Artificial intelligence (AI) is currently presented as a solution to all our needs and problems and education is no exception. Large language models such as ChatGPT are already used in schools, often to the dismay of teachers, which prompted OpenAI, the tool’s parent company, to release a guide on how to use generative AI in the classroom.
Companies such as Khan Academy lure customers with individualised tutoring systems (like Khanmigo) that promise high-quality one-on-one teaching on an unprecedented scale. There are also hopes that AI will improve schooling by monitoring students’ mood and engagement, providing real-time feedback to teachers or even assigning grades automatically.
There is no guarantee that such promises are anything more than clever marketing designed to entice schools and governments to invest significant resources into new products. After all, AI does not always work as advertised, with ChatGPT generating false information and making up sources or facial recognition systems routinely misclassifying users belonging to marginalised groups.
However, let us assume that AI does indeed have the potential to revolutionise education by improving students’ performance and giving teachers more opportunities to, well, teach, rather than mark assignments and write reports.
Ceann comhairle election key task as 34th Dáil convenes for first time
Your EV questions answered: Am I better to drive my 13-year-old diesel until it dies than buy a new EV?
Workplace wrangles: Staying on the right side of your HR department, and more labrynthine aspects of employment law
The great trifle revival: ‘Two creamy, delicious things on top of a boozy, fruity, delicious thing’ - what’s not to like?
Not every school is going to equally reap these benefits and Ireland’s dual-language education system introduces unique challenges. We worry that widespread use of educational AI could advantage English-language schools and leave Gaelscoileanna behind.
Contemporary AI models rely on vast amounts of training data. Based on instructions from their developers, they can identify patterns which help them predict the likelihood that, for example, a given image depicts a dog rather than a turtle.
In the case of ChatGPT, its makers needed to use 183,000 books, billions of web pages and numerous other texts before the model was able to formulate coherent sentences; it does so by assessing how likely given words are to follow each other in a particular context.
Due to the relatively low number of native speakers, developers of AI tools operating in Irish do not have access to anywhere near the same level and variety of written source material as is available in English. This is problematic as greater amounts of data allow for the creation of more sophisticated tools.
While the version of ChatGPT launched in November 2022 is estimated to have used about 570 gigabytes of training data, the newest, more advanced GPT-4 model required 10 times as much information. It is incredibly difficult to teach AI a language, because each one, including Irish, involves its own quirks and idiosyncrasies. Non-English versions of tools such as ChatGPT are already more prone to factual and grammatical errors and further advances are likely to exacerbate the differences in the availability of data. Recent research has shown that the web is becoming polluted with poor AI translations, which negatively affects minority languages the most.
Assuming that educational AI will be developed by private companies as currently happens, is there a large enough user base for Irish-language AI that would guarantee satisfactory return on investment and incentivise developers to work on such tools?
Is it possible to satiate the data hunger of Irish-language educational AI tools and maintain parity with their English-language counterparts?
At the same time, we cannot just feed the model any data and hope for the best. Specialised AI tools require specialised training and educational applications need highly contextualised information. If we expect AI to help in teaching subjects such as geography, history or physics and to do so in Irish, we need to first rely on existing materials that would demonstrate the required standards and set proper expectations.
To generate a single new math problem, AI needs to analyse thousands of existing ones. However, the Government has already lamented the low availability of Irish-language subject-specific teaching materials. And while future AI tools might help us close this gap, they cannot create textbooks out of thin air.
Furthermore, assuming that educational AI will be developed by private companies as currently happens, is there a large enough user base for Irish-language AI that would guarantee satisfactory return on investment and incentivise developers to work on such tools?
Recent data indicates that almost 52,000 pupils attend primary schools with at least some subjects taught in Irish. It is doubtful that any profits companies can extract from this cohort will offset the cost associated with the research and engineering, as well as the enormous energy and infrastructural cost of developing and operating Irish-language educational tools. In addition, the decentralised nature of the Irish school system means that not every school need adopt specific tools, which further reduces the potential user base. By contrast, developers of English-language AI systems can target not only Irish pupils, but also their peers in other English-speaking countries, which in turn impacts where companies choose to invest.
All these concerns lead us to worry that without a good plan, Irish-language schools will be faced with an impossible choice: embrace second-rate educational AI or forgo the potential benefits of automation altogether. Of course, general concerns related to AI and digital technologies mean that it might be a good idea to reduce our reliance on them and countries such as Sweden have already expressed similar sentiments. But being mindful of potential ethical and practical issues does not justify throwing the baby with the bathwater.
Countries could band together to develop their own AIs beyond the control of giant for-profit corporations which could better serve local languages. Examples under development from the Netherlands, Sweden and Finland hope to democratise AI by making it more open. We suggest that the Government should devote more attention to developments in educational AI and provide concrete resources to create Irish-language AI systems that would meet the needs of students and educators.
The future of Irish-language schooling requires a proactive stance and a clear strategy on how to ensure that every school, no matter its main language, benefits from the promises of new technologies.
Dr Michał Wieczorek is an IRC Government of Ireland Fellow working on ethics of educational technology and Dr Eamon Costello is an associate professor of digital learning at Dublin City University.
- Listen to our Inside Politics Podcast for the latest analysis and chat
- Sign up for push alerts and have the best news, analysis and comment delivered directly to your phone
- Find The Irish Times on WhatsApp and stay up to date
