Dr Hannah Fry (UCL Centre for Advanced Spatial Analysis) discusses the limitations of AI, and how to teach Boris Johnson about exponential growth, in a wide-ranging interview for the Financial Times.
In a blur of flame-red hair, fizzy enthusiasm and effusive apologies, Hannah Fry arrives a little late at Little Social, a modish restaurant just off London’s Regent Street. We had hit on the bistro to meet because it was close to the Liberty department store, where Fry was due to have her ears pierced. I ask her whether this has indeed happened. It has, but on an earlier visit. Then, she drops some shocking news.
"The actual truth of it is that I was really sick earlier this year, I had cancer," she says. A routine smear test had flagged cervical cancer necessitating radical surgery and several weeks of recuperation. "It was very close to being very bad but I got really lucky. I got away with it," she says.
After coming through such difficult times, Fry thought about all the things she would like to do with the rest of her life but hadn’t done, such as quitting her job or travelling the world. "Unfortunately, in the pandemic you’re limited for choice so I just thought I’d get my ears pierced," she smiles.
Mercifully, Fry has rebounded strongly from her illness and has resumed her hectic schedule of writing, lecturing and broadcasting. With her sassy sense of humour, the 37-year-old academic has emerged as one of Britain’s most recognisable mathematicians (admittedly, a small subset of celebrities). Her TED talk on "The Mathematics of Love" has attracted 5.4m views and her televised Royal Institution science lectures in 2019 taught her young audience (among other things) how to win at pulling a Christmas cracker and avoid bags of green gunge. Fry spends her working life explaining numbers, probabilities and risk. But she says she gained a very different perspective when she became the number, as she puts it.
With disarming directness, she says there is now a one in 10 chance the cancer will return and, if it does, there is no cure. She finds herself thinking, like most people in such circumstances, that 90 per cent is close to 100 per cent so that is never going to happen. But when she reframes those same odds as one in 10 then "you suddenly feel you’re in a line-up of 10 people and one person’s going to get shot". Fully understanding the human dimension to any data set is a valuable lesson for anyone who deals with statistics. "It’s very easy when you are working with data to just think of people as though they are numbers."
One realisation she reached during her illness was that she was happy with her life, with a young family, a rewarding academic career and the opportunity to do other "cool stuff" on radio and television. "I was doing this thing where I was in a helicopter flying over the desert in Dubai at dawn. You’re not going to get that in an academic desk job, are you?" Although her sickness was horrible to live through, she hopes the enforced pause enabling her to reflect on life’s priorities might prove valuable.
As we consult the menu, I ask Fry if she is a foodie. "Honestly, no," she laughs. "I could pretend to be, but no." Overcoming her indifference, she orders a vegetable salad and Herdwick lamb rump, while I opt for a crab salad and Cornish cod with confit tomato. We both order glasses of white wine, a Sauvignon Blanc from the Ardèche for her and a Beaujolais Chardonnay for me.
Little Social is described in the Michelin guide as having "a clubby, comforting feel and a smart, stylish look, with on-trend copper, dark wood, and duck egg blue upholstery ensuring it fits in perfectly with its Mayfair surroundings". That seems about right. You can tell it is a high-class establishment: the dishes come in minute, exquisite portions, small works of art.
I ask Fry how she became interested in maths. She says she’s not completely sure but suggests that people’s relationship with the subject is an unstable equilibrium. A little nudge in one direction or the other will result in them either loving or hating it.
In her own case, her Irish mother, who had a "very interesting idea of what constitutes a fun summer holiday", insisted that Fry read one page of a maths textbook every day before going out to play. When she returned to school she did well in the subject and worked all the harder, a kind of reinforcement learning. Being good at maths soon became part of her identity.
When I ask if she is going to impose a similar learning regime on her two daughters, aged four and two, she demurs. Their teachers, and her husband, will be there for the "big stuff" when it comes to maths, she says, leaving Fry to concentrate on the "fun stuff". "I want my daughters to see maths as a playground of ideas," she says.
Fry was able to explore the creative dimensions of maths while studying for a PhD at University College London. For four years, Fry wrestled with the Navier-Stokes equations, which describe the way that fluids move, from drops of honey to the formations of galaxies. She compares maths research with geographic exploration: the solution exists somewhere and just needs to be discovered.
Sir Andrew Wiles, the Cambridge scholar who in 1995 published a proof for Fermat’s last theorem that had baffled mathematicians for 358 years, described his research as stumbling around in the scrubland for years before turning a corner and realising he had found a well-manicured garden. Fry remembers watching a documentary in which Wiles talked about that moment and recalls how his voice caught in his throat. "I just found that so powerful because he knew that he could live 1,000 lifetimes and never have that moment again," she says.
Fry experienced her own "lower-rent, pound shop" version of that discovery moment during her research. After six frustrating months rewriting and rewriting code, she hit on the solution when pushing the door open to enter the toilets one day. "I was like: ’Oh, my God, I know exactly what’s wrong with it,’" she says. Comparing the code to Lego bricks, she says she had been focusing too much on the structure of the bricks rather than the joins.
Fry has remained at UCL and lectures at the Centre for Advanced Spatial Analysis, studying complex social and economic systems. During lockdown, she taught an online masters course to up to 120 students from home, sitting on her sofa. She says she feels bad for those students who missed out on the full university experience. "However hard you try, you just can’t replicate it."
UCL has been supportive of her broadcasting career, she says, although she suspects some pure mathematicians roll their eyes at the way she presents the subject. "I’m sure there is lots of tutting but they just don’t tut to my face." What she is adamant about is that maths must be open to everyone, not just to "white man-shaped holes". Maths belongs to her as much as anyone else. "I don’t need permission to talk about something that I think is interesting."
One thing she finds interesting is how maths-based computer algorithms are being used in automated decision-making systems. Fry has spent a lot of time talking with the experts at Google’s DeepMind in London, who have been pioneering research in artificial intelligence, and is now working on her second podcast series about the company’s research. She has also written Hello World (2018), investigating how to be human in the age of the machine.
She says she is astonished at the rate of progress in the field of deep learning, which has become so good at text, voice and image recognition. "Ten years ago, these systems couldn’t recognise a cat in a picture and now we’re questioning how pervasive and invasive facial recognition has become. It’s mad to me," she says.
But she also highlights the chicanery of some people in the field and the limitations of AI systems. One methodology she uses to assess whether claims about AI are bogus is her so-called "magic" test. "If you take out all the technical words and replace them with the word ’magic’ and the sentence still makes grammatical sense, then you know that it’s going to be bollocks."
Even the most sophisticated companies, such as Amazon, find it difficult to draw meaningful conclusions from their data: its book-recommendation service is "rubbish", Fry says. When I tell her that Amazon keeps recommending her books to me, she replies: "Does it? Terrible, that’s absolutely awful."
She worries about the "amazing position of dominance" giant tech companies have acquired, threatening privacy and democracy if left unchecked. She says the global power these companies have amassed hasn’t been seen "since empires were a thing". However, while admitting that she might be "using optimism as a coping strategy", she insists that AI can still be used for enormous good.
Some researchers have highlighted how human biases have been unwittingly encoded into AI systems, arguing they should not be used in areas such as predictive policing or judicial sentencing. But Fry counters that if carefully designed they can still act as a useful check on poor human decision-making. "What you’re comparing them to is human judges making absolutely God-awful decisions because they cannot make fair and consistent decisions for toffee," she says.
Declaring her lamb to have been "absolutely delicious", Fry resists the temptations of dessert and opts for a caffe latte. By now the restaurant has filled up with later lunchers laden with posh shopping bags.
The power of science to improve our lives has been showcased by the Covid-19 pandemic and the rapid development of vaccines. But it has also exposed the limitations of public understanding. Fry has been one of the spikiest critics of what she calls the "theatre of data", designed to add a layer of authority to a system that owes more to human impulse.
She says she has been "blown away" by the poor scientific knowledge displayed by some politicians. One chart presented by Downing Street last year with a misleading y-axis prompted her to tweet: "I think I just vomited into my mouth."
Her particular bugbear has been the reluctance to use logarithmic axes, which she argues are a better way to represent exponential growth because they give a clearer view of the pandemic’s trajectory. I ask her how she would best explain the concept of exponential growth to the prime minister given the chance. "The thing you have to understand about exponential growth is that it feels like nothing is happening for ages and then it’s like an unstoppable truck that’s just slamming into a wall," she says.
Her favourite example is of an imaginary lily pad on a pond doubling its area every day. It starts growing in one minuscule corner and covers the whole surface of the pond after a month. By day 20, the lily pad is still almost invisible. By day 28, it covers a quarter of the pond, by day 29 it covers half, and by day 30 it covers it all.
It is a similar story with the hospitalisation numbers during the latest upsurge in the Covid-19 pandemic in the UK, she suggests. "Sure, they’re at a much lower level than they were in January but they are on this trajectory. We’re at day two, day three of this lily pad and the pattern exists."
Fry’s ability to communicate in such vivid terms was partly honed in an academic comedy club. She has also struck up a sparky radio double act with biologist Adam Rutherford and the two of them have just written a book together called Rutherford and Fry’s Complete Guide to Absolutely Everything* (*Abridged). Although she teases Rutherford that biology "is just giving stuff Latin names", the two share a passion for making science more accessible.
Fry is also working on a BBC documentary about her illness, which she wants to call When Your Number’s Up. She started recording in isolation at home and found it therapeutic, she says. Realising she could not just film an hour of herself crying, she began to explore our relationship with the fear of disease, our sense of mortality and our insistence on reducing all risk as far as possible. "We will see if I can get any of that in. It will probably just end up being an hour of crying."
As if all that were not enough to keep her busy, Fry writes essays for the New Yorker magazine, most recently on the limitations of data in predicting human behaviour and the power of graphs. Her writing routine consists of taking her 10-year-old cockapoo Molly for a morning walk and dictating 200 words or so into a voice recorder, which sets her up for the day.
She is forever thinking of imaginative ways to bring the wonders of science and maths to life, resisting the "Power Point and elbow patch" approach to the subject. "There’s so much humour and strangeness about who we are and how we want the world to be that is just missing from these very formal depictions of science, which I think is a bit of a shame really," she says, sipping the last of her coffee.
With that, Fry sweeps off on her mission to explore and explain the "weirdness of reality".
This article originally appeared in the Financial Times on 30 July 2021.