AI-generated image using Bing and the prompt “image of female hands typing on laptop computer showing an image of the bogeyman on the screen, with the words 'AI' on his shirt, cartoon, with red and orange” .
ChatGPT, Bard, “insert obscure large language model here”; if you work in the language industry (why else would you be reading this post?), you are probably a bit sick of hearing about generative AI, but fully aware that we can’t just pretend it doesn’t exist if we want to thrive in the long term. Everybody is talking about it, and in Mantua, at METM23, so were we.
Of course, that wasn’t my first encounter with these tools. When the hype started, I tested ChatGPT in the silly little ways we all did at the beginning and watched a few webinars by people who are far more enthusiastic about the whole thing than I’m ever likely to be. Some suggested using it for outlining copy, others to use it as a qualified first draft, and so many propose it as a terminology research tool. I’ve tried those applications, and I found the output to not be good enough yet when compared to my current systems. So, currently I basically just use it for creating images for my blog posts.
METM23’s programme featured a keynote speech by researcher Luisa Bentivogli, which was very interesting, even though she spent rather too long explaining how LLMs work and (to my taste) not enough time on the part I was really interested in: their intrinsic and extrinsic risks, as well as the social impact.
But that wasn’t all the conference had to offer. I also managed to grab a spot in the enlightening workshop “How human is your workflow? Critical AI use for language professionals”, facilitated by Elina Nocera, Theresa Truax-Gischler and Allison Wright.
Elina discussed some of the problems around the ethics of AI, particularly copyright infringement. Her main argument was that “AI is not and cannot be a shortcut for building a brand, business or reputation people trust”. As this is precisely what I set out to do for my clients, it was reassuring to hear someone I respect thinking along the same lines!
Theresa presented the historical and legal aspects of AI. She reminded us that disruptions caused by new writing technologies are nothing new, and that the key to navigating these changes is what she calls “critical AI writing literacy”. This includes awareness of how these tools are changing writing, their biases and ethical issues, and their limitations.
In a particularly interesting slide, she listed several problems with AI, such as flawed logical argument, poor conceptual and factual accuracy, and its inability to understand meaning outside the form. The solution to these problems is a very simple one: use a human!
Finally, Allison demonstrated how she incorporated MT (DeepL) and an AI tool (Quillbot) into her translation process for a picture book project. That was my favourite part of the whole workshop, as I love learning about other translator’s processes. There’s always something new that you can incorporate into your own, and it gives you a great peek into another person’s brain.
She uses Quillbot’s paraphrase function during various stages of the translation, to look for ways of enriching her text. This seemed a brilliant way of using AI, so I was keen on trying it out. But, in the practical section of the workshop, I was disappointed to see that the tool only produces decent results into English. So, once again, all well and good, but no use for me just yet. I’ve since learned of a Brazilian Portuguese-specific tool (Clarice.ai), which I will test and see if it’s any better.
To sum it up, METM23 was the chill pill I think everyone needs: AI may sound like the bogeyman, but we all know he doesn’t really exist. The technology has its uses, is evolving and we would be foolish to just ignore it, but humans are the ones who actually understand the text.
If there were to be a bogeyman, thought, it wouldn’t be AI and what it can (or cannot) do: it would what our clients might think it can do. Our focus, then, should be to curb the hype (and the consequent fearmongering), figure out how best to educate the general public and demonstrate the value skilled humans can bring.
And, just in case you’re wondering… No. Other than the image on the top, no AI was used in the creation of this post!