Background info: The interview is following up the investigation on the AI potential contribution to the education sciences, described in the article „Ariadna Experiment”, issued in December 2022*.
Our interviewee was Dr. Olimpius Istrate, the lead scientist of the research team undertaking the Ariadna experiment. He is associate professor at the University of Bucharest, Faculty of Psychology and Education Sciences, giving courses on virtual learning environments, computer-assisted instruction, new media in education and training. As the president of the Institute for Education foundation (Romania) since 2009, he is initiating development research and evaluation programmes aiming to improve education.
In short, what is the Ariadna experiment and how it relates Artificial Intelligence to education?
Dr. Olimpius Istrate: In December 2022, as everybody else, we were of course intrigued by the suddenly revealed new potential of AI. Among many other things of which we were aware since 2021 – personalisation and support in learning, feedback and assessment, progress monitoring, classroom management, cybersecurity and safety, and so on – we started to wonder if it could actually help in reshaping the theoretical framework of social sciences as we know it, and in particular the education science.
For example, what if a new, simple yet efficient theory of learning would be revealed by AI, based on a combination of selected elements from neuroscience, cognitive constructivism, today’s (digital) social habits, and who knows what other pieces of the learning puzzle that we couldn’t put together up to now? What would be needed to trigger this new theory? Is it about a smart request workflow that a scientist is formulating in an AI interface? Is it about the AI’s training datasets? Or is it about the inner construction of AI applications (to which we would need to contribute)? If we could co-design a large-language model to better replicate creativity and reasoning, what would be the input elements?
Ariadna experiment is a combination of action-research and longitudinal evaluation, an attempt to get answers to such questions, knowing that AI is at the beginning.
In the first stage of our investigation, we have generated and analysed a significant number of texts approaching theory and practice of digital education. We are aiming to estimate the ability of AI to produce scientific innovative content, to monitor the progress in this matter over a long period of time, to understand the changes and the effects on theoretical knowledge and on research in education/ social sciences, but also on teaching professionals, students, education process, education management.
What differences between the 2022 texts and 2023 texts do the researchers expect?
There are two main directions to follow within our research. One is the level of creativity that AI is proving, and we are looking to find new scientific content that could actually be significant for the development of the education science domain. Combinational and exploratory creativity are already there – the aspects that astonished the world several months ago. But when it comes to transformational creativity, the innovative dimension that we are mostly looking for, then we have another story. If any glimpse of transformational creativity, it is indistinctive, random, amorphous, heavily dependant of the human interlocutor and their interpretation. We hope to see a big difference in the second iteration of our research, at the end of this year.
The other expectation is related to the capacity to ”think”; aware that the AI’s ”thinking” is (for now?) a mimetic function, we were trying to discover valuable judgement in our samples. Is AI able to imitate complex reasoning? To what extent this is bringing value to scientists working at conceptual level? Beyond deductive, abductive and common sense reasoning, we are seeking for signs of stable, valid inductive reasoning, of monotonic and non-monotonic reasoning, of critical thinking, of solid decompositional reasoning. This would be a strong basis for link-based hypothesis testing, for example, helping scientists to meaningfully process huge amount of data and establish novel relationships between ideas, new approaches, valid solutions.
At the end of this year, we certainly hope to see a leap in AI’s ”creativity” and ”reasoning”, and there are good signs in both these matters. Furthermore, along with this approach, we better understand what we do every day and what is the worth of our scholar work. We would know what to request from a good AI program, and, most of all, how to train the future generations of scientists.
Do the researchers have strict definitions for the terms „quality, relevance, novelty and usefulness”?
Not at this time, no. Whilst we do intend to preserve a canonical framework for knowledge building, we were quite loose regarding the outcomes of this investigation. We are open to identify, recognise, record, interpret any authentic sign of scientific content that is ”valid” science, being it relevant, of quality, new, useful, opportune, or an indefinite combination of any. Anyway, I think a permissive approach is required in this matter.
What hypotheses are formulated around ‘trusting AI’?
A possible answer to your question is that the trust in AI relates to the level of control needed on behalf of a human being, and right now AI tools need a lot of attention. AI is at an early stage of a long process towards reaching a satisfactory level of artificial general intelligence (AGI) – necessary, in our opinion, for transformational creativity and for reasoning – consequently, for a significant contribution in the fields of social sciences. We are not there yet, not even close.
In our study, the issue is not necessarily around trusting AI to independently (re)build science, but rather around trusting AI to bring forth valid and (mostly) constructive, helpful, and prolific hypotheses, approaches, tools, solutions or conclusions. We are not there yet either.
Do the researchers have connections to AI21/OpenAI?
No, this is not the case. We don’t have connections with any of the many existing AI tools, but we do hope to have at some point a contribution to the way they are built, trained, conditioned.
Are students involved in this work/are you planning to share output with students?
This is a good question and a good point to take into consideration. Recently, we were thinking to add the student’s perspective. On one hand, to onboard them in the research team, with specific tasks related to their role as learners. On the other hand, to use some of students’ work that resembles much with AI’s output, and we already have a small database with essays and articles –there are yet some ethical issues to solve here. But we will surely have a mixed team, starting with the next academic year.
What implications do you anticipate for how universities test their students’ knowledge & comprehension (for instance, via essays that could easily be generated via an AI-tool)?
The second stage of the evaluation research will run in November-December 2023, appraising and comparing AI’s scientific outputs, but also adding new components and questions to the initial set. We already have some preliminary conclusions, as the first investigation stage was unexpectedly prolific, transforming the approach into exploratory research.
Regarding your question, I would remind that the education systems – and higher education institutions in particular – have always been looking for ways to innovate, to improve, to better prepare the graduates, to use the learning time more efficient… Far more than just remember, comprehend, analyse and so on, today we want our students to be able to solve complex problems, to think critically and creatively, to continuously learn and adapt. Aims such as design thinking, conjecture reasoning, divergent thinking, social and communication skills should be at the core of today’s study programmes, no matter the field of specialisation, and AI is part of the equation, nowadays and also tomorrow. AI can help teaching staff to design, develop and evaluate education situations, AI can help with the administrative matters, AI can help students learn better. AI is already in the landscape of almost any occupational field, therefore we are obliged to prepare our students to be aware of its potential and limits, and to make the most of it. In short, AI comes with solutions and challenges, but mostly with the opportunity to rethink teaching, learning and assessment in universities – educational goals, content, methodology, context of learning, evaluation of the opportunity, efficiency, and added-value of the HE study paths.
As concerns the contribution of AI to the development of social sciences disciplines, the preliminary interpretation was clear. In these early stages of AI, we might consider the generated texts rather incomplete reports, school essays, integrated reading notes, or work-in-progress materials. They do not have sufficient theoretical foundation, ideational force, generative value. However, one conclusion is that the AI tools that generate (scientific) text are, to a significant extent, similar to what many scholars are regularly doing when they elaborate articles and deliver lectures – to a certain extent, this is ”recycling language” – and the main lesson that AI is teaching us is to stop writing papers if we don’t have anything new to say; furthermore, to stop asking students to write ”artificial” essays, pointless activities within their learning paths. We should be aware of our own potential and limits, and do our best to be authentic creators and authentic professors in this new context. Perhaps in the academic world the ”confrontation” with AI is like ”the mirror test” in psychology – we should see ourselves in the AI’s reflection, and try to become – and to educate! – better humans, better artists, better professionals, more creative scientists.
* Istrate, O., Velea, S., Ștefănescu, D. (2022). Ariadna Experiment. The Role of Artificial Intelligence in Education Sciences. Journal of Digital Pedagogy, 1/2022. Bucharest: Institute for Education. https://digital-pedagogy.eu/ariadna-experiment-the-role-of-artificial-intelligence-in-education-sciences/