Wales Wide Web

Last week I was at the EDEN Digital Learning Europe Conference in Graz. Sadly difficulties in transport limited by time and I missed the second day of the conference. But I am copying this report on LinkedIn by Felix Kwihangana from the University of Manchester of the EDEN Oxford Debate, where he supported the motion “Codes of Ethics for Generative Artificial Intelligence in Education are Useless/ Toothless/ A waste of time, chaired by Paul Prinsloo.

The debate provided an opportunity to explore and interrogate the complex issues around Ethical guidelines for hashtag#Generative hashtag#AI in education, their merits and demerits and everything in-between. I was privileged to work with Elisabeth Anna Guenther and Dr Martina Plantak in supporting the motion against an impressive team of debaters (Eamon Costello, Victoria Marín and Palitha Edirisingha).

In supporting the motion, we argued that hashtag#GenAI ethical guidelines in HE are often reactive, exclusive of non-western ways of knowing, based on a limited understanding of Generative AI, becoming obsolete before they are enacted due to the speed at which Generative AI is developing, and used as virtue signalling tools by institutions motivated by maintaining control rather than encouraging exploration and discovery. Using some historical cases (Alan Turing prosecution, The Prohibition), we argued that the ever changing values of society and the fast pace of Generative AI development could make Generative AI codes of ethics not only useless but also dangerous, when looked at within the historical lens of damage done in the name of "ethics", "values" and "norms" that societies end up dropping anyway. Needless to say, the opposing team had equally strong counterarguments, which made the debate worth its name! 

GPT 4 is quite useful for some things. I have been developing four Open Educational Resources around Labour Market Information, designed for careers professionals in different European countries. I was asked to include Issues for Reflection and a short multiple choice quiz on each of the OERs. I fed GPT4 the content of each OER and asked for 6 issues for reflection and 6 quiz questions. Fast as a flash they were done and are (in my view) very good. If I had to have done it without the AI it would have taken me at least half a day.

For other things GPT4 is less useful. And I have to say that its English, although grammatically good, is both stilted and plain. It also has the tendency to use somewhat odd English words, which I had always ascribed to it writing American English. But it seems not. In a Guardian newspaper newsletter, Alex Hern reports on work by AI influencer Jeremy Nguyen, at the Swinburne University of Technology in Melbourne, who has highlighted ChatGPT’s tendency to use the word “delve” in responses.

I have to say that I don't think I have ever used delve in anything I have written And talking to my Spanish English speaking friends none of them even new what the work means, Anyway Jeremy Hguyen says no individual use of the word can be definitive proof of AI involvement, but at scale it’s a different story. When half a percent of all articles on research site PubMed contain the word “delve” – 10 to 100 times more than did a few years ago – it’s hard to conclude anything other than an awful lot of medical researchers using the technology to, at best, augment their writing.

And according to a dataset of 50,000 ChatGPT responses, its not the only one. It seems the ten most overused words are: Explore, Captivate, Tapestry, Leverage, Embrace, Resonate, Dynamic, Testament Delve, and Elevate.

Now back to my hypothesis that its the fault of our American cousins. According to Alex Hearn an army of human testers are given access to the raw outputs from Large Language Models like ChatGPT, and instructed to try it out: asking questions, giving instructions and providing feedback. This feedback may be just approving of disapproving the outputs, but can be "more advanced, even amounting to writing a model response for the next step of training to learn from." And, here is the rub: "large AI companies outsource the work to parts of the global south, where anglophonic knowledge workers are cheap to hire."

Now back to the word "Delve."

There’s one part of the internet where “delve” is a much more common word: the African web. In Nigeria, “delve” is much more frequently used in business English than it is in England or the US. So the workers training their systems provided examples of input and output that used the same language, eventually ending up with an AI system that writes slightly like an African.