Instructors have raised widespread concern about the impact of generative artificial intelligence on undergraduate education.

As we focus on undergraduate classes, though, we must not lose sight of the profound effect that generative AI is likely to have on graduate education. The question there, though, isn’t how or whether to integrate AI into coursework. Rather, it’s how quickly we can integrate AI into methods courses and help students learn to use AI in finding literature; identifying significant areas of potential research; merging, cleaning, analyzing, visualizing, and interpreting data; making connections among ideas; and teasing out significant findings. That will be especially critical in STEM fields and in any discipline that uses quantitative methods.

The need to integrate generative AI into graduate studies has been growing since the release of ChatGPT last fall. Since then, companies, organizations, and individuals have released a flurry of new tools that draw on ChatGPT or other large language models. (See a brief curated list below.) If there was any lingering doubt that generative AI would play an outsized role in graduate education, though, it evaporated with the release of a ChatGPT plugin called Code Interpreter. Code Interpreter is still in beta testing and requires a paid version of ChatGPT to use. Early users say it saves weeks or months of analyzing complex data, though.

OpenAI is admirably reserved in describing Code Interpreter, saying it is best used in solving quantitative and qualitative mathematical problems, doing data analysis and visualization, and converting file formats. Others didn’t hold back in their assessments, though.

Ethan Mollick, a professor at the University of Pennsylvania, says Code Interpreter turns ChatGPT into “an impressive data scientist.” It enables new abilities to write and execute Python code, upload large files, do complex math, and create charts and graphs. It also reduces the number of errors and fabrications from ChatGPT. He says Code Interpreter “is relentless, usually correcting its own errors when it spots them.” It also “ ‘reasons’ about data in ways that seem very human.”

Andy Stapleton, creator of a YouTube channel that offers advice to graduate students, says Code Interpreter does “all the heavy lifting” of data analysis and asks questions about data like a collaborator. He calls it “an absolute game changer for research Ph.D.s.”

Code Interpreter is just the latest example of how rapid changes in generative AI could force profound changes in the way we approach just about every aspect of higher education. Graduate education is high on that list. It won’t be long before graduate students who lack skills in using generative AI will simply not be able to keep up with those who do.

Other helpful research tools

The number of AI-related tools has been growing at a mind-boggling rate, with one curator listing more than 6,000 tools on everything from astrology to cocktail recipes to content repurposing to (you’ve been waiting for this) a bot for Only Fans messaging. That list is very likely to keep growing as entrepreneurs rush to monetize generative AI. Some tools have already been scrapped or absorbed into competing sites, though, and we can expect more consolidation as stronger (or better publicized) tools separate themselves from the pack.

The easiest way to get started with generative AI is to try one of the most popular tools: ChatGPT, Bing Chat, Bard, or Claude. Many other tools are more focused, though, and are worth exploring. Some of the tools below were made specifically for researchers or graduate students. Others are more broadly focused but have similar capabilities. Most of these have a free option or at least a free trial.

• Research assistants like Research Rabbit, ARIA (an add-on to Zotero), Litmaps, Connected Papers, and ClioVis help scholars find and organize literature, visualize connections among ideas, share collections of research, and keep up with new research.
• Search tools like Consensus, Elicit, Scite, SciSpace, Snowball, System Pro, You, Metaphor, and Inciteful help guide researchers toward specific types of literature.
• Tools for interacting with articles. These allow researchers to query literature with natural language questions and to find ideas or connections they might have overlooked. They include Explainpaper, which allows you to highlight text and ask for explanations; PaperBrain; Sharly, which has a generous free version; ChatPDF; MapDeduce; Census GPT, which connects to U.S. Census data; Humata; Upword; OpenRead; PDF.ai; Kadoa; and Cerelyze, which allows you interact with articles and translate method into code.
• Summary tools provide quick synopses of articles, papers, and books, making it easier to work through large amounts of literature. These include Scholarcy, which highlights key areas of academic articles and allows you to build a personal library; Lateral; MapDeduce; ShortForm.ai, which uses a browser plugin; TinyWow, which offers a suite of free AI-powered tools; Summarize.tech and Eightify, which summarize YouTube videos; and BooksAI, which provides summaries of books.
• Several data tools allow researchers to query their quantitative data with natural language questions. These include Julius; Formulas HQ; Chatwithdata; Chatcsv; ChatNode; Tomat; Chartify; DataSquirrel; Chaindesk; and JADBio, which focuses on genomics, medicine, health, and drug discovery. Web-scraping tools like Simplescraper help automate time-consuming data-gathering.
• Writing tools like Jenni, Paperpal, Writefull, and Wisio promise to make academic writing easier.

How to use Code Interpreter

You will need a paid ChatGPT account. Jon Martindale of Digital Trends explains how to get started. An OpenAI forum offers suggestions on using the new tool. Members of the ChatGPT community forum also offer many ideas on how to use ChatGPT, as do members of the OpenAI Discord forum. (If you’ve never used Discord, here’s a guide for getting started.)

A year after the release of a know-it-all chatbot, educators have yet to find a satisfying answer to a nagging question: What are we supposed to do with generative artificial intelligence?

One reason generative AI has been so perplexing to educators is that there is no single step that all instructors can take to make things easier. Here are a few things what we do know, though:

• 

  Students are using generative AI in far larger numbers than faculty, and some are using it to complete all or parts of assignments. A recent Turnitin poll said 22% of faculty were using generative AI, compared with 49% of students.

• Students in other developed countries are far more likely to use generative AI than students in the U.S., two recent polls suggest.
• Students are as conflicted as faculty about generative AI, with many worried about AI’s impact on jobs, thinking, and disinformation.
• Many faculty say that students need to know how to use generative AI but also say they have been reluctant to use it themselves.
• Detectors can provide information about the use of generative AI, but they are far from flawless and should not be the sole means of accusing students of academic misconduct.

Perhaps the biggest lesson we have learned over the past year is that flexibility in teaching and learning is crucial, especially as new generative AI tools become available and the adoption of those tools accelerates.

We don’t really have an AI problem

It’s important to understand why generative AI has made instructors feel under siege. In a forthcoming article in Academic Leader, I argue that we don’t have an AI problem. We have a structural problem:

Unfortunately, the need for change will only grow as technology, jobs, disciplines, society, and the needs of students evolve. Seen through that lens, generative AI is really just a messenger, and its message is clear: A 19th-century educational structure is ill-suited to handle changes brought on by 21st-century technology. We can either move from crisis to crisis, or we can rethink the way we approach teaching and learning, courses, curricula, faculty roles, and institutions.

That’s not a message most faculty members or administrators want to hear, but it is impossible to ignore. Colleges and universities still operate as if information were scarce and as if students can learn only from faculty members with Ph.D.s. The institutional structure of higher education was also created to exclude or fail students deemed unworthy. That’s much easier than trying to help every student succeed. We are making progress at changing that, but progress is slow even as change accelerates. I’ll be writing more about that in the coming year.

Faculty and staff are finding ways to use AI

Many instructors have made good use of generative AI in classes, and they say students are eager for such conversations. Here are a few approaches faculty have taken:

• Creating AI-written examples for students to critique.
• Allowing students to use AI but asking them to cite what AI creates and separately explain the role AI played in an assignment.
• Having students use AI to create outlines for papers and projects, and refining goals for projects.
• Allowing full use of AI as long as students check the output for accuracy and edit and improve on the AI-generated content.
• Having students design posters with AI.
• Using examples from AI to discuss the strengths and weaknesses of chatbots and the ethical issues underlying them.
• Using paper and pencils for work in class. In recent discussions with CTE ambassadors, the term “old school” came up several times, usually in relation to bans on technology. As appealing as that may seem, that approach can put some students at a disadvantage. Many aren’t used to writing by hand, and some with physical impediments simply can’t.
• For non-native English speakers, generative AI has been a confidence-builder. By evaluating their writing with a grammar checker or chatbot, they can improve phrasing and sentence construction.
• Some faculty members say that generative AI saves time by helping them create letters of recommendation, event announcements, and case studies and other elements for class.

Sara Wilson, an associate professor of mechanical engineering and a CTE faculty fellow, uses what I think is probably the best approach to AI I’ve seen. In an undergraduate course that requires a considerable amount of programming, she allows students to use whatever tools they wish to create their coding. She meets individually with each student – more than 100 of them – after each project and asks them to explain the concepts behind their work. In those brief meetings, she said, it is fairly easy to spot students who have taken shortcuts.

Like faculty, students are often conflicted

Many students seem as conflicted as faculty over generative AI. In a large introductory journalism and mass communications class where I spoke late this semester, I polled students about their AI use. Interestingly, 21% said they had never used AI and 45% said they had tried it but had done little beyond that. Among the remaining students, 27% said they used AI once a week and 7% said they used it every day. (Those numbers apply only to the students in that class, but they are similar to results from national polls I mention above.)

In describing generative AI, students used terms like “helpful,” “interesting,” “useful” and “the future,” but also “theft,” “scary,” “dangerous,” and “cheating.” Recent polls suggest that students see potential in generative AI in learning but that they see a need for colleges and universities to change. In one poll, 65% of students said that faculty needed to change the way they assess students because of AI, the same percentage that said they wanted faculty to include AI instruction in class to help them prepare for future jobs.

Students I’ve spoken with describe AI as a research tool, a learning tool, and a source of advice. Some use AI as a tutor to help them review for class or to learn about something they are interested in. Others use it to check their writing or code, conduct research, find sources, create outlines, summarize papers, draft an introduction or a conclusion for a paper, and help them in other areas of writing they find challenging. One CTE ambassador said students were excited about the possibilities of generative AI, especially if it helped faculty move away from “perfect grading.”

Time is a barrier

For faculty, one of the biggest challenges with AI is time. We’ve heard from many instructors who say that they understand the importance of integrating generative AI into classes and using it in their own work but that they lack the time to learn about AI. Others say their classes have so much content to cover that working in anything new is difficult.

Instructors are also experiencing cognitive overload. They are being asked to focus more on helping students learn. They are feeling the lingering effects of the pandemic. In many cases, class sizes are increasing; in others, declining enrollment has created anxiety. Information about disciplines, teaching practices, and world events flows unendingly. “It’s hard to keep up with everything,” one CTE ambassador said.

Generative AI dropped like a boulder into the middle of that complex teaching environment, adding yet another layer of complexity: Which AI platform to use? Which AI tools? What about privacy? Ethics? How do we make sure all students have equal access? The platforms themselves can be intimidating. One CTE ambassador summed up the feelings of many I’ve spoken with who have tried using a chatbot but weren’t sure what to do with it: “Maybe I’m not smart enough, but I don’t know what to ask.”

We will continue to provide opportunities for instructors to learn about generative AI in the new year. One ongoing resource is the Generative AI and Teaching Working Group, which will resume in the spring. It is open to anyone at KU. CTE will also be part of a workshop on generative AI on Jan. 12 at the Edwards Campus. That workshop, organized by John Bricklemyer and Heather McCain, will have a series of sessions on such topics as the ethics of generative AI, strategies for using AI, and tools and approaches to prompting for instructors to consider.

We will also continue to add to the resources we have created to help faculty adapt to generative AI. Existing resources focus on such areas as adapting courses to AI, using AI ethically in writing assignments, using AI as a tutor, and handling academic integrity. We have also provided material to help generate discussion about the biases in generative AI. I have led an effort with colleagues from the Bay View Alliance to provide information about how universities can adapt to generative AI. The first of our articles was published last week in Inside Higher Ed. Another, which offers longer-term strategies, is forthcoming in Change magazine. Another piece for administrators will be published this month in Academic Leader.

Focusing on humanity

If generative AI has taught us anything over the past year, it is that we must embrace humanity in education. Technology is an important tool, and we must keep experimenting with ways to use it effectively in teaching and learning. Technology can’t provide the human bond that Peter Felten talked about at the beginning of the semester and that we have made a priority at CTE. Something Felten said during his talk at the Teaching Summit is worth sharing again:

“There’s literally decades and decades of research that says the most important factor in almost any positive student outcome you can think about – learning, retention, graduation rate, well-being, actually things like do they vote after they graduate – the single biggest predictor is the quality of relationships they perceive they have with faculty and peers,” Felten said.

Technology can do many things, but it can’t provide the crucial human connections we all need.

In an ambassadors meeting in November, Dorothy Hines, associate professor of African and African-American studies and curriculum and teaching, summed it up this way: “AI can answer questions, but it can’t feel.” As educators, she said, it’s important that we feel so that our students learn to feel.

That is wise advice. As we continue to integrate generative AI into our work, we must do so in a human way.

Doug Ward is associate director of the Center for Teaching Excellence and an associate professor of journalism and mass communications.

If you are sitting on the fence, wondering whether to jump into the land of generative AI, take a look at some recent news – and then jump.

• Three recently released studies say that workers who used generative AI were substantially more productive than those who didn’t. In two of the studies, the quality of work also improved.
• The consulting company McKinsey said that a third of companies that responded to a recent global survey said they were regularly using generative AI in their operations. Among white-collar professions that McKinsey said would be most affected by generative AI in the coming decade are lawyers and judges, math specialists, teachers, engineers, entertainers and media workers, and business and financial specialists.
• The textbook publisher Pearson plans to include a chatbot tutor with its Pearson+ platform this fall. A related tool already summarizes videos. The company Chegg is also creating an AI chatbot, according to Yahoo News.
• New AI-driven education platforms are emerging weekly, all promising to make learning easier. These include: ClaudeScholar (focus on the science that matters), SocratiQ (Take control of your learning), Monic.ai (Your ultimate Learning Copilot), Synthetical (Science, Simplified), Upword (Get your research done 10x faster), Aceflow (The fastest way for students to learn anything), Smartie (Strategic Module Assistant), and Kajabi (Create your course in minutes).

My point in highlighting those is to show how quickly generative AI is spreading. As the educational consultant EAB wrote recently, universities can’t wait until they have a committee-approved strategy. They must act now – even though they don’t have all the answers. The same applies to teaching and learning.

A closer look at the research

Because widespread use of generative AI is so new, research about it is just starting to trickle out. The web consultant Jakob Nielsen said the three AI-related productivity studies I mention above were some of the first that have been done. None of the studies specifically involved colleges and universities, but the productivity gains were highest in the types of activities common to colleges and universities: handling business documents (59% increase in productivity) and coding projects (126% increase).

One study, published in Science, found that generative AI reduced the time professionals spent on writing by 40% but also helped workers improve the quality of their writing. The authors suggested that “ChatGPT could entirely replace certain kinds of writers, such as grant writers or marketers, by letting companies directly automate the creation of grant applications and press releases with minimal human oversight.”

In one of two recent McKinsey studies, though, researchers said most companies were in no rush to allow automated use of generative AI. Instead, they are integrating its use into existing work processes. Companies are using chatbots for things like creating drafts of documents, generating hypotheses, and helping experts complete tasks more quickly. McKinsey emphasized that in nearly all cases, an expert oversaw use of generative AI, checking the accuracy of the output.

Nonetheless, by 2030, automation is expected to take over tasks that account for nearly a third of current hours worked, McKinsey said in a separate survey. Jobs most affected will be in office support, customer service, and food service. Workers in those jobs are predominantly women, people of color, and people with less education. However, generative AI is also forcing changes in fields that require a college degree: STEM fields, creative fields, and business and legal professions. People in those fields aren’t likely to lose jobs, McKinsey said, but will instead use AI to supplement what they already do.

“All of this means that automation is about to affect a wider set of work activities involving expertise, interaction with people, and creativity,” McKinsey said in the report.

What does this mean for teaching?

I look at employer reports like this as downstream reminders of what we in education need to help students learn. We still need to emphasize core skills like writing, critical thinking, communication, analytical reasoning, and synthesis, but how we help students gain those skills constantly evolves. In terms of generative AI, that will mean rethinking assignments and working with students on effective ways to use AI tools for learning rather than trying to keep those tools out of classes.

If you aren’t swayed by the direction of businesses, consider what recent graduates say. In a survey released by Cengage, more than half of recent graduates said that the growth of AI had left them feeling unprepared for the job market, and 65% said they wanted to be able to work alongside someone else to learn to use generative AI and other digital platforms. In the same survey, 79% of employers said employees would benefit from learning to use generative AI. (Strangely, 39% of recent graduates said they would rather work with AI or robots than with real people; 24% of employers said the same thing. I have much to say about that, but now isn’t the time.)

Here’s how I interpret all of this: Businesses and industry are quickly integrating generative AI into their work processes. Researchers are finding that generative AI can save time and improve work quality. That will further accelerate business’s integration of AI tools and students’ need to know how to use those tools in nearly any career. Education technology companies are responding by creating a large number of new tools. Many won’t survive, but some will be integrated into existing tools or sold directly to students. If colleges and universities don’t develop their own generative AI tools for teaching and learning, they will have little choice but to adopt vendor tools, which are often specialized and sold through expensive enterprise licenses or through fees paid directly by students.

Clearly, we need to integrate generative AI into our teaching and learning. It’s difficult to know how to do that, though. The CTE website provides some guidance. In general, though, instructors should:

• Learn how to use generative AI.
• Help students learn to use AI for learning.
• Talk with students about appropriate use of AI in classes.
• Experiment with ways to integrate generative AI into assignments.

Those are broad suggestions. You will find more specifics on the website, but none of us has a perfect formula for how to do this. We need to experiment, share our experiences, and learn from one another along the way. We also need to push for development of university-wide AI tools that are safe and adaptable for learning.

The fence is collapsing. Those who are still sitting have two choices: jump or fall.

AI detection update

OpenAI, the organization behind ChatGPT, has discontinued its artificial intelligence detection tool. In a terse note on its website, OpenAI said that the tool had a “low rate of accuracy” and that the company was “researching more effective provenance techniques for text.”

Meanwhile, Turnitin, the company that makes plagiarism and AI detectors, updated its figures on AI detection. Turnitin said it had evaluated 65 million student papers since April, with 3.3% flagged as having 80% to 100% of content AI-created. That’s down from 3.5% in May. Papers flagged as having 20% or more of content flagged rose slightly, to 10.3%.

I appreciate Turnitin’s willingness to share those results, even though I don’t know what to make of them. As I’ve written previously, AI detectors falsely accuse thousands of students, especially international students, and their results should not be seen as proof of academic misconduct. Turnitin, to its credit, has said as much.

AI detection is difficult, and detectors can be easily fooled. Instead of putting up barriers, we should help students learn to use generative AI ethically.

Doug Ward is the associate director of the Center for Teaching Excellence and an associate professor of journalism and mass communications.