by Hannah Fry
The procedure for updating the neurons is known as the ‘backpropagation algorithm’. We start with the final neuron that outputs the probability that the image is a dog. Let’s say we fed in an image of a dog and it predicted that the image had a 70 per cent chance of being a dog. It looks at the signals it received from the previous layer and says, ‘The next time I receive information like that I’ll increase my probability that the image is a dog’. It then says to each of the neurons in the previous layer, ‘Hey, if you’d given me this signal instead I would have made a better prediction’. Each of those neurons looks at its input signals and changes what it would output the next time. And then it tells the previous layer what signals it should have sent, and so on through all the layers back to the beginning. It is this process of propagating the errors back through the neural network that leads to the name ‘the backpropagation algorithm’.
For a more detailed overview of neural networks, how they are built and trained, see Pedro Domingos, The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World (New York: Basic Books, 2015).
12 Alex Krizhevsky, Ilya Sutskever and Geoffrey E. Hinton, ‘ImageNet classification with deep convolutional neural networks’, in F. Pereira, C. J. C. Burges, L. Bottou and K. Q. Weinberger, eds, Advances in Neural Information Processing Systems 25 (La Jolla, CA, Neural Information Processing Systems Foundation, 2012), pp. 1097–1105, http://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf. This particular algorithm is known as a convolutional neural network. Rather than the entire image being fed in, the algorithm first applies a host of different filters and looks for local patterns in the way the picture is distorted.
13 Marco Tulio Ribeiro, Sameer Singh and Carlos Guestrin, ‘“Why should I trust you?” Explaining the predictions of any classifier’, Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, 2016, pp. 1135–44, http://www.kdd.org/kdd2016/papers/files/rfp0573-ribeiroA.pdf.
14 This was compared to the assessment of a panel of experts, whose collective analysis was considered to be the ‘ground truth’ for what was contained in the slides.
15 Trafton Drew, Melissa L. H. Vo and Jeremy M. Wolfe, ‘The invisible gorilla strikes again: sustained inattentional blindness in expert observers’, Psychological Science, vol. 24, no. 9, Sept. 2013, pp. 1848–53, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3964612/.
16 The gorilla is located in the top-right-hand side of the image.
17 Yun Liu, Krishna Gadepalli, Mohammad Norouzi, George E. Dahl, Timo Kohlberger, Aleksey Boyko, Subhashini Venugopalan, Aleksei Timofeev, Philip Q. Nelson, Greg S. Corrado, Jason D. Hipp, Lily Peng and Martin C. Stumpe, ‘Detecting cancer metastases on gigapixel pathology images’, Cornell University Library, 8 March 2017, https://arxiv.org/abs/1703.02442.
18 Dayong Wang, Aditya Khosla, Rishab Gargeya, Humayun Irshad and Andrew H. Beck, ‘Deep learning for identifying metastatic breast cancer’, Cornell University Library, 18 June 2016, https://arxiv.org/abs/1606.05718.
19 David A. Snowdon, ‘The Nun Study’, Boletin de LAZOS de la Asociación Alzheimer de Monterrey, vol. 4, no. 22, 2000; D. A. Snowdon, ‘Healthy aging and dementia: findings from the Nun Study’, Annals of Internal Medicine, vol. 139, no. 5, Sept. 2003, pp. 450–54.
20 The idea density – a proxy for linguistic complexity – was calculated by counting up the number of unique ideas each nun used per string of ten words. There’s a nice overview here: Associated Press, ‘Study of nuns links early verbal skills to Alzheimer’s, Los Angeles Times, 21 Feb. 1996, http://articles.latimes.com/1996-02-21/news/mn-38356_1_alzheimer-nuns-studied.
21 Maja Nielsen, Jørn Jensen and Johan Andersen, ‘Pre-cancerous and cancerous breast lesions during lifetime and at autopsy: a study of 83 women’, Cancer, vol. 54, no. 4, 1984, pp. 612–15, http://onlinelibrary.wiley.com/wol1/doi/10.1002/1097-0142(1984)54:4%3C612::AID-CNCR2820540403%3E3.0.CO;2-B/abstract.
22 H. Gilbert Welch and William C. Black, ‘Using autopsy series to estimate the disease “reservoir” for ductal carcinoma in situ of the breast: how much more breast cancer can we find?’, Annals of Internal Medicine, vol. 127, no. 11, Dec. 1997, pp. 1023–8, www.vaoutcomes.org/papers/Autopsy_Series.pdf.
23 Getting an exact statistic is tricky because it depends on the country and demographic (and how aggressively your country screens for breast cancer). For a good summary, see: http://www.cancerresearchuk.org/health-professional/cancer-statistics/statistics-by-cancer-type/breast-cancer.
24 Quotations from Jonathan Kanevsky are from personal communication.
25 ‘Breakthrough method predicts risk of DCIS becoming invasive breast cancer’, Artemis, May 2010, http://www.hopkinsbreastcenter.org/artemis/201005/3.html.
26 H. Gilbert Welch, Philip C. Prorok, A. James O’Malley and Barnett S. Kramer, ‘Breast-cancer tumor size, overdiagnosis, and mammography screening effectiveness’, New England Journal of Medicine, vol. 375, 2016, pp. 1438–47, http://www.nejm.org/doi/full/10.1056/NEJMoa1600249.
27 Independent UK Panel on Breast Cancer Screening, ‘The benefits and harms of breast cancer screening: an independent review’, Lancet, vol. 380, no. 9855, 30 Oct. 2012, pp. 1778–86, http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(12)61611-0/abstract.
28 Personal communication.
29 Andrew H. Beck, Ankur R. Sangoi, Samuel Leung, Robert J. Marinelli, Torsten O. Nielsen, Marc J. van de Vijver, Robert B. West, Matt van de Rijn and Daphne Koller, ‘Systematic analysis of breast cancer morphology uncovers stromal features associated with survival’, Science Transitional Medicine, 19 Dec. 2014, https://becklab.hms.harvard.edu/files/becklab/files/sci_transl_med-2011-beck-108ra113.pdf.
30 Phi Vu Tran, ‘A fully convolutional neural network for cardiac segmentation in short-axis MRI’, 27 April 2017, https://arxiv.org/pdf/1604.00494.pdf.
31 ‘Emphysema’, Imaging Analytics, Zebra Medical, https://www.zebra-med.com/algorithms/lungs/.
32 Eun-Jae Lee, Yong-Hwan Kim, Dong-Wha Kang et al., ‘Deep into the brain: artificial intelligence in stroke imaging’, Journal of Stroke, vol. 19, no. 3, 2017, pp. 277–85, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5647643/.
33 Taylor Kubota, ‘Deep learning algorithm does as well as dermatologists in identifying skin cancer’, Stanford News, 25 Jan. 2017, https://news.stanford.edu/2017/01/25/artificial-intelligence-used-identify-skin-cancer/.
34 Jo Best, ‘IBM Watson: the inside story of how the Jeopardy-winning supercomputer was born, and what it wants to do next’, Tech Republic, n.d., https://www.techrepublic.com/article/ibm-watson-the-inside-story-of-how-the-jeopardy-winning-supercomputer-was-born-and-what-it-wants-to-do-next/.
35 Jennings Brown, ‘Why everyone is hating on IBM Watson, including the people who helped make it’, Gizmodo, 14 Aug. 2017, https://www.gizmodo.com.au/2017/08/why-everyone-is-hating-on-watsonincluding-the-people-who-helped-make-it/.
36 https://www.theregister.co.uk/2017/02/20/watson_cancerbusting_trial_on_hold_after_damning_audit_report/
37 Casey Ross and Ike Swetlitz, ‘IBM pitched its Watson supercomputer as a revolution in cancer care. It’s nowhere close’, STAT, 5 Sept. 2017, https://www.statnews.com/2017/09/05/watson-ibm-cancer/.
38 Tomoko Otake, ‘Big data used for rapid diagnosis of rare leukemia case in Japan’, Japan Times, 11 Aug. 2016, https://www.japantimes.co.jp/news/2016/08/11/national/science-health/ibm-big-data-used-for-rapid-diagnosis-of-rare-leukemia-case-in-japan/#.Wf8S_hO0MQ8.
39 ‘Researchers validate five new genes responsible for ALS’, Science Daily, 1 Dec. 2017, https://www.sciencedaily.com/releases/2017/12/171201104101.htm.
40 John Freedman, ‘A reality check for IBM’s AI ambitions’, MIT Technology Review, 27 June 2017.
41 Asthma facts and statistics, Asthma UK, 2016, https://www.asthma.org.uk/about/media/facts-and-statistics/; Asthma in the US, Centers for Disease Control and Prevention, May 2011, https://www.cdc.gov/
vitalsigns/asthma/index.html.
42 ‘Schoolgirl, 13, who died of asthma attack was making regular trips to A&E and running out of medication – but was NEVER referred to a specialist even when her lips turned blue, mother tells inquest’, Daily Mail, 13 Oct. 2015, http://www.dailymail.co.uk/news/article-3270728/Schoolgirl-13-died-asthma-attack-not-referred-specialist-lips-turned-blue.html.
43 My Data, My Care: How Better Use of Data Improves Health and Wellbeing (London: Richmond Group of Charities, Jan. 2017), https://richmondgroupofcharities.org.uk/publications.
44 Terence Carney, ‘Regulation 28: report to prevent future deaths’, coroner’s report on the case of Tamara Mills, 29 Oct. 2015, https://www.judiciary.gov.uk/publications/tamara-mills/.
45 Jamie Grierson and Alex Hern, ‘Doctors using Snapchat to send patient scans to each other, panel finds’, Guardian, 5 July 2017, https://www.theguardian.com/technology/2017/jul/05/doctors-using-snapchat-to-send-patient-scans-to-each-other-panel-finds.
46 Even if you get around all of those issues, sometimes the data itself just doesn’t exist. There are thousands of rare diseases with an underlying genetic cause that are effectively unique. Doctors have enormous difficulty spotting one of these conditions because in many cases they will have never seen it before. All the algorithms in the world won’t solve issues with tiny sample sizes.
47 Hal Hodson, ‘Revealed: Google AI has access to huge haul of NHS patient data’, New Scientist, 29 April 2016, https://www.newscientist.com/article/2086454-revealed-google-ai-has-access-to-huge-haul-of-nhs-patient-data/.
48 Actually, much of the blame for this so-called ‘legally inappropriate’ deal has been laid at the door of the Royal Free Trust, which was probably a bit too eager to partner up with the most famous artificial intelligence company in the world. See the letter from Dame Fiona Caldicott, the national data guardian, that was leaked to Sky News: Alex Martin, ‘Google received 1.6 million NHS patients’ data on an “inappropriate legal basis”’, Sky News, 15 May 2017, https://photos.google.com/share/AF1QipMdd5VTK0RNQ1AC3Dda1526CMG0vPD4P3x4x6_qmj0Zf101rbKyxfkfyputSPvqdA/photo/AF1QipP1_rnJMXkRyy3IuFHasilQHYEknKgnHFOFEy4T?key=U2pZUDM4bmo5RHhKYVptaDlkbEhfVFh4Rm1iVUVR.
49 Denis Campbell, ‘Surgeons attack plans to delay treatment to obese patients and smokers’, Guardian, 29 Nov. 2016, https://www.theguardian.com/society/2016/nov/29/surgeons-nhs-delay-treatment-obese-patients-smokers-york.
50 Nir Eyal, ‘Denial of treatment to obese patients: the wrong policy on personal responsibility for health’, International Journal of Health Policy and Management, vol. 1, no. 2, Aug. 2013, pp. 107–10, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3937915/.
51 For a description of the procedures, see http://galton.org/essays/1880-1889/galton-1884-jaigi-anthro-lab.pdf.
52 Francis Galton, ‘On the Anthropometric Laboratory at the late international health exhibition’, Journal of the Anthropological Institute of Great Britain and Ireland, vol. 14, 1885, pp. 205–21.
53 ‘Taste’, https://permalinks.23andme.com/pdf/samplereport_traits.pdf.
54 ‘Sneezing on summer solstice?’, 23andMeBlog, 20 June 2012, https://blog.23andme.com/health-traits/sneezing-on-summer-solstice/.
55 ‘Find out what your DNA says about your health, traits and ancestry’, 23andMe, https://www.23andme.com/en-gb/dna-health-ancestry/.
56 Kristen v. Brown, ‘23andMe is selling your data but not how you think’, Gizmodo, 14 April 2017, https://gizmodo.com/23andme-is-selling-your-data-but-not-how-you-think-1794340474.
57 Michael Grothaus, ‘How23andMe is monetizing your DNA’, Fast Company, 15 Jan. 2015, https://www.fastcompany.com/3040356/what-23andme-is-doing-with-all-that-dna.
58 Rob Stein, ‘Found on the Web, with DNA: a boy’s father’, Washington Post, 13 Nov. 2005, http://www.washingtonpost.com/wp-dyn/content/article/2005/11/12/AR2005111200958.html.
59 After having his DNA tested, the young man learned that a particular pattern on his Y-chromosome – passed from father to son – was also shared by two people with the same surname (distant relatives on his father’s side). That surname, together with the place and date of birth of his father, was enough to track him down.
60 M. Gymrek, A. L. McGuire, D. Golan, E. Halperin and Y. Erlich, ‘Identifying personal genomes by surname inference’, Science, vol. 339, no. 6117, Jan. 2013, pp. 321–4, https://www.ncbi.nlm.nih.gov/pubmed/23329047.
61 Currently, genetic tests for Huntington’s disease are not available from any commercial DNA testing kits.
62 Matthew Herper, ‘23andMe rides again: FDA clears genetic tests to predict disease risk’, Forbes, 6 April 2017, https://www.forbes.com/sites/matthewherper/2017/04/06/23andme-rides-again-fda-clears-genetic-tests-to-predict-disease-risk/#302aea624fdc.
Cars
1 DARPA, Grand Challenge 2004: Final Report (Arlington, VA: Defence Advanced Research Projects Agency, 30 July 2004), http://www.esd.whs.mil/Portals/54/Documents/FOID/Reading%20Room/DARPA/15-F-0059_GC_2004_FINAL_RPT_7-30-2004.pdf.
2 The Worldwide Guide to Movie Locations, 7 Sept. 2014, http://www.movie-locations.com/movies/k/Kill_Bill_Vol_2.html#.WkYiqrTQoQ8.
3 Mariella Moon, What you need to know about DARPA, the Pentagon’s mad science division, Engadget, 7 July 2014, https://www.engadget.com/2014/07/07/darpa-explainer/.
4 DARPA, Urban Challenge: Overview, http://archive.darpa.mil/grandchallenge/overview.html.
5 Sebastian Thrun, ‘Winning the DARPA Grand Challenge, 2 August 2006’, YouTube, 8 Oct. 2007, https://www.youtube.com/watch?v=j8zj5lBpFTY.
6 DARPA, Urban Challenge: Overview.
7 ‘DARPA Grand Challenge 2004 – road to …’ , YouTube, 22 Jan. 2014, https://www.youtube.com/watch?v=FaBJ5sPPmcI.
8 Alex Davies, ‘An oral history of the DARPA Grand Challenge, the grueling robot race that launched the self-driving car’, Wired, 8 March 2017, https://www.wired.com/story/darpa-grand-challenge-2004-oral-history/ .
9 ‘Desert race too tough for robots’, BBC News, 15 March, 2004, http://news.bbc.co.uk/1/hi/technology/3512270.stm.
10 Davies, ‘An oral history of the DARPA Grand Challenge’.
11 Denise Chow, ‘DARPA and drone cars: how the US military spawned self-driving car revolution’, LiveScience, 21 March 2014, https://www.livescience.com/44272-darpa-self-driving-car-revolution.html.
12 Joseph Hooper, ‘From Darpa Grand Challenge 2004 DARPA’s debacle in the desert’, Popular Science, 4 June 2004, https://www.popsci.com/scitech/article/2004-06/darpa-grand-challenge-2004darpas-debacle-desert.
13 Davies, ‘An oral history of the DARPA Grand Challenge’.
14 DARPA, Report to Congress: DARPA Prize Authority. Fiscal Year 2005 Report in Accordance with 10 U.S.C. 2374a, March 2006, http://archive.darpa.mil/grandchallenge/docs/grand_challenge_2005_report_to_congress.pdf.
15 Alan Ohnsman, ‘Bosch and Daimler to partner to get driverless taxis to market by early 2020s’, Forbes, 4 April 2017, https://www.forbes.com/sites/alanohnsman/2017/04/04/bosch-and-daimler-partner-to-get-driverless-taxis-to-market-by-early-2020s/#306ec7e63c4b.
16 Ford, Looking Further: Ford Will Have a Fully Autonomous Vehicle in Operation by 2021, https://corporate.ford.com/innovation/autonomous-2021.html.
17 John Markoff, ‘Should your driverless car hit a pedestrian to save your life?’, New York Times, 23 June 2016, https://www.nytimes.com/2016/06/24/technology/should-your-driverless-car-hit-a-pedestrian-to-save-your-life.html.
18 Clive Thompson, Anna Wiener, Ferris Jabr, Rahawa Haile, Geoff Manaugh, Jamie Lauren Keiles, Jennifer Kahn and Malia Wollan, ‘Full tilt: when 100 per cent of cars are autonomous’, New York Times, 8 Nov. 2017, https://www.nytimes.com/interactive/2017/11/08/magazine/tech-design-autonomous-future-cars-100-percent-augmented-reality-policing.html#the-end-of-roadkill.
19 Peter Campbell, ‘Trucks headed for a driverless future: unions warn that millions of drivers’ jobs will be disrupted’, Financial Times, 31 Jan. 2018, https://www.ft.com/content/7686ea3e-e0dd-11e7-a0d4-0944c5f49e46.
20
Markus Maurer, J. Christian Gerdes, Barbara Lenz and Hermann Winner, Autonomous Driving: Technical, Legal and Social Aspects (New York: Springer, May 2016), p 48.
21 Stephen Zavestoski and Julian Agyeman, Incomplete Streets: Processes, Practices, and Possibilities (London: Routledge, 2015), p. 29.
22 Maurer et al., Autonomous Driving, p. 53.
23 David Rooney, Self-guided Cars (London: Science Museum, 27 Aug. 2009), https://blog.sciencemuseum.org.uk/self-guided-cars/.
24 Blake Z. Rong, ‘How Mercedes sees into the future’, Autoweek, 22 Jan. 2014, http://autoweek.com/article/car-news/how-mercedes-sees-future.
25 Dean A. Pomerleau, ALVINN: An Autonomous Land Vehicle In a Neural Network, CMU-CS-89-107 (Pittsburgh: Carnegie Mellon University, Jan. 1989), http://repository.cmu.edu/cgi/viewcontent.cgi?article=2874&context=compsci.
26 Joshua Davis, ‘Say hello to Stanley’, Wired, 1 Jan. 2006, https://www.wired.com/2006/01/stanley/; and, for more detail, Dean A. Pomerleau, Neural Network Perception for Mobile Robot Guidance (New York: Springer, 2012), p. 52.
27 A. Filgueira, H. González-Jorge, S. Lagüela, L. Diaz-Vilariño and P. Arias, ‘Quantifying the influence of rain in LiDAR performance’, Measurement, vol. 95, Jan. 2017, pp. 143–8, DOI: https://doi.org/10.1016/j.measurement.2016.10.009; https://www.sciencedirect.com/science/article/pii/S0263224116305577.
28 Chris Williams, ‘Stop lights, sunsets, junctions are tough work for Google’s robo-cars’, The Register, 24 Aug. 2016, https://www.theregister.co.uk/2016/08/24/google_self_driving_car_problems/.
29 Novatel, IMU Errors and Their Effects, https://www.novatel.com/assets/Documents/Bulletins/APN064.pdf.
30 The theorem itself is just an equation, linking the probability of a hypothesis, given some observed pieces of evidence, and the probability of that evidence, given the hypothesis. A more comprehensive introductory overview can be found at https://arbital.com/p/bayes_rule/?l=1zq.