This isn’t a technical view artificial intelligence (AI). I’m exploring the opportunities and challenges of AI in solving real problems, user-centred design, and particularly new considerations for organisations providing services in Welsh.
First, I believe the first thing we should all do is ignore the hype and exaggeration.
The Jevons Paradox suggests that AI will create more work rather than eliminate roles. The recent history of translation is an example of this; through machine translation, translators have been using aspects of AI for years to translate more and more, helping organisations comply with Welsh Language Standards.
Now that AI technology is more widely available, it’s easy to see opportunities to make “savings” by fully automating elements of work. But surely future work won’t disappear—it’ll just change into different kinds of work: reasoning, human, thoughtful work.
Embedding or marginalising
There’s a (small but real) danger that Welsh could be marginalised in the process of using AI to become more efficient organisations or companies. It’s easy to automate things you don’t understand—other people’s roles, invisible but important work, provision for minorities and the disadvantaged…
What are the implications for Welsh and its everyday users? Less time translating? More time considering the nature of Welsh provision and the appropriate methods to maintain it?
Organisations are tempted to find easy answers in a fast-moving world.
We need to embed the Welsh language and consider how we use AI to deliver bilingual public services. But we also need to think about the people using our services and the implications of AI on their lives.
Roles of the Commissioner and Welsh speakers
In the future, the role of the Welsh Language Commissioner and Welsh speakers within organisations should involve helping everyone understand that language is not just a medium of communication (or words), but a Welsh expression of ideas and perceptions of the world we live in. Only through this can we deliver effective services.
AI can translate words from one language to another, but it cannot understand cultures, motivations, and challenges across linguistic, economic, and social boundaries.
One of the strategic aims of the Commissioner is to promote the Welsh language in the workplace. Although the intentions of many organisations are commendable, there is a risk that Welsh will be given no more than a tokenistic presence; we can see and hear more Welsh but fundamentally, nothing changes. But what if promoting Welsh meant moving or elevating our Welsh speakers into roles that are crucial for delivering good bilingual services—roles that embed Welsh into organisational processes from within and change the culture in order to provide better services?
If we’re designing services for users—the people of Wales—then we need to think of technology as a way to facilitate that goal. We musn’t generalise the population or exclude entire groups from our understanding of Welsh society.
Designing for an English-speaking audience and then translating the interface into Welsh is neither sufficient nor appropriate. We need to understand the unique Welsh context and design for it.
Efficiency is not the (only) goal for service providers, despite the never-ending financial challenges, but effectiveness of services for everyone who uses them. Without that, there’s no sustainability or long-term benefit for the organisation either.
Accessing information with GenAI
Will people in the future get their information mainly through interfaces like ChatGPT? Many already do. What are the implications for Welsh?
As content designer Nia Campbell noted, we’re in a new era where people don’t search, they chat. They don’t browse, they delegate.
What are the implications for Welsh and compliance with the Welsh language Standards?
- Will there be a Welsh interface for platforms like ChatGPT? Will it be fully usable in Welsh?
- Will Welsh-language sources be needed, or can ChatGPT translate sources on the fly and provide a service in the user’s language (whichever that may be)?
- What do we as Welsh speakers demand or desire? What will be an acceptable service for us in the future?
- Will we rely on huge international companies to provide for us, or do we need to explore alternative models?
Content and user-centred design
To paraphrase Liberty-Belle Howard’s reflections: AI, in the form of chatbots and so on, is often seen as a technical project. But that’s where the gap lies. The content that underpins them is crucial.
In Wales, we have information in both Welsh and English. How do AI models interact with that information? In this regard, we need to reconsider the nature of bilingual information within AI-supported services.
I liked how Liberty-Belle framed insights in 3 areas. Here they are, adapted slightly for this context:
Interface level:
- With chatbots – AI is both the interface and the background. We need to design prompts, meta-prompts, and error messages.
- Can something like ChatGPT be a Welsh interface? Or could it translate information in real time, or does it need bilingual information from the start?
Service level:
- Source information must not be more incorrect or inconsistent in one language than another.
- What if AI can’t answer a question – where does the user journey go? Can that journey be taken in Welsh? Who is the real person answering the phone call at the end of the digital journey?
- We need to understand user complexity and motivation in every language.
- We need to understand the risks of using AI.
- How can we ensure AI doesn’t introduce or increase bias, especially favouring majority languages at the expense of Welsh?
Organisational level:
- We need to prepare content for AI that is complete, accurate, and clear.
- We need to prepare content for both people and machines. How will it be presented, summarised, cited, paraphrased, and referenced?
- We need clear content governance processes to ensure outdated or irrelevant information isn’t accessed and presented to people. That’s hard in one language, but harder with bilingual content.
- Design is more than interfaces and content—it’s about organising people, facilitating conversations, and mutual understanding so we can solve problems for users and organisations.
The future
Here are some ideas for steps that need to be taken. This isn’t just about developing technology. It’s about steering toward a future that’s desirable for us as a linguistic community with the help of new technology. Only people can do that. And only we can do that on behalf of the Welsh language.
How about these proposals?
- We need a register of AI used in the public sector. We need to consider and understand the impact of each use on Welsh, as well as accessibility, the environment, and equality.
- Welsh speakers must be involved in developing and testing AI use in the public sector to ensure it doesn’t negatively affect people in Wales or their rights to Welsh-language services.
- We need to consider outcomes for people in Wales, not just bilingual outputs and organisational efficiency.
- We need to encourage and allow Welsh speakers providing services to redefine our roles and be more than just facilitators that help organisations to comply with Standards at a basic level.
- Non-Welsh speakers who own services need to stop treating Welsh provision as a box to tick and recognise the importance and invaluable value of Welsh users’ and staff’s cultural perspectives and understanding.
- We need to seize huge opportunities to expand Welsh provision for people with other needs: people using transcribers, screen readers, and subtitles.
- We need to continually consider, as technology changes, what it means to provide Welsh-language services and what compliance with Standards means in that context.
- We need more investment to develop the technology and the teams that create and govern it. This includes not only large models but also smaller, local models that solve specific problems, provide local and national context and help reduce complexity for users.
- We need to collaborate with UK Government teams to take advantage of opportunities to include the Welsh language in new projects.
- We need to be open. We can’t guarantee Welsh availability on AI platforms in the hands of huge international organisations.
If we care about the future of the Welsh language, we must engage with the field, the opportunities, and the implications. If we don’t, others will decide for us.