Joint Committee on Education, Further and Higher Education, Research, Innovation and Science debate -
Tuesday, 9 Jul 2024

Apologies have been received from Deputy Sorca Clare and Senators Aisling Dolan and Fiona O'Loughlin. Senator Malcolm Byrne is substituting for Senator O'Loughlin.

I remind members to please ensure their mobile phones are switched off for the duration of the meeting as they interfere with the broadcasting equipment of the Houses, even if they are on silent mode. Are the minutes of the meeting of 3 July agreed? Agreed.

We are meeting this morning with the BT Young Scientist winner and other entrants on AI projects. On behalf of the committee I welcome Ms Eugenie Kelleher, St. Mary’s College, Macroom, County Cork; Ms Layla-Grace White and Ms Leah Jennings, Ballinteer Community School, Dublin 16; Mr. Padráig Meade, Scoil Pól, Kilfinane, Count Limerick; and Mr. Seán O'Sullivan, Coláiste Chiaráin, Croom, County Limerick. The witnesses are here to discuss the future of artificial intelligence in post-primary education.

Finally, I congratulate Seán on winning the accolade of the 2024 BT Young Scientist of the Year. I also welcome the parents, siblings and teachers of the young scientists, and former committee witnesses, who are their teachers, and officials from the Department of Education to the Public Gallery.

The format of the meeting will be that I will invite the witnesses to make brief opening statements lasting five minutes and there will be a timer on the screens. Ms Kelleher, Ms Jennings and Ms White will make a joint statement. They will be followed by Mr. Meade and Mr. O'Sullivan. The opening statements will be followed by questions from the committee members. Each member has a specific slot to ask questions and for the witnesses to respond. As the witnesses are probably aware, the committee will publish the opening statements on its website following the meeting.

Before we begin, I remind members of the long-standing parliamentary practice to the effect that they should not comment on, criticise or make charges against a person or entity outside the House or an official either by name or in such a way as to make him, her or it identifiable.

Witnesses are reminded of the long-standing parliamentary practice to the effect that they should not criticise or make charges against any person or entity by name or in such a way as to make him, her or it identifiable or otherwise engage in speech that might be regarded as damaging to the good name of the person or entity. Therefore, if their statements are potentially defamatory with regard to an identifiable person or entity, they will be directed to discontinue their remarks. It is imperative they comply with any such direction.

Ms Kelleher may begin. She will be followed by the other witnesses.

I thank the Chairperson and committee for the invitation to discuss the integration of AI in post-primary education. I also appreciate the opportunity to present my findings and insights from my research project titled "An investigation into the effect of Artificial Intelligence Aids on learning achievements, creativity, and innovative thinking in Irish secondary school students", which I conducted for the BT Young Scientist and Technology Exhibition 2024.

In today's world, it seems as though AI is advancing at an exponential rate. Its potential to transform various aspects of our lives is undeniable. However, with the inevitability of AI's integration into education, given its rapid development, it is crucial that this integration is done in a gradual and thoughtful way in order to harness its benefits while also effectively addressing the challenges it presents. Throughout the duration of my research, I have developed a comprehensive understanding of the role that AI can play in enhancing educational experiences.

My project primarily focused on generative AI tools such as ChatGPT and Snapchat AI and their effect on learning achievements, creativity and innovative thinking among post-primary students. Students do not all learn in the same way and an important aspect of AI, which we must consider for its integration into post-primary education, is the fact that it can be utilised to accommodate students' individual learning methods. For example, if we consider a student who is eager to engage in class discussions but finds their methods of learning do not align with the teacher's methods of teaching, this is not, of course, the fault of the student or teacher but, with the use of AI software, the student can be presented with the lesson content in a way that resonates with their learning needs, such as visualisations or personalised explanations tailored to their unique style of learning. This demonstrates how AI software can be utilised to enhance student learning in the post-primary educational landscape.

Through my research, I observed that many students are already incorporating AI tools into their academic work. One key finding from my questionnaire data was that 36.3% of students reported using AI-powered tools to enhance their creativity in educational settings. Students found that while generative AI tools offer substantial benefits such as providing new ideas, enhancing technical skills and streamlining the creative process, they also raise significant concerns about academic integrity, over-reliance and potential laziness among students. For example, 76.6% of participants in my study acknowledged that AI tools provide new ideas and inspiration, demonstrating their positive impact on creativity. However, 88.2% of participants expressed concerns about AI's role in fostering creativity and innovation, worrying about its potential to promote dishonesty and hinder genuine creative thinking.

Additionally, my research revealed significant gender-based differences in attitudes to AI use. Male students, for instance, were more likely to use generative AI tools for English assignments, indicating a potential gender imbalance in adopting AI tools. Furthermore, the recent leaving certificate reforms, which emphasise project work and practical assessments, highlight the importance of addressing AI's role in education. As these new assessment methods become more prevalent, the risk of academic dishonesty through AI use increases. Ensuring the authenticity of students' work in this context is crucial to maintaining the integrity of our education system.

To effectively integrate AI into post-primary education and address these concerns, I propose the following recommendations. First, clear guidelines and policies should be developed for the use of AI tools in education to protect academic integrity. Second, training programmes for educators should be implemented to help them to understand and manage the use of AI in their classrooms. Third, we should ensure all students have equal access to AI resources and are encouraged to use them responsibility. Ralph Ellison once said, "Education is all a matter of building bridges." As we integrate AI into our education system, we must build bridges carefully, ensuring they lead to enhanced learning, creativity and integrity for all students.

I thank members for their time and consideration. I look forward to discussing these issues further and answering any questions they may have.

First, we would like to thank the committee for inviting us to discuss our project at this year's BT Young Scientist and Technology Exhibition regarding the future of artificial intelligence in post-primary education. Leah Jennings and I have just finished transition year in Ballinteer Community School, County Dublin.

Our teachers were asked to identify which essays they believed were generated by AI using basic prompts. We tested this with written work from multiple subjects, including English, Irish, French, history and geography. Our results were mixed and the quality of essays generated varied.

However, this could be improved with more specific instructions to the AI.

Having conducted this study, it is our belief that as AI large language models continue to evolve at a rapid pace, this is not something we can ignore in the context of education in Ireland. It may be necessary to come up with a nationwide plan for how we decide to use artificial intelligence as a tool for learning so students gain a greater understanding of its uses as it becomes an important tool in more and more careers.

As we move towards a new leaving certificate with increased continuous assessment in more subjects, the question is whether there is a need to verify submissions from students to rule out the use of AI. We believe, as students, that we would like to be certain we were completing projects that fairly reflected the work we had put into them.

I thank the committee for the opportunity to speak on this topic. My project at the BT Young Scientist competition was entitled “An investigation into the uses of artificial intelligence in the modern world and the benefits or problems it might cause”. The project aimed to see how AI is used in day-to-day life, including how it is used in school. The way I was able to tackle this question was by conducting an anonymous survey across the school from all year groups and seeing how students in the school used AI, as well as the potential problems or benefits they believed may arise from using and developing AI in the future.

I was able to get 120 responses to the surveys. My first question was: "Can you explain, in your own words, what artificial intelligence is?" The results from this question were positive overall as almost all of the students shared a common and simple understanding of what AI is. A sample of some of the responses were: “The ability of computers and technology being able to do tasks humans can do”; “Machines that are made to think like humans”; and “The ability of computers and technology being able to do tasks humans can do.” One statistic that did surprise me was that the most common term used in responses was robot. I think this makes sense, however, as AI is typically associated with robots all of the time.

The next question was: “Have you heard of apps such as ChatGPT before?” The results of this question did not surprise me as 95 of the 120 responses, or just shy of 80%, were that they had heard of ChatGPT before. This gave me useful information because one year after the application launched, the majority of people have heard of it.

The next question was: “If selected yes, where have you heard of these apps?” Out of all the responses that were given, social media, and more specifically TikTok, was the most common answer. This might already answer my question as to whether AI has got more media coverage in recent months, as a quick look at Google Trends shows us that the search term “AI” skyrocketed on its website in late 2022 and early 2023. Other popular answers included "friends" or "family".

The next question was: “Do you use any AI websites to help with homework?” The results of this did catch my attention due to how even it was, with an almost perfect 60-60 split. I was expecting the survey to be much more one-sided in terms of the people who do not use AI to help with homework but having a near-even split of the results indicates to me that AI is much more common nowadays than, say, in the last five years.

The next question was: “If selected yes, which subject would you use AI to help you with the most and which AI programme?” Out of all the responses, there were two very clear results to be seen: first, languages such as English, Irish and Spanish were by far the most common subjects where AI was used; and, second, the most popular forms of AI used were translation websites, such as Google Translate and ChatGPT. This made me understand it is more than likely that students were using these AI programmes to help with either translating words from one language to another or essay writing for English.

The next question was: “How does AI help you complete this homework?” My previous point was proven by the responses to this question as the most common answer was to translate or to get answers for the questions. This makes a lot of sense as Google Translate will give a near-perfect translation immediately, whenever asked, and ChatGPT has been known to be able to give detailed answers, whenever asked. The use of these apps is probably so common due to how convenient they are and how quick to give answers, given many people would rather time efficiency when answering a question and not spend too long on a question.

The next question was: “In your opinion, is Artificial Intelligence a good or a bad thing? Explain.” This question was by far the most interesting one answered, in my opinion, as I got to see how my peers felt about AI in modern life. Overall, the results were very similar to how the media portrays AI in life. I believe the following response best sums up people's opinions:

I think that artificial intelligence can be the greatest or worst invention by mankind. I think that if it is controlled and regulated properly it could change all of our lives for the better in an unimaginable way. It has an infinite number of applications in any sector in the world and can infinitely benefit us in our everyday life. I think the positives far outweigh the negatives.

From the results of the survey, I was able to see that a lot of my fellow students had used AI in the past and that the most popular subjects that people used AI for were either languages or more essay-oriented subjects.

From the results of the survey, I was able to see that a lot of my fellow students had used AI in the past and that the most popular subjects that people used AI for were either languages or the more essay-oriented subjects. There was also great awareness shown for how AI has both positive and negative aspects with the clear, most prominent negative being the idea of becoming overly reliant on AI and, thus, not learning. If I was to recommend how AI should be used for education for the future, apps like Duolingo are a great place to start, as the AI can personalise the learning experience for students across all levels and make learning a language easier and less daunting. I also believe that apps like ChatGPT can be utilised efficiently, if the app is seen as a learning tool and not as the complete answer.

Good afternoon, a Chathaoirligh, and esteemed members of the committee. I am a 17-year-old student attending Coláiste Chiaráin secondary school in Croom, County Limerick. I am honoured to present my project, "VerifyMe: A new approach to authorship attribution in the post-ChatGPT era," which I presented at the BT Young Scientist and Technology Exhibition 2024 at which I was fortunate to be named the BT Young Scientist and Technologist 2024.

The rapid advancements in AI, particularly in natural language processing, have significantly blurred the lines between human- and AI-generated texts. This challenge became the focus of my project, as conversations with my teachers revealed their struggle to distinguish between student-written work and text generated by AI tools like ChatGPT. They often claimed they could recognise if a student used AI because the writing did not match their unique writing style or ability. This insight highlighted a central flaw in current AI content detection systems in that they do not verify against the author's individual writing style.

VerifyMe addresses this gap by employing a structured methodology to analyse stylometric "signatures" for authorship verification. Stylometry, the study of linguistic style, leverages an author's distinct word choice, syntax and sentence structure patterns to create a textual "fingerprint". My project extracts these stylistic markers from submitted texts, including lexical, syntactic, readability and vocabulary richness features, and uses an AI model trained to evaluate the similarities between sample texts and the presented piece of writing. Unlike existing AI content detection systems, VerifyMe is not designed to identify AI-generated text. It is an authorship verification system that predicts whether a presented text matches the specific writing style of the given author, regardless of whether the text was written by a friend, plagiarised, AI-generated, or a mixture of both.

The urgency of this issue is underscored by recent developments in AI content obfuscation tools such as Humanizer, which makes AI-generated text undetectable by existing AI content detectors. These tools highlight the need for a robust authorship verification system such as VerifyMe, which can discern true authorship by comparing texts to the unique writing style of the author presenting the work.

Current AI content detectors have been shown to be unreliable, with instances of human-written work being falsely flagged as AI-generated and vice versa. This has caused significant issues in academic settings, as evidenced by cases at the University of California, Davis, where students faced false accusations of cheating. Moreover, the absence of independently audited evaluation data sets or official oversight further questions the reliability of these systems.

As a further example, as of the 20 July 2023, OpenAI decommissioned its AI content detection system, Classifier, due to low accuracy. Upon evaluation on its English text challenge set, Classifier correctly identified - true positives - 26% of AI-written text as likely AI-written, while incorrectly labelling human-written text - false positives - as AI-written 9% of the time. Additionally, it was discovered that Classifier disproportionately negatively affected writers who had learned English as a second language and people whose writing followed a more predictable and algorithmic structure. This research stands in stark contrast to the numerous AI content-detection systems that claim evaluation accuracies upwards of 95%.

VerifyMe demonstrates consistent performance in authorship verification, even against advanced AI systems such as GPT-4, by analysing multiple texts from a given author. This stylometric approach holds promise for maintaining academic integrity in the face of evolving AI technologies. Looking forward, it is crucial to address the challenges posed by rapid AI advancements in education. This includes enhancing AI literacy among teachers and students, addressing resource gaps in schools and promoting ethical AI usage. Additionally, developing innovative assessment methods and fostering industry partnerships can help integrate AI effectively into education while ensuring academic honesty.

In terms of future opportunities, AI-enhanced learning tools can greatly assist in personalising education, providing custom-tailored learning experiences for students. Teachers can benefit from AI by reducing their administrative workload, allowing them to focus more on teaching and student interaction.

Furthermore, AI can be a powerful tool for creative collaboration and ideation, aiding students in developing critical thinking and problem-solving skills. However, we must also consider the ethical implications of AI in education. It is essential to teach students about the responsible use of AI and emphasise the importance of academic honesty and integrity, especially in subjects such as English where it is crucial for students to be able to cultivate their own unique writing style.

I also believe that by collaborating with AI industry leaders and integrating their insights into policy and research we can create further opportunities for the application of AI in education. I recommend that we pilot new approaches to examinations and assessments that address AI-related challenges and initiate research projects aimed at developing authorship verifications systems such as VerifyMe, trained on more extensive datasets. My evaluations found VerifyMe achieved an accuracy of 85% and it was trained using the British academic written English corpus and further fine-tuned using the Reuters_50_50 corpus. For background, the British academic written English corpus consisted of approximately 3,000 samples of student writing. It was collected between 2004 and 2007 in the UK and it consisted of approximately 1,000 unique authors. The Reuters_50_50 corpus I used has a curriculum-learning training approach. It is more specialised and focused on financial writing from the Reuters news agency with 50 authors and 1,000 unique texts. With more extensive datasets, we would be able to train a more capable system in authorship verification, especially as these AI content detection systems continue to advance.

The advancements in AI present both challenges and opportunities for education. By embracing these changes, thoughtfully and responsibly, we can enhance the learning experience while safeguarding the principles of academic integrity. I thank the committee for its time and attention. I look forward to discussing how VerifyMe can contribute to the future of education and academic integrity in our changing world.

Gabhaim buíochas leo as ucht teacht os comhair an choiste. I thank everyone for coming before the committee. I congratulate them for all their work. Getting to present it today is a big day and I am delighted to see everyone was able to bring family members and so on. Anyone who has ever done a project or run for an election knows it takes a family effort, even if it is just putting up with our stress. It is really good to see everyone here today.

This is really interesting. It will probably be a problem more for the witnesses generation than ours. It is their generation which will have to come up with ways to deal with AI. Some of it has been fascinating. I would ask the two women in front of me about getting teachers to look at what was and was not AI generated to see which texts were which. They mentioned some was in Irish, some in English and some was in other languages. I think it was French. Was there a difference in how Irish was perceived? Was it easier or less easy to see it in Irish rather than other languages?

It was the same with French. The French was very advanced. However, we did find that ChatGPT wrote the strongest Irish essay that our teacher said students would not be able to produce whereas the Snapchat AI produced something less advanced than ChatGPT and it was less obvious to the teacher that it was AI.

That is interesting. It is hard enough to correct everybody's homework without also having to think about that. I was interested in AI not yet having the ability to copy one's own speech and how one would write.

Obviously, if you are correcting 20 or 30 sets of homework, it will be really hard to think every time about how different students do their homework, and that takes a lot of effort. My understanding, however, from what the witnesses said is that they believe that is only a matter of time. WhatsApp nearly predicts what you are going to say write next or how you would structure a sentence. The witnesses believe it will be just a matter of time before AI is used more often. Do they mean that if I started using it now to write speeches, the more I used it, the more it would understand me, or do they mean that it will just take time in a broader sense?

I think it will happen broadly as we train these systems. For example, when I was evaluating current AI content-detection systems, I found there was a huge accuracy drop-off between OpenAI's GPT-3 and GPT-4 models, simply because the latter was more capable. We can say these models are just predicting the next token but, in doing that, they are also being optimised to mimic human writing to a more and more believable level. This was point-in-time research showing that existing systems, even when provided with past examples of a person's writing, are unable to mimic their style perfectly such that an author-verification system can differentiate. We need to explore what it means to use these tools in writing speeches, for example. If there were a version of GPT-4o that had been finely tuned on every speech the Deputy had ever written, there could be a time in the future when it was able to mimic her style exactly, so that raises questions as to the ethics surrounding that. These all need to be discussed.

Mr. O'Sullivan is dead right on that and it is very interesting, even in respect of the ethics. He mentioned the creativity aspect. If someone is tired and is trying to get something written at the last minute, they are not going to be very creative and will try just to get the job done. They could use AI to get ideas and so on but, at the same time, if they end up being over-reliant on it, they are not going to put in the same level of creative effort. Is that something that came strongly back? It was stated that a total of 76.6% of respondents said it had provided new ideas but that more than 80% felt it was hindering their creativity, although a lot of self-awareness is required to realise that. What are the witnesses' thoughts on that?

I found in my study that students found it helpful in providing them with an initial prompt, which they could then use to do the work on their own. That needs to be discussed in the context of what the ethics are in coming up with the original idea and how much of the project comprises the initial brainstorming. The students found they could use the initial prompt and then do most of the work on their own, without AI.

It was stated that AI needs to be seen as a tool for help and assistance, rather than as the primary tool. For school students in particular, how can that balance be struck? Reference was made to education on it. Does that need to start happening even at primary school level to help people understand that it can be used as a tool but that it should not be seen as the only tool?

Yes. If someone were to use AI to help with homework, it should not be used to just copy and paste the entire answer but, rather, if someone were trying to figure out the answer to a question, he or she might see how the AI produced the answer and then explore how it had got to that answer. I think that is the best way in which we could use the likes of ChatGPT for helping with homework. The likes of Duolingo are almost foolproof for helping people to learn a language, given they are seamless and are able to personalise every lesson for all levels.

I welcome all our guests and commend them on the excellent work they have done. Obviously, huge opportunities and risks are associated with the use of AI in post-primary education. I would have thought the opportunities relate, as Ms Kelleher mentioned, to assisting people to learn and assisting teachers to teach people and prompt them in a certain way.

The risks, however, seem to be universally identified with a fear that people will use it to do assessment work or present themselves as having knowledge they do not have. In what way does Ms Kelleher believe there will be opportunities for teachers to teach using AI? What benefits will it have for kids in secondary school?

On the question on teachers using AI to teach, I believe there are more benefits for students using it to learn in their own way. You can prompt AI and input lesson content into ChatGPT or whatever AI software is being used and it can present the information in a different way so you can understand it better. For example, word-based lesson content could be presented visually so it could be understood and learned better.

I completely agree. Some teachers have even started in that regard. If they suspect a piece of work is AI-generated, they make the student sit down and do it in front of them, with no phone, in an exam-type situation. However, AI has positives in that it can aid learning. Even outside the education field and in various workplaces, it has so many positives where problem-solving is concerned. However, with leaving certificate reform AI will have to be kept in mind, especially where there are classroom-based assessments that are essay-style, because it will not be fair. There would be a return to what happened during the Covid pandemic, when there was a higher college dropout rate because students could not sustain the levels indicated by their results.

I do not know whether Mr. O'Sullivan is aware of a report produced by the University of Reading on research based on 33 exam answers submitted by AI. The examiners did not detect any of them. In fact, 32 out of the 33 were much better than the answers of the humans doing the exams. I would have believed it will always be the case that the technology for AI will be one step ahead of detection. Does Mr. O'Sullivan agree?

That was the motivation behind this project. As each new model of our current AI content-detection systems comes out, we almost have to rebuild the binary classifiers we have made. Last year, I created an AI content-detection system trained using the contemporary corpus of American English and samples from GPT-3, yet I found it could be bypassed through paraphrasing and obfuscation techniques, as will continue to be the case as these systems continue to develop. Therefore, it would be very interesting to explore just to what degree AI can mimic an author's writing. From the research, I found that GPT-3 and GPT-4 are not able to mimic it to such a believable level that an authorship verification system cannot make a differentiation. As these technologies continue to develop and to be integrated into every part of our lives and careers, I am not sure what will occur.

I agree that the AI content detection systems are always moving against a moving target. That is why I pivoted it in this project. One’s individual writing style never changes. Well, that is a stretch. It develops over time but you can take markers of that and you can reference those markers. It is an excellent question. I am continuing to develop my project and that is exactly what I will be exploring.

Absolutely. That was another motivation. In secondary school, especially in Ireland, there is quite an intimate relationship between the teacher and the student, especially, for example, in the English classroom. When I was discussing this with my teachers, they said outright in our classes that they would be able to recognise whether the writing was created by us or not. We have our styles and ways of formulating our language. That is not mimicked in a third-level setting. That is where potentially an authorship verification system like this could be applied.

Looking at chess, for example, we have chess bots that can beat every player that has ever lived. However, nobody is as interested in watching that as watching Magnus Carlsen playing a game of chess. Regarding creativity on a broader level, like music and art, I could perhaps see a future where there is a personalised AI experience such that the AI is able to elicit emotions within you that no human would be able to create. However, I think we value the fact that a piece of work was created by a human in the first place.

I thank the witnesses for their work; their family, friends and teachers for their support; and BT for its continued support of this competition. I think any of us who go to BT Young Scientist are often blown away by the range. You think you have heard it all and then suddenly somebody comes up with a question that makes you go, “Oh, I never thought about that.”

I will follow on from what Deputy O’Callaghan was saying, and we were all talking about exams. When we think about exams, in many ways, the leaving certificate is trying to cram large amounts of data into all of our brains so that we all just spew it back out in a series of exams at the end of a five- or six-year period. Given where generative AI is going, do the witnesses think that we will need the leaving certificate in the future? If AI will be able to manage all of that data, how will we test human intelligence? Does anyone have any thoughts on that?

No worries. I am just curious.

Ms Kelleher made a valid recommendation on training programmes in AI for educators. Who should provide those? There is a concern about the level of knowledge of those in the education system around generative AI. Should there be partnerships with the technology companies and AI platforms?

Yes, absolutely. AI or technology providers should provide some general training to teachers around that so they can flag or know how to flag certain ways in which sentences are formed by the AI that is different from how humans would do it in a conversational sense. I found that AI in general does not understand the art of human conversation. Certain words would be repeated in a paragraph whereas a person would use a synonym for that word in general.

That would be one of the ways in which something that is partially AI-generated in some way could be flagged.

AI is pretty similar to the way people dealt with technology and the Internet several years ago. Everyone feared it and said it was a bad thing but now we cannot live without it. Teachers nearly rely on the Internet, PowerPoint and slides to teach. Maybe AI can be introduced in a similar way. It could be a really useful thing in the context of teaching.

There could be a complete over-reliance on AI. If people constantly use ChatGPT, they will not develop the necessary skills for essay writing. In an exam setting, as it has always been hammered into us, you will not have access to your phone on the day. That is one of the main ethical issues, along with general laziness happening to students as well.

Mr. Seán O'Sullivan obviously talked about his project, VerifyMe, and the models used. He was using large data sets. A lot of teachers talk about how they are only allowed to photocopy a certain number of pages out of a document. While AI is a tool which can be used in education, how do you get around some of the copyright or intellectual property concerns around feeding all those data sets in to inform education practices?

Okay, that is good. I wish to ask Mr. O'Sullivan a general question. Does he think educators, and even more widely, policy makers, get it? AI is transforming our lives. It is not something that is happening in the future. Does he think, from his experience, there is a sufficient level of understanding at policy-making level around the impact, and the potential impact, of AI?

There is a prevalent lack of AI understanding among teachers and students inside of our secondary school environment. While I have not done a survey on this, from my own experience, the vast majority of my teachers have not even tried ChatGPT. There is a predominately negative rhetoric around how AI might hinder teachers from their own unique processes of teaching. Training is required. I am not sure whether that burden can be placed on the companies which are developing these frontier models, but it is something that needs to be explored. AI can help teachers with lesson plans. Large language models are incredible for ideation. There has been a lot of research in which used large language models have been used to just generate lists upon lists of ideas. A rules-based system can then go through and decide which of these ideas are actually feasible. For teachers, this could be applied to lesson plans and to personalising the learning experience for each student. There is currently a lot of potential in this area. From my own experience, my teachers just have not been exposed to the technology to explore it.

Does any other witness wish to speak? I might ask Mr. O'Sullivan one final question, on which any of the witnesses are welcome to come in. VerifyMe is obviously a fascinating programme with regard to the possible use, or misuse, of large language models. We are obviously worried in politics about the use of deepfakes and similarly about large language models being misused. Does Mr. O'Sullivan think VerifyMe could have a role in politics? We have an election coming up. Does he want to adapt the app?

I am not sure. As this continues to develop, we need to develop tools for verifying that works have come from an individual author. Deep fakes are continuing to develop at a rate whereby even for those detectors, it is like running against a moving target. With campaigns and deep fakes, a deep fake video of the Senator could be posted on social media and people would not be able to tell. Perhaps for campaigns, people will be more inclined to see the politician in person and go to rallies in person. I am not sure. It will be interesting to see how this plays out in the future.

I read all of the opening statements in great detail. I thank the witnesses for coming in. This meeting feels like a bit of an exam. It is interesting that AI would have been a single topic many years ago but all of the witnesses have come at it from different angles. They have considered the creativity aspect, how to adjust for people's own writing styles and how teachers can tell if students are using AI or not. We are adults in the world and politicians. I never go anywhere without my phone. I am never in a situation in an exam where somebody is testing my knowledge to see how much is in my brain and whether I can get it out. What we in the world are doing is to see how we can use the tools around us to make society the best place it can be. The witnesses have talked about people's different leavening styles, which I had not thought about. That approach has helped in a classroom situation where students have more of a visual style rather than a written style. I had not considered that point.

Do they think it is more the case that education has to adapt to the reality of AI rather than the other way around? From many of their answers, it seems the education system is trying to control the space. The system is stating that it does not want students to do continuous assessment because we cannot trust them not to use AI. When we are in the world, everybody trusts us. LinkedIn asks users if they want something to be rewritten using AI. I have no idea, nor do I care, whether somebody's LinkedIn post is rewritten by AI. A person made the decision to do that in the first place and used a tool. Is it more the case that education has to change and accept reality rather than the students being put in a situation where they are not trusted and cannot go through a more mentorship-style approach with their teachers? That is a broad question but I would love to hear some thoughts on that.

That is a very good point. It is about learning about AI and its use. Most of us do not really understand that. Before I read the submissions, I did not understand most of its implications or what the advances might be in the future. Does anybody have any thoughts on the proposition that it is the education system that might need to change?

It must be really hard for teachers then, if they are doing the same thing. The students were doing one project once. I have done a BT Young Scientist project and it is a huge challenge, but if a teacher is having to change how they are teaching day-in, day-out, that is much harder than maybe the education system looking 20 years down the line and making the changes now. I thank Ms White and Ms Jennings. What do Mr. Meade and Mr. O'Sullivan think?

Teachers and education overall will have to adapt because there is no way AI will just disappear. It should be faced. It should definitely be encouraged in certain scenarios so students know how to use it and get the best benefits out of it without becoming lazy or overreliant on it. The way I have always seen it is that the tools are there in front of us and it is up to whoever to use the tools in the right way or the wrong way. AI can actually develop someone's writing skills so they are better, but at the same time it could also be very detrimental to their learning abilities.

Taking assessments as an example, in-person assessments will have to be weighted more than more continuous ones. Maybe continuous assessment could be done in the form of in-class assessments where the teacher is overseeing because AI is not being used for that. Some of the most capable language models we have are behind paywalls and subscriptions. While they do have free access, it is extremely limited. With Anthropic's Claude 3.5, a paid subscription allows someone to make five times more requests to the model.

Moving forward, AI has incredible opportunities. It is already being used by Moderna, for example. That is completely outside education, but it is accelerating so many aspects of our world already and in education it will incredible for students. If you are doing a homework assignment and do not completely understand what you covered in maths that morning, you could, if your teacher uploads problem sets, upload them and attach them to ChatGPT, Claude or Google's Gemini and they can provide a personalised experience tailored to you to help you understand that content.

When it comes to assessing people it will probably have to go back to in-person. Our education system is a filtering system and these are all hurdles people have to jump over. For example, if you are applying to college abroad there are standardised exams. I think we will see in-person assessments weighted far more heavily and if it is continuous it will still be continuous, but it will be done in person.

Okay. I thank Mr. Meade and Mr. O'Sullivan very much. There are a lot of questions there about the value of assessments and of assessing human beings, but as Mr. O'Sullivan says, part of it is entry to college, which is a really challenging thing to get right. I thank all the students so much. I have really enjoyed the conversation and thank them so much for the work they are doing. I put "Write me a Green Party manifesto" into AI and it came back pretty close to what our manifesto is, but just much shortened.

For me, it was just seeing how fast AI is advancing and seeing in my school environment more and more students turning to use it or trying to test it out for themselves. It occurred to me to wonder what problems could come from this exponential advancement of AI, so I decided to explore that in my project.

Initially, the idea for the project came from the SAG-AFTRA strikes in America, when screenwriters were worried about losing their jobs to the likes of ChatGPT because it was so seamless and easy. They were worried their jobs would be taken from them. I decided to delve into the likes of education when I saw many people questioning what the point of exams was any more when you could easily put a question into ChatGPT and get an answer. That was where I got the idea for delving into AI in education in day-to-day life.

In February 2023, I tried to train my own AI content detector because I found that existing detectors could be bypassed through obfuscation and paraphrasing and other adversarial approaches like that. I found that even after training a model on specific paraphrased AI data, it could be bypassed through instructing the model to mimic an author's style. Even as the model moves from GPT-3 to the more capable GPT-4, and to GPT-4o and all the other frontier models, not just from up in AI, this system was built by taking a step back and asking what are we trying to do when we detect AI-generated content. We are trying to attribute authorship, and that is what I explored in this project.

First, we need enhanced education and training to promote AI literacy across all education, for both students and teachers. We need industry partnerships and to collaborate with AI industry leaders, such as OpenAI, Anthropic, Google, Meta and Microsoft. We need to inform policy, influence research and create further opportunities for AI and its application to post-primary education in Ireland. Last, we need more research into AI integrity in an Irish setting. Many of these projects that are to be seen here today were conducted in individual secondary schools. To assess the usage of large language models, for example, by Irish students when they are creating their homework assignments, we need to pilot research into how prevalent usage of large language models is across the island of Ireland in all secondary schools. This would be really telling in respect of how we would need to inform our assessment methods. Through that pilot, we might find new approaches to examinations and assessments to address AI-related challenges.

If AI is integrated into education in some way, we need to ensure that all students have equal access to AI software. Every student needs to have access to the most recent GPT from OpenAI. The Department of Education needs to supply schools with the necessary software and hardware to ensure that this happens.

If you take a teacher in any secondary school, who could have 35 years of teaching experience, I presume he or she would have huge challenges in grappling with all of this. This is totally new to such teachers. New teachers coming in directly from teacher training are very much up on tech. They have lived and been immersed in the new tech era compared with someone who is 55 or 60 years of age. I find myself to be okay on a mobile phone but the witnesses would be fields ahead of me on AI and everything like that. What are the witnesses' thoughts on those teachers with many years of experience compared with new teachers and do they see that? I do not want the witnesses to comment on any of their individual teachers, although they probably would not mind. I am interested in what they think.

Some of the more experienced teachers would probably be able to tell between authentic work from students. They are used to seeing students' work and what they have to present, over artificial intelligence and how perfect their writing is.

The majority of younger teachers also can recognise AI as they have grown up with it and as they are becoming teachers, the AI is with them. They can tell, especially if they have studied AI. If there was a course for them to be able to tell the difference between students' work and AI, it would make their work easier.

When students come into first year and meet all their teachers, the teachers do not know what their individual writing style is yet or how it has developed over the years. A child going into first year who already is using ChatGPT or Snapchat AI or whatever kind of AI software can take advantage of the fact that his or her teachers do not know him or her or his or her individual writing style and the child essentially can use it without repercussions. The teacher cannot tell if what they have written is in their own writing style.

For a teacher with 30-plus years' experience, a lot could come down to knowing how a student should typically write. That could be a very easy way to recognise where students might be using AI, for example in their wording. Even if it is a first-year student who the teacher does not know anything about, the teacher might be able to use that student’s peers’ writing to judge if the student is using AI or not. Teachers who are younger and are coming up have experience with AI and should be able to identify it more easily, as they have used it a lot.

The subjects where AI is used can be split into where essay writing is a main requirement, such as English or History, and languages such as Spanish, Irish, French or German, for example. For essay writing, AI saves a lot of time and students can either spend 40 minutes trying to do an essay or use ChatGPT and have it finished in five minutes. With translations, they simply think they cannot understand so they copy and paste and put it into Google Translate to have a fair idea where they are going.

Even if students are using AI for their homework assignments, that will be evident in their first midterm or November assessments, when there are discrepancies between their performances in their exams and the quality of their writing in their homework submissions. Again, if we train and educate our teachers in how to respond to these situations, they will be able to identify specific cases where their students are getting assistance with their homework and will be able to talk to them about it. They will be able to question if a student was not feeling as well that day as they have in the past.

To go back to the original question about teachers who in their 20s or earlier are incredibly aware of these generative AI technologies as they continued to develop, compared with somebody who has been teaching for the last 30 years, these systems are designed to be so user-friendly that if a teacher is handed access to ChatGPT and is told he or she can ask this bot anything, such teachers will see the endless opportunities it has for them in the classroom. This also applies to lesson planning and even outside of education. The key is exposure and that can be done by initiatives run by principals in their own schools.

Department-wide, showing the teachers and principals and SNAs of secondary schools what can be done right now with these language models is the best way of showing people their capabilities regardless of whether they have been teaching for the last couple of decades, or they are new to teaching and know exactly what these models can do.

Over recent years, this committee has done a huge amount of work on mental health supports in schools. Do you feel that the overuse of AI could possibly lead to greater exclusion and less collaboration among students in your own peer group? Could this cause more anxiety and additional mental health issues among students? Is there also a danger that AI could be used adversely in terms of cyberbullying? We know this is a serious issue in post-primary schools and in some primary schools as well. The Minister for Education has an idea around students in primary schools not having mobile phones and things like that. Some schools allow students to have a mobile phone on their person during the day while other schools have a different policy in that they are locked away for the day. I am interested to hear the witnesses' views on that.

The development we have been seeing with generative AI in terms of language models could possibly be used for spam, for social media campaigns. Our first real exposure to AI turned against us. I do not quite mean it in that sense, but it could be used to exploit our own psychology using social media. For example, what Norma Foley has been saying about primary school students not needing a phone is important. Our current social media are designed by teams of psychologists and engineers to leverage the most fundamental circuitry in our brains. That is moving in another direction again. Large language models in an educational setting currently could be providing an advantage to students who have access to the more capable models in comparison with the students who might be slightly disadvantaged and do not. I am not sure.

I am just wondering, on another tangent, what the witnesses think doing this project has given them.

How will they take the information they learned forward? Do they think they know more than other people about the subject? They obviously do because they have done a lot of work. Is it going to perhaps inform what work they might do in the future? I am interested to hear about that because they are obviously very passionate. Is it giving them a passion for learning in general or for this subject, and that they might continue forward with it in future?

After the end of the BT Young Scientist competition, one of the judges who interviewed me heavily suggested that I continue along the path of using AI or something along the lines of software developer or engineer because she thought, based the way I was able to do my research and knew so much about the topic, that it could genuinely benefit me in future. I am looking to maybe do a career in software development at the University of Limerick.

I found it interesting that following Senator O'Reilly's question both Mr. Meade and Mr. O'Sullivan were more enthusiastic about careers in software and technology, whereas the girls were not opposed to it but were thinking about different directions. One of the challenges we have is getting more women into technology.

Absolutely. If we look at the numbers who are studying the subject of technology, or even any of the science, technology, engineering or maths subjects for the leaving certificate, the numbers tend to be smaller, which follows on in higher education. Why do the witnesses think that is? If we want to encourage more girls to go into technology, how do they think we can do it? I will let the girls answer before the boys.

Even from a young age, society conditions women and young boys and girls into careers. You see it even with toys and things like that. A mother or father would give a certain toy to a boy and a different toy to a girl. Girls get dolls and boys play with stereotypically male things, such as tools. Society conditions us from a young age to go into different career paths.

Maybe a module about technology could be introduced. For example, in transition year, we have modules such as legal studies.

If there was a module specifically for technology and how it developed in the future, that might tick more women's interests.

To add to that point, my brother goes to an all-boys school and I go to an all-girls school. In the boys' school, tech graph is available and woodwork and metalwork, but they are not available in a girls' school. Home economics is available in a girls' school but it is not available in my brother's school. That needs to change. Woodwork and metalwork should be available in a girls' school and home economics should definitely be available in a boys' school.

Returning to the question about how generative AI use might damage the mental health of young people, OpenAI revealed that in 2023 it had a model called voice engine. Given 15 seconds of any person's voice, it could fine tune the model to mimic their voice perfectly on any given text that was presented to it. I do not think it is planning on releasing it publicly, but from a policymaking standpoint you would almost want to be in direct communication with these companies to understand the exact technology they are developing right now so that you can start forecasting the policies that need to be created to prevent the potential negative effects it could have on the mental health of Irish young people. We must be careful not to leave that decision to these companies. There could be an opportunity to collaborate with these companies to ensure AI, as it continues developing as quickly as it is, will ultimately be for the benefit of humanity.

I agree 100%. It is an issue this committee has discussed over a long time and it keeps coming back to the social media platforms and the bullying that can happen on them. It amazes me that schools do not have a strict policy on the use of mobile phones in their own premises. If I had my way, no student would have a phone in school. Every school principal talks about the great policy they have and how they are on top of this, but if they wanted to make real meaningful change, and it is not as big an issue in primary schools although it is becoming one, then school principals, boards of management and chairs of boards of management should have a very strict policy on the use of mobile phones. I am not sure whether Senator Byrne would agree. In fairness to the Minister for Education, she has tried to bring all the social media platforms to the table but they do not care so long as they sell their platforms and their wares. They can talk all they like but they do not really care about bullying in or out of school, but it is a huge issue. It is great to hear young people's voices saying what they have said. It is up to school principals and boards of management to take a very strict approach on mobile phones. Many things happen in schools that would not happen if mobile phones were not as freely available.

I thank everyone for coming before the committee today to share their knowledge and research findings. Does Senator O'Loughlin wish to come in?

On behalf of all the committee members, I want to say to our witnesses that they are impressive young people. They are a credit to their parents, teachers and schools. They should all be very proud. Primary school teaching is a great career to go into. My daughters want to do the same.

Whether that is because of the holidays or not, I do not know. I hope none of the witnesses were on holidays from which they had to come back for the Oireachtas committee hearing. When I saw that Mr. Meade and Mr. O’Sullivan were from County Limerick, I was surprised they had come to Dublin after Limerick's poor performance over the weekend. I congratulate Ms Kelleher; Cork had a brilliant win. I will not comment on how set they are for the All-Ireland final.

I thank the witnesses. They should be very proud of their achievements. I know that everybody, from their teachers and parents to their schools, is very proud of their achievements. Winning the BT Young Scientist award is no mean feat. It is a huge achievement and they should be very proud of themselves, and I know I speak on behalf of all the meeting participants in that regard.

We will now adjourn until we meet in private session at 12 noon on Tuesday, 24 September 2024, when we will talk about school holidays. I hope an election will not be called between now and 24 September. I thank members for their input to the committee over this session since January of this year. I hope everyone will have a lovely summer.