What does a picture of a (very nice) group of people have to do with my blog this week? It was taken at a songwriting retreat, where the main goal of the three days was to collaborate on a song and finally perform it to the group. I collaborated with Pete, on the far right. Our topic was 'Years' and we wrote a song called, 'Time's not on my side any more'.

The genesis of our song and this blog was talking about AI with my friend during the week (before going on an extremely hard bike ride in the back country behind Gibbston). As she walked past me on our way to breakfast, I was working on a funding proposal. “What are you doing?” she asked.

“Trying to get a section of a proposal down to 1120 words from the 2500 words the researcher gave me.”

“Why don’t you get ChatGPT to do that for you?”

The short answer was because I didn't think ChatGPT can do that sort of editing well. ChatGPT can reduce the number of words but it struggles to deliver a text that is both shorter and includes the necessary detail for a research funding proposal. In general, I consider large language models can yet do the work I do, although I could be wrong. The short answer I provided was not good enough for my friend so I had to do a lot of thinking to break down the components of why I was doing the editing without AI assistance.

1) I understand the research proposals I work on by interacting with them. I rarely just read a proposal – I start working on the text at the same time I start reading it. In the process of rewriting I clarify whether I can understand the concepts in the document – to restructure text requires understanding what you are reading and editing. In my experience, improving a piece of text cannot be done at any deep level simply using comments boxes. Improvement requires being deeply embedded into what you are reading, which requires interaction i.e. either rewriting or summarising separately in your own words. Anyone who has studied for exams will remember the best way to really know your material is to rewrite it and summarise it, not just read it.

So, if I got ChatGPT to reduce my 2500 word text to 1120 words, I wouldn’t necessarily know whether the reduced version carried the same meaning as the original text because I wouldn't know what the original text actually meant. I certainly wouldn’t know if all the information had been conveyed or whether it was structured the best way. The only way I’d know that is if I tried to reduce the text myself and then compared the ChatGPT version to my version, at which point why I would use ChatGPT because I would have already done the work?

2) To create a good summary requires understanding at a deeper level than the level of the summary you are creating – it is very difficult to summarise or teach at the most complex level you understand. Normally you summarise/teach at least ‘one level up’. So if I hand over to ChatGPT to create a summarised version of material, my deepest level of knowledge will be at the summary level.

I found this conclusion frightening. Powered vehicles have meant we gradually lose our physical ability to move ourselves from place to place. Now people seek ‘artificial’ ways to maintain/regain physical ability e.g. by going to the gym (often driving there!). In the same way, AI may mean we give away our mental ability to understand information at a deep level in the course of work. Will we then have to find artificial ways of exercising our brains because we have got AI to do the heavy lifting? Will we be reading and answering questionnaires on complex pieces of text as a voluntary exercise so we don’t lose our mental capacities? Or will we not even notice we are losing our ability, in the same way that the reading age media targets has gradually decreased over time without anyone apparently minding.

In the 1940s, paper articles were aimed at college graduates. Flesch and Gunning, readability experts, persuaded the papers to 'increase their readability' to improve circulation. After their work, the majority of large newspapers could be understood by early high school students. However, the goal now is text readable by late primary/intermediate school children. If you want more challenging reading you need to go to the New York Times – see the picture below of the reading age of different news outlets.