At present, artificial intelligence (AI) is being depicted as a panacea for all our requirements and challenges, including those present in the education sector. Big language models such as ChatGPT, despite causing consternation amongst teaching staff, are being employed within schools. This has compelled OpenAI, the creator of this tool, to issue a manual on leveraging generative AI in an academic setting.
Various organisations including Khan Academy tempt potential clients with personalised coaching systems (for instance, Khanmigo) which offer unparalleled scale in delivering high-calibre one-on-one instruction. In addition, it is hoped that AI will elevate the education system through monitoring pupils’ engagement and emotional state, giving teachers moment-to-moment feedback, or even assigning grades autonomously.
However, there are no assurances that these ambitious claims are anything beyond ingenious advertising strategies aimed at inducing schools and governments to sink substantial assets into novel products. AI doesn’t always perform the way it’s pitched – for instance, ChatGPT is known for producing incorrect data and fabricating sources, while facial recognition systems have a habit of improperly labelling users from marginalised communities.
Assuming that AI does indeed hold the potential to transform education by enhancing students’ achievement and providing teachers with increased opportunities to focus on teaching rather than grading assignments or writing reports, its benefits would not be evenly distributed across all schools. There are concerns that the extensive deployment of AI in education could favour English-language schools and neglect Gaelic schools in Ireland’s dual-language education approach.
Modern AI models are heavily dependent on large volumes of training data. From their developers’ instructions, these models can identify trends allowing them to predict, for example, if a certain image is more likely to represent a dog than a turtle.
In the case of ChatGPT, to enable the model to generate coherent sentences, its developers had to use resources such as 183,000 books, billions of internet pages and countless other texts. The model then calculates the likelihood of certain words succeeding each other in a specified context.
Drawing from the original text, it can be deduced that AI tools operating in Irish are somewhat handicapped due to the relative shortage of native speakers contributing to a dearth of written source material. In contrast, data availability in English is extensive, promoting the development of more intricate tools. To illustrate, while the November 2022 version of ChatGPT used approximately 570 gigabytes of training data, the more sophisticated GPT-4 model required a volume of data tenfold.
Teaching AI a language is a complex process, as each language, including Irish, possesses its own unique attributes and peculiarities. Non-English versions of tools such as ChatGPT already exhibit a higher propensity for factual and grammatical errors. Further advancements without addressing the data shortfall could accentuate these issues. Recent studies have revealed an influx of poor AI translations web-wide, with minority languages, like Irish, bearing the brunt.
Should AI tools continue to be developed by private companies, the question arises whether there will be a sizeable enough Irish-speaking audience to promise healthy return on investment and galvanise developers to prioritise such tools. Furthermore, can Irish language AI tools be provided with enough data to keep up with their English counterparts, satisfying their data needs?
Yet, feeding the tool data indiscriminately won’t suffice, as AI requires specific training and educational applications need contextual information. Should we envisage AI facilitating teaching subjects such as geography, history, or physics in Irish, there’s a need to first utilise available resources that meet the required standards and set fitting expectations.
To create a new math problem, AI has to scrutinise thousands of existing ones. However, the lack of Irish-language subject-specific teaching materials has been a point of discontent for the government, causing a hindrance. While future AI tools could potentially narrow this gap, they can’t miraculously conjure textbooks.
Once again, it begs the question: if private companies continue to spearhead AI development, will there be ample users for Irish-language AI to warrant a satisfactory return on investment and motivate developers to invent such tools?
Fresh insight reveals that in excess of 52,000 students are enrolled in primary schools where Irish features in the curriculum. The substantial expense involved in research, engineering and the energy and infrastructural costs of generating and implementing Irish-language educational tools raises scepticism about any potential profitability for corporations. Moreover, the diversification within the Irish schooling system means there’s no obligation for each school to adopt specific tools, which minimises the potential market size. Comparatively, creators of English-language AI systems have an increased target audience, encompassing not just Irish students, but also their counterparts in other anglophone countries. This directly influences corporate investment decisions.
These factors evoke concerns that without a suitable strategy, Irish-language schools will be trapped in a difficult predicament: accept subpar educational AI, or entirely sacrifice the possible advantages of automated processes. Clearly, it may be prudent to decrease our dependence on AI and digital technologies due to generalized concerns pertaining to them, a viewpoint already voiced by countries like Sweden. However, being conscious of potential ethical and practical issues shouldn’t warrant the dismissal of all the pros these technologies bring.
There is potential for countries to collaborate on developing their own AIs that can cater to local languages, without being governed by large profit-driven corporations. For instance, current endeavours from the Netherlands, Sweden and Finland aim to democratise AI, through making it more accessible. Futuristic Irish-language schooling warrants a proactive approach and a defined roadmap on how to ensure each school, irrespective of its principal language, reaps the rewards of emerging technologies.
Dr Michał Wieczorek, an IRC Government of Ireland Fellow specialising in the ethics of educational technology, and Dr Eamon Costello, an associate professor of digital learning at Dublin City University, suggest the Government invest more focus and tangible resources in advancing educational AI and generate Irish-language AI systems to fulfill the demands of learners and instructors.