Introduction  

What is already known about this topic? 

AI is not a new field   

The technology of Large Language Models (LLMs) underpins the growth of the variant of AI known as generative AI (GenAI)   

What this report adds 

A brief review of AI and GenAI in the field of education  

Implications for practice or policy 

There are significant opportunities to use GenAI in education and to use the technology as a catalyst to rethink assessment design and teaching and learning practice   

In this report we provide a review of how the higher education sector in Australia is addressing the challenge of artificial intelligence related to practices in teaching and learning. To do this, we provide a historical background to modern general purpose artificial intelligence tools (such as DALL-e or ChatGPT), review the use of artificial intelligence in educational contexts and provide some practice guidelines from an Australian context that can provide pointers for the productive application of artificial intelligence in the field of higher education.  

The E arly P ioneers  

The work of Charles Babbage (1791-1871) and Ada Lovelace (1815-1852) has traditionally been associated with the development of machines (conceptual and real) that have been identified as the first modern computers ( Hollings et al. 2018 ). Babbage and Lovelace contributed to the development of the ‘Analytical Engine ’, a conceptual machine which was never developed and Lovelace has been attributed with the design of procedures which have since become part of the lexicon and practice of modern computing such as loops. The work of Babbage and Lovelace was instrumental in providing the foundations for modern computer science ( Swade , 2017).    

History of Artificial Intelligence  

In the early and mid – 20th century, the work of Turing (1912-1954) and Godel (1906-1978) (amongst others) developed the theoretical foundations of modern computer science and Turin g , in particular, identifie d the logical features of a ‘thinking’ machine in his seminal paper (1950) which begins with the question ‘Can machines think?’. To address this question, Turing devises a thought experiment called the ‘Imitation Game’ or what is commonly referred to as the ‘Turing Test’ to define and measure the nature of ‘thinkin g ’ . Text-based questions are posed to a computer and a human respondent and if the questioner is unable to tell the difference between the machine and human responses, the computer, Turing posits, will have passed the test. The move from the description of a logical machine to a working program that simulated problem-solving strategies is represented by Newell and Simon’s Logical Theorist ( Gugerty, 2006 ) and in the same year the term Artificial Intelligence is coined at the Dartmouth conference (Moor, 2006). Participants at the conference included those that would go on to provid e the foundations of the modern academic discipline of AI such as Marvin Minsky, John McCarth y and Shannon Bell. From the 1950s up to the ‘deep learning’ models that underpin much of the current AI systems (ChatGPT, for example), several AI systems were developed. These are outlined below and demonstrat e the range of cognitive attributes associated with human behaviour and practice.  

ELIZA, a  program developed by Joseph Weizenbaum in the 1960s (Weizenbaum, 1966) that simulates how a Rogerian psychologist might interact with a person. The program used matching rules to drive the interaction and is a progenitor of ChatBots. There are several online versions of the program and this is one example .

SHRDLU and Winograd Schemas,   SHRDLU is another program that demonstrates the growing sub-field of natural language processing in AI through the 1960s and 1970s. It is a program written by Terry Winograd that processes simple commands and questions related to a block world. A Winograd schema relates to a linguistic challenge which demonstrates the balance between syntax and real-world knowledge that humans bring to bear in the process of comprehension. The Winograd schema test is seen as an updated version of the Turing Test.

The 1980s through to the early 2000s can be characterised by a symbol processing approach using frames of understanding or expert schemas to simulate reasoning associated in particular domains (expert systems) through to more bottom-up approaches associated with connectionism, neural networks and machine learning. These approaches have now led to the paradigm known as ‘deep learning’ which is an approach that can discover structures of patterns in large data sets (LeCun et al. 2015).   

Artificial Intelligence is not without its critics and perhaps the most well-known of these is Hubert Dreyfus (1929-2017). His is a thoroughgoing critique of the philosophical foundations of the AI project and he has argued that human intelligence is driven as much by subconscious processes (recognising a face, for example) as it is by higher level processes that can be formalised into rules. However, as symbol-processing approaches to AI are being increasingly replaced by ‘deep learning’ approaches, it could be argued that Dreyfus’ concerns are being tested by the latest AI models. One of these is ChatGPT.  

ChatGPT (and its variants) is an example of a large language model (LLM) which uses deep learning as a training mechanism to improve its output based on massive data sets from the internet. More specifically, the program generates the most statistically appropriate responses to questions based on its identification of patterns in large data sets (Rospigliosi, 2023). As responses from ChatGPT become increasingly more plausible and valid, issues related to academic integrity are likely to become more complex but the possibilities for the application of ChatGPT in a range of teaching and learning practices are also likely to increase. The increase in the application of LLMs, however, also means that it is incumbent on the community to sharpen its understanding of the claims that are made about the ‘intelligence’ of these systems. Dron (2023) provides a critical view and argues that the nature of intelligence demonstrated by LLMs is a form of reified knowledge without any underlying meaning. As Dron (p.4) says:

“…an LLM is nothing but signals, with nothing that is signified underneath. The symbols  have no meaning.”

A unified position on the nature of AI, therefore, is unlikely given the different perspectives on the nature of intelligence but it is important to maintain a critical perspective as these technologies become more embedded in our daily lives.

In the following section we review how AI is currently being used in educational contexts and we provide an Australian perspective on the potential for the application of AI in the higher education space.  

AI and Education  

Today, AI can potentially offer new opportunities in higher education, including, personalisation of online learning and innovation in learning and teaching practices. However, rapid technological developments and the rise of generative AI, particularly Chat GPT, can pose challenges that require changes in policies, assessment design and the consideration of the ethical implications of the technology. Various organisations have been established to explore the potential of AI technologies within education settings, guided by principles of efficiency, access, scale and inclusion.  

The National AI Institute for Adult Learning and Online Education (AI-ALOE) in the United States is developing an AI-based transformative model for online adult learning. Grounded in theories of human cognition and learning, AI-ALOE’s goal is to enhance the application of AI to make adult education more accessible and equitable. AI-ALOE is currently exploring the opportunities afforded by ChatGPT. For instance, they promote the safe and responsible use of ChatGPT as a tool through examples that address common challenges in online learning environments such as increasing levels of teacher and social presence. These are two examples of their current projects.    

JILL WATSON , the world’s first virtual teaching assistant (Design and Intelligence Lab, Georgia Institute of Technology) was developed in 2016, with the aim of amplifying the teacher’s presence by answering students’ questions anytime and anywhere. Since its development, Jill Watson, has been successfully deployed in 40 online classes, servicing more than 12,000 students. Harnessing the power of ChatGPT, a new Jill (Jill+) is being developed to ingest class content and materials with the aim of answering questions from students more efficiently and reliably.   

SAMI, the virtual social agent (Design and Intelligence Lab, Georgia Institute of Technology) was created to enhance and create opportunities for social interactions and connectedness in online learning environments. Using natural language processing, SAMI analyses learners’ posts to construct a profile for each learner and then recommends social connections based on commonalities. Since its deployment, SAMI has been used in more than 30 online classes with more than 8,000 students. There are plans to outsource SAMI’s natural language processes to ChatGPT in SAMI+ to enable more meaningful matches.  

Internationally, the United Nations Educational, Scientific and Cultural Organisation (UNESCO) has led efforts to deliver global standards to maximise the benefits of new technologies and innovations, such as AI, while minimising the risks, and ensuring their contributions are inclusive, sustainable and lead to peaceful outcomes. UNESCO’s work highlights the importance of a human-centred approach such as ensuring that systems are in place to ensure human oversight of AI systems. Within an educational context, UNESCO has also developed a framework to foster the readiness of education communities in their use of AI. Their publication on Artificial Intelligence and Education: Guidance for Policy-makers looks at the opportunities and challenges of AI in education.  

In Australia, the Australasian Council on Open Distance and eLearning (ACODE) plays a key role in enhancing policy and practice in higher education around technology enhanced learning and teaching. One of ACODE’s key objectives is to disseminate and share knowledge and expertise across institutional, national and international levels (ACODE, 2023). In 2023, ACODE held a national workshop and series of roundtables for the tertiary sector to promote discussion on the challenges, strategies and opportunities of the use of AI in learning and teaching, focusing on academic integrity and productivity. These areas were identified as key areas of concerns from a pre-workshop survey. As a result of the roundtable discussions, the following items were identified as key recommendations for practice (Venaruzzo et al., 2023):  

1. Embrace AI in learning, teaching, and assessment, but consider potential risks and challenges that come with it, such as academic integrity concerns and workload issues.   
2. Foster a culture of transparency, collaboration, and partnership between educators, students, and AI experts, to ensure that AI is used ethically and effectively.   
3. Develop evidence-based support systems and guidelines for AI use in education, and regularly update them to keep up with the latest developments and challenges.   
4. Identify and provide appropriate training and professional development opportunities for educators to build their AI competencies, confidence and fluency.   
5. Consider the potential impact of AI on equity and accessibility and ensure that AI solutions are designed to benefit all students, regardless of their background or circumstances.   
6. Collaborate with external bodies, such as accrediting bodies and regulatory agencies, to align educational responses to AI across primary, secondary, and tertiary education sectors.   
7. Continuously monitor and evaluate the impact of AI on learning, teaching, and assessment, and be open to making necessary adjustments based on the evidence.   
8. Institutions prioritise assessment redesign, by adopting more authentic forms of assessment to minimise the option for students to use AI based tools in generating assessment content.  

The groundwork for modern computing began in the 1800s, with the work of Babbage and Lovelace and pioneering work continued in the field most notably associated with the work of Alan Turing.  During the 1950s, the term, ‘Artificial Intelligence’ was established at the Dartmouth conference and some of the early concepts and concerns about AI continue to inform current practice and the direction of the field. Within educational settings, the technology of AI is rapidly evolving and continues to challenge institutions with the ways in which it can be applied in the area of learning and teaching and this is illustrated in the recently published ACODE guidelines. The unfolding narrative of AI and education is likely to be a feature of the field for a long time to come.  

References  

Australasian Council on Open, Distance and e-Learning (ACODE) (2023). What we do. https://www.acode.edu.au    

AI-ALOE (n.d). AI-ALOE Builds on ChatGPT for Innovations in Adult Learning and Online Education . https://aialoe.org/chatgpt/    

Dron, J. The Human Nature of Generative AIs and the Technological Nature of Humanity: Implications for Education.  Preprints   2023 , 2023100433. https://doi.org/10.20944/preprints202310.0433.v1

Gugerty, L. (2006). Newell and Simon’s Logic Theorist: Historical Background and Impact on Cognitive Modeling. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 50(9), 880–884. https://doi.org/10.1177/154193120605000904  

Hollings, C., Martin, U., & Rice, A. (2018, July 26). Ada Lovelace and the Analytical Engine. Ada Lovelace: Celebrating 200 Years of a Computer Visionary. https://blogs.bodleian.ox.ac.uk/adalovelace/2018/07/26/ada-lovelace-and-the-analytical-engine/  

LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436–444 (2015). https://doi.org/10.1038/nature14539  

Moor, J. (2006). The Dartmouth College artificial intelligence conference: The next fifty years. AI Magazine, 27(4), 87–91.  

Rospigliosi, Pericles ‘asher’ (2023) Artificial intelligence in teaching and learning: what questions should we ask of ChatGPT?, Interactive Learning Environments, 31:1, 1-3, DOI: 10.1080/10494820.2023.2180191  

Swade, Doron, ‘Turing, Lovelace, and Babbage’, The Turing Guide (Oxford, 2017; online edn, Oxford Academic, 12 Nov. 2020), https://doi.org/10.1093/oso/9780198747826.003.0033  

Turing A. (1950). Computing machinery and intelligence, Mind, 59(236), 433–460.  

Fengchun, M., Holmes, W., Huang, R. & Zhang, H. (2021). AI and education: guidance for policy-makers. UNESCO, https://unesdoc.unesco.org/ark:/48223/pf0000376709    

United Nations Educational, Scientific and Cultural Organisation (UNESCO) (2023). Artificial Intelligence. https://www.unesco.org/en/artificial-intelligence    

Venaruzzo, L., Ames, K., Leichtweis, S (2023). Embracing AI for student and staff productivity An ACODE Whitepaper based on the ACODE 88 Workshop and Roundtables Australasian. ACODE. https://www.acode.edu.au/pluginfile.php/13426/mod_resource/content/5/ACODE88-Whitepaper.pdf  

Wiener, N., Cybernetics or Control and Communications in the Animal and the Machine, Actualités Sci. Ind. No. 1053; Hermann et Cie, Paris; MIT Press, Cambridge, MA and John Wiley & Sons, New York, NY, 1948