I’m human. It seems odd to have to say this, but recent research has shown that 6.96% of 857,434 news articles from July 1, 2024, were AI-generated (published by a Brooklyn based company I really admire, called Pangram Labs). Worse than that, in the same research they showed a quarter of published content on one day last summer was at least in part AI generated. Gulp.
Figure 1 https://www.pangram.com/blog/one-day-of-ai-news
We’ve all seen it though – that colleague who thought punctuation was something that happened to car tyres, who suddenly discovered spelling, grammar, apostrophes and the Oxford (or Harvard, if you so desire) comma. The perfect covering letter for a job application, which managed to incorporate all the keywords that the job posting mentioned. Or a colleague suddenly writing a rather nice and well-written blog…
Here in Canary Towers, we also incorporate AI-generated content and we give a big ‘This is AI generated content’ label when we present such material. Meeting summarisation and action point generation is, in truth, very impressive, but being the pragmatists that we are, we do still give the meeting owner the ability to edit the content so they can emphasise the key point of the meeting – which the AI cannot necessarily divine.
We have also just added additional AI content. After a meeting, we present back to the meeting organiser some statistics against our FiRE score (Focus, Inclusion, Respect and Energy) and we show ways that the score can be improved as below
The latest addition is as follows:
It’s largely harmless and (we hope) helpful. We wrestled with naming people who were in the actual meeting in this summary and settled on the side of doing just that. So, future summaries may well suggest that John or Jane (the ‘quieter’ members of the meeting above) might be asked to contribute more. Or perhaps, if they regularly don’t engage, maybe ask them if they really need to attend the meeting.
It’s a little more ‘direct’ but if we are trying to ‘make meetings better’ (and we are) then direct is actually just code for sensible efficiency/productivity gains.
AI-generated content will also make generalised statements without knowing about certain cultural aspects, let’s take ‘video on’ encouragement for meeting participants, here’s a generated reply from Copilot:
It’s good, largely harmless, a bit dull, but it does give some references – and as a pointer it feels fine. In fact, we could just pass that on as advice for meetings with low video on engagement, couldn’t we?
But hold on. We have a client in the Middle East, who encourages colleagues to work from home, men and women. It’s a conservative country so the expectation of women would be that their hair is covered if they left their home, but inside their home their hair would not need to be covered. So actually this bland advice misses any kind of cultural nuance.
Nevertheless, we have incorporated such advice into Meeting Canary but with the provision that we have also made sure that our corporate subscribers have the ability to tailor the advice given to their own brand/culture values.
And finally how can I prove that none of this was AI generated? I turned to the wonderful pangram and ran all the above though their checker:
Figure 2 – Look I’m human: https://www.pangram.com/
But the video advice above (which I deliberately included as an image not as the text it contains, so as to not skew the results) when tested on its own scored:
Figure 3 – Aha, it’s a robot: https://www.pangram.com/
And if I included the words from Copilot as text so I have a mixed human/synthetic prose it scores:
Figure 4 Yup them again https://www.pangram.com/