledge, but that doesn’t have to be the case, nor will it be in the future. In a just released study, “How Will Language Modelers like ChatGPT Affect Occupations and Industries?” by Ed Felten (Princeton), Manav Raj (University of Pennsylvania) and Robert Seamans (New York University), the case is made that generative AI has the potential to impact teaching in many fields. The researchers applied an AI Occupational Exposure measure, developed in 2021, to determine which human requirements for a position most overlapped with generative AI capabilities. In one of the listings reviewed by the researchers of most likely to be affected career fields, 14 of the top 20 occupations were faculty positions in a wide range of fields. A CBS News report on the paper discusses the relative extent to which this predicts job augmentation or rather job substitution.
Longer-term advances in applying AI to learning are coming along, but at a measured—or sometimes a rather stumbling—pace. Neuralink, a subsidiary company owned by Elon Musk, continues its research into direct brain-computer interfacing. However, it has run into repeated regulatory obstacles. The FDA has just blocked a request to begin human trials in implanting chips into human brains. Some pessimism is centered on the company’s understanding of current regulations and attention to regulators.
Johns Hopkins researchers are examining the potential of using human stem cells to power a kind of organic computing:
It’s called organoid intelligence, or OI, and it uses actual human brain cells to make computing “more brain-like.” OI revolves around using organoids, or clusters of living tissue grown from stem cells that behave similarly to organs, as biological hardware that powers algorithmic systems. The hope—over at Johns Hopkins, at least—is that it’ll facilitate more advanced learning than a conventional computer can, resulting in richer feedback and better decision-making than AI can provide.
Of course, there are both technological and ethical considerations to be addressed as this model moves forward.
Much of the interesting work in the coming months will be to design interfaces to adapt the technology to actual work roles in both supportive and possibly in replacement modes. Generative AI already creates lesson plans, grades assignments, advises students, and answers learner questions. Can it competently take on class management and the other associated administrative tasks? And, certainly, generative AI has an important role to play in research. Will it become a formal or informal co-investigator and co-author of research? What status will we give generative AI in higher ed? And, what will happen to the human faculty aspirants who fail to measure up? I am looking forward to discussing these questions as we prepare for our generative AI partnership in higher ed.