Mathematical personality

Last week to the Royal Institution to hear Hannah Fry talk about algorithms. A sell out occasion, involving HRH the Duke of Kent, but just one penguin suit. A plus was that, in honour of HRH, we got the director doing the introductions and questions, a director who might be a bit tiresome in other ways, but is a lot less tiresome than a bouncy blonde on his staff. A chap we had heard previously and noticed at reference 1.

Goat very crowded, to the point of there being very few seats left in the usually quiet bar upstairs. I felt slightly bad about getting myself served before the young lady next to me downstairs, but she was not paying attention, not noticing when the barman's eyes lighted upon her. So I let him serve me instead. Onto the facilities at the Institution to be reminded that they are one of the few places still to offer paper towels instead of hot air machines. This despite the scientific flavour of the place.

Hannah Fry is a young associate professor (whatever one of those is here in England) who did mathematics, then fluid dyanmics and now advanced spatial analysis. But I suspect she spends more time being a television personality, with the result that I found her a bit irritating in the flesh. Despite which, she had some good material. I offer a selection.

She opened by telling us about an experiment to train novices to recognise tumours in tissue slides. Which turned out to be a lot easier than one might think, with a few weeks being quite enough, nothing like the time taken to train radiographers. And, picking up the curious phenomenon of the wisdom of the crowd touched on at reference 5, if you averaged the scores of all 100 novices (or whatever the number was), you got near perfect results. With the novices in question being pigeons.

Then she had us tell her which of two cantatas was composed by the real Bach, with the other being composed by a computer. I had no idea, despite having heard a few over the years, and the vote was pretty even. Her line was that the computed cantata had been computed by doing on chord changes something like predictive text on telephones, a wheeze devised by a real composer, but which would be detected by any half-way serious musician as being far too simple minded for the master.

Then we had computers doing image analysis, with the message seeming to be that computers were not very good at nonsense, like a tree full of pink sheep. She also had a couple of countryside examples, with the computer quite wrongly seeing sheep in one and cows in the others. Except that someone in the audience (a plant?) explained at question time that she came from the country and she knew that the sheep scene was very suitable for sheep and the cow scene was very suitable for cows. So the computer had not been that dumb and the cheap laugh that Fry got at its expense was not really fair.

Then we had the nun study. Not particularly relevant but a longitudinal study of nearly 700 nuns, a study which includes the donation of their brains after death, all kept in tubs in a large storage room, which made a slightly unsettling picture. But given the large amount of other information about these nuns, and the fact that their life styles were all very much the same, a wonderful source of information about Alzheimer's, the focus of this study. An example of big data at work. See claimed that it had been shown that one could predict the risk of Alzheimer's from the essays they wrote as much younger women - lots of complicated sentences means low risk - but I have been unable to check this claim. See references 6 and 7.

Then we had cars. Her point here is that we have computers drive the car and get them to raise an alert when things look wrong. But humans who have been dozing off do not react well in these circumstances, and it might be better for the human to drive the car, but have the computer in the back seat ready to warn you if it thinks something is not quite right. Computers very good at doing lots of routine checking of the road ahead, something that humans can be quite bad at.

Lastly we had getting computers to read scans and look for tumours, which seemed to be a battle between getting all the tumours and not getting too many false positives. Her point here was that a human very rarely made a mistake if told that a scan needed checking. But they did make mistakes when doing mass screening, and she gave the example of humans missing sketches of gorillas which had been added to scans - a trick born of the well known gorilla in the basketball game played at management courses around the world.

Her bottom line seemed to be that the best thing to do was to work with AI and its algorithms to get the best out of humans and the best out of AI. Much the same as the line advanced by Rajpoot at the event noticed at reference 4.

But all a bit glib. Too much the presenter of a television show and not enough of the scientist, which was no doubt lurking underneath. She was also punting for her new book (snapped above), a punting which included a signing session. I passed.

Back for a little something at the Marquis, also very crowded. The barman knew nothing about the Sancerre I had been offered on my last visit, back in September, and so I had to settle for common or garden sauvignon blanc. There was also a minor incident in the toilets involving the attendance of both the police and an ambulance. No violence, so we assumed substance related.

PS: during the longueurs of the talk, I had a go at counting the people in the bay, front left. The trick on this occasion was to count the rows often enough that the row counts were lodged in  memory, then add up the rows. So 6, 7, 8, 9, 11, 12, 13, 13. Plus 3 out front. So 79 or 82 depending. First four rows totaled to a handy 30, but I found it hard to get a grip on the last two rows for some reason.

Reference 1: https://psmv3.blogspot.com/2018/06/hot-air.html.

Reference 2: http://www.hannahfry.co.uk/. For Hannah Fry.

Reference 3: https://www.ucl.ac.uk/bartlett/casa/. For spatial analysis. Where they develop new responses to pressing world issues.

Reference 4: https://psmv3.blogspot.com/2018/10/popular-science.html.

Reference 5: https://psmv3.blogspot.com/2018/09/getting-it-wrong.html.

Reference 6: Healthy Aging and Dementia: Findings from the Nun Study - David A. Snowdon - 2003. Both sadly and unusually, neither Bing nor Google found a freebie and I drew the line at $35. So the best I could do was the abstract which follows and the short article at reference 7. Abstract: the Nun Study is a longitudinal study of 678 Catholic sisters 75 to 107 years of age who are members of the School Sisters of Notre Dame congregation. Data collected for this study include early and middle-life risk factors from the convent archives, annual cognitive and physical function evaluations during old age, and postmortem neuropathologic evaluations of the participants' brains. The case histories presented include a centenarian who was a model of healthy aging, a 92-year-old with dementia and clinically significant Alzheimer disease neuropathology and vascular lesions, a cognitively and physically intact centenarian with almost no neuropathology, and an 85-year-old with well-preserved cognitive and physical function despite a genetic predisposition to Alzheimer disease and an abundance of Alzheimer disease lesions. These case histories provide examples of how healthy aging and dementia relate to the degree of pathology present in the brain and the level of resistance to the clinical expression of the neuropathology. Comment: or put another way, an apparent disconnect between bodily structure and function. A gap to be filled by the soul?

Reference 7: https://danablog.org/2011/09/21/world-alzheimers-day-do-as-the-nuns-do/.