AI for Teaching and Learning in Higher Education: Preparing Students for an AI Future

From helping educators and students to track progress and activity to answering questions about content and concepts, AI is already being used to support educational outcomes for learners. 

Automating supports based on progress and performance

Tools like OnTask allow educators to set up emails tailored to notify and encourage students based on if/then rules. For instance, IF a student hasn't completed key learning activities by a set date, THEN that student is sent an email pre-written by the instructor that encourages them to do so. If a student seems to be struggling with activities related to a particular concept, an email could be automatically generated to suggest additional resources and supports. This is already available for academics at UniSA. 

Personalised, cognitive-based learning with custom-trained AIs

Because GPT AI models can be trained on specific content or work with other systems, the outputs AI creates can be based on credible, high-quality, vetted content.

The Khan Academy, an online educational platform that provides video lessons and interactive exercises, has done just that with a tool aimed at both educators and students. Educators (who have to pay to use the platform unless paid for by their district) can seek ideas on developing lesson plans and content based on Khan Academy's materials and students can use the AI dubbed Khanmigo to seek clarity on concepts and materials without getting their work done for them. 

Eight South Australian high schools have trialled an AI app, designed in partnership with Microsoft, to show them how to use AI to support learning with guardrails built for student safety. 

Researchers at UniDistance Suisse (Baillifard et al., 2023) recently published findings of research they did into the implementation of an AI tutor in a neuroscience course fed with course materials and found that students who actively engaged with the AI tutor led to an average improvement of up to 15 percentile points. 

Harvard University is integrating AI into their introduction to coding course, to help students de-bug coding and to answer basic questions.

AI is also being combined with other learning technologies such as virtual- and augmented-reality-enabled immersive learning as is the case with the South Australian educational technology company Lumination and educational systems, schools and industry are trialling various applications of these tools. 

Organisations are already able to use services like Microsft's Azure to create and train their own AI models on internal data to produce results.

Individuals are also able to do so - for a variety of purposes. For instance, inventor and futurist Ray Kurzweil has spent a decade or more on a very personal project: being able to connect with his deceased father (whom he lost at age 22) by feeding an AI his father Fred's journals, musical compositions, lecture notes and letters and training it so that the AI produces a simulacrum of a conversation with his father.  

Personalised, cognitive-based learning with general purpose, generative AIs

Even general-purpose, generative AIs can serve as a jumping-off point for further exploration of a topic or concept when fed a deliberately structured prompt, keeping in mind the risks of "hallucinations", inherent biases and limited access to current and/or paywalled data as well as privacy and IP concerns. There is also the very real concern that students will rely too heavily on AI writing output and miss out on learning the fundamental writing and analysis skills that help them use AI effectively. Educators need to develop AI skills and experience of their own precisely because of these risks. And teaching students to use AI to support learning means integrating AI use into coursework. (Mollick and Mollick, 2023) 

AI can be used as a feedback generator, tutor, team coach and even let your students take on the role of teacher. In addition to their excellent YouTube series, Ethan and Lilach Mollick have written a series for Harvard Business Education publishing on these topics that includes the wording for detailed prompts for these user scenarios. The fifth video in their YouTube series on Practical AI for Instructors and Students serves as an introduction to how to use general purpose, generative AI tools for personalised, cognitive-based learning and guidance for educators on how to effectively guide and integrate the use of AIs in the classroom.

AI agents vs Software

Currently, if you want to create a document, presentation, or spreadsheet and create or edit images to accompany them - you need three or four different programs. You may need an entirely different device to send photos and messages when you're out and about or to turn your air conditioning on before you get home or to record your ideas for a presentation you've been asked to give or to confirm you really DO remember that 3 pm dental appointment.

Microsoft founder and self-confessed lover of software Bill Gates predicts that in a few years, this will change completely - you will simply tell your device, in everyday language, what you want to accomplish and it will deliver. Depending on your comfort levels with cognitive offloading and sharing your personal information (and your employer's IP and privacy policies), your device will understand your unique context and needs. He sees a near future where we all have a truly personal assistant with us at all times.

There are issues that will need plenty of discussion and resolution, and Gates acknowledges this. These include the widening of the digital divide between the haves and have-nots, privacy, loss of control and the erosion of skills as well as the vulnerability of systems to attack and the risk of building our society on technologies that can let us down. We only have to look at the Optus outage in early November 2023 that caused small businesses that are cash-free to lose a half day's or more of income, took infrastructure such as trains offline and interfered with the operations of dozens of hospitals. 

Using AI as a tool, not an author

Generative, general purpose AIs can be useful tools because of their ability to do basic spelling and grammar checks, tone, basic coding, provide contextualised feedback, create exemplars, create drills on foundational knowledge, help us overcome creative blocks (show me an image that represents X concept), summarise videos or difficult papers and use the give and take/iterative process to wrestle with a concept or idea.

Where AI use goes over the ethical line is when either its output is misrepresented as someone's original work and/or in academic work where permitted AI use has been defined and the user goes beyond that permitted use.

Students are asking for guidance on what permitted AI use is in their courses. See Episode 2: Artificial Intelligence: Friend or foe? of the UniSA student podcast series "The Study Room". 

Guides like the AI Assessment Scale described in the Assessment Design section of this guide (see heading "Defining permitted AI use") can establish a shared understanding between the teaching team and students about what acceptable AI use looks like for a given activity or assessment.

It bears repeating to students as often as possible: current (May 2024) general purpose, generative text AIs produce a remix of other people's ideas, not an original view based on experience, robust interrogation of research data or discussions with experts and elders. In his May 2024 titled 'Partial regurgitation and how AI really work', neuroscientist and author Marcus's describes AI output as, at best, poor quality paraphrasing:

What the LLM does is more akin to what some high school students do when they plagiarize: change a few words here or there, while still sticking close to the original.

LLMs are great at that. By clustering piles of similar things together in a giant n-dimensional space, they pretty much automatically become synonym and pastische (sic) machines, regurgitating a lot of words with slight paraphrases while adding conceptually little, and understanding even less.

Having competent humans in the loop to direct, reject, edit and tweak AI output is still very much desirable, but we're in the early days of its public use.

What is in store for entry-level office jobs that involve writing, administration, coding and graphic design in the next few years? The answer is - no one is certain quite yet. 

The jobs that are most impacted are those that involve activities that AI can do more efficiently than humans. Repetitive, data-driven tasks, entry-level programming, analytics, engineering, graphic design, human resources, marketing, advertising, some aspects of customer support, legal research and of course teaching and learning design. Some experts (Eloundou et al., 2023) posit that software and tooling built on top of large language models could complete between 47% and 56% of all worker tasks in the US (obviously depending on the industry and application) significantly faster at the same level of quality. 

Research by the Boston Consulting Group (Fabrizio Dell'Acqua et al., 2023) on 18 different tasks done at the consulting company, found that of the 758 consultants involved in the study, those who used AI outperformed those who did not - by a considerable margin and by every way performance is measured. Interestingly too - those consultants who measured the worst before using AI had the biggest increase in their performance, indicating that AI use can be a leveller.

However, when posed a problem the experimenters knew AI would give a wrong but convincing answer to - many did worse. Another study on the use of high-quality AI by 181 recruiters (Fabrizio Dell'Acqua, 2023) found that users became over-reliant on AI and missed out on potentially qualified applicants.

Researcher Ethan Mollick's article on AI's "Jagged Frontier" looks at both these studies as well as how he uses AI depending on the task at hand. His conclusion thus far is:

"On some tasks AI is immensely powerful, and on others, it fails completely or subtly. And, unless you use AI a lot, you won’t know which is which."  

IBM's 2023 Global Adoption Index reports that the financial services industry are most likely to be using AI - with about 50% of IT professionals in that industry saying hat their company is actively deploying AI, with telecommunications companies not far behind. As AI advances, more skilled professionals are available for deployment and issues such as data security and trust and transparency are addressed AI use by companies will grow accordingly. 

The OECD's March 2024 report on Using AI in the Workplace cites automation and job displacement as a risk that differs from previous technologies because AI automates many non-routine cognitive tasks, all occupations are likely to be impacted and the rapidity of adoption means there is little time for society to adjust. While there is not yet sufficient evidence that AI will have a negative impact on jobs - the OECD report estimates that the jobs most at risk for AI automation account for 27% of total employment.