by Hannah Fry
27 State Council Guiding Opinions Concerning Establishing and Perfecting Incentives for Promise-keeping and Joint Punishment Systems for Trust-breaking, and Accelerating the Construction of Social Sincerity, China Copyright and Media, 30 May 2016, updated 18 Oct. 2016, https://chinacopyrightandmedia.wordpress.com/2016/05/30/state-council-guiding-opinions-concerning-establishing-and-perfecting-incentives-for-promise-keeping-and-joint-punishment-systems-for-trust-breaking-and-accelerating-the-construction-of-social-sincer/.
28 Rachel Botsman, Who Can You Trust? How Technology Brought Us Together – and Why It Could Drive Us Apart (London: Penguin, 2017), Kindle edn, p. 151.
Justice
1 John-Paul Ford Rojas, ‘London riots: Lidl water thief jailed for six months’, Telegraph, 7 Jan. 2018, http://www.telegraph.co.uk/news/uknews/crime/8695988/London-riots-Lidl-water-thief-jailed-for-six-months.html.
2 Matthew Taylor, ‘London riots: how a peaceful festival in Brixton turned into a looting free-for-all’, Guardian, 8 Aug. 2011, https://www.theguardian.com/uk/2011/aug/08/london-riots-festival-brixton-looting.
3 Rojas, ‘London riots’.
4 Josh Halliday, ‘London riots: how BlackBerry Messenger played a key role’, Guardian, 8 Aug. 2011, https://www.theguardian.com/media/2011/aug/08/london-riots-facebook-twitter-blackberry.
5 David Mills, ‘Paul and Richard Johnson avoid prison over riots’, News Shopper, 13 Jan. 2012, http://www.newsshopper.co.uk/londonriots/9471288.Father_and_son_avoid_prison_over_riots/.
6 Ibid.
7 Rojas, ‘London riots’. ‘Normally, the police wouldn’t arrest you for such an offence. They wouldn’t hold you in custody. They wouldn’t take you to court,’ Hannah Quirk, a senior lecturer in criminal law and justice at Manchester University wrote about Nicholas’s case in 2015: Carly Lightowlers and Hannah Quirk, ‘The 2011 English “riots”: prosecutorial zeal and judicial abandon’, British Journal of Criminology, vol. 55, no. 1, 2015, pp. 65–85.
8 Mills, ‘Paul and Richard Johnson avoid prison over riots’.
9 William Austin and Thomas A. Williams III, ‘A survey of judges’ responses to simulated legal cases: research note on sentencing disparity’, Journal of Criminal Law and Criminology, vol. 68, no. 2, 1977, pp. 306–310.
10 Mandeep K. Dhami and Peter Ayton, ‘Bailing and jailing the fast and frugal way’, Journal of Behavioral Decision-making, vol. 14, no. 2, 2001, pp. 141–68, http://onlinelibrary.wiley.com/doi/10.1002/bdm.371/abstract.
11 Up to half of the judges differed in their opinion of the best course of action on any one case.
12 Statisticians have a way to measure this kind of consistency in judgments, called Cohen’s Kappa. It takes into account the fact that – even if you were just wildly guessing – you could end up being consistent by chance. A score of one means perfect consistency. A score of zero means you’re doing no better than random. The judges’ scores ranged from zero to one and averaged 0.69.
13 Diane Machin, ‘Sentencing guidelines around the world’, paper prepared for Scottish Sentencing Council, May 2005, https://www.scottishsentencingcouncil.org.uk/media/1109/paper-31a-sentencing-guidelines-around-the-world.pdf.
14 Ibid.
15 Ibid.
16 Ernest W. Burgess, ‘Factors determining success or failure on parole’, in The Workings of the Intermediate-sentence Law and Parole System in Illinois (Springfield, IL: State Board of Parole, 1928). It’s a tricky paper to track down, so here is an alternative read by Burgess’s colleague Tibbitts, on the follow-up study to the original: Clark Tibbitts, ‘Success or failure on parole can be predicted: a study of the records of 3,000 youths paroled from the Illinois State Reformatory’, Journal of Criminal Law and Criminology, vol. 22, no. 1, Spring 1931, https://scholarlycommons.law.northwestern.edu/cgi/viewcontent.cgi?article=2211&context=jclc. The other categories used by Burgess were ‘black sheep’, ‘criminal by accident’, ‘dope’ and ‘gangster’. ‘Farm boys’ were the category he found least likely to re-offend.
17 Karl F. Schuessler, ‘Parole prediction: its history and status’, Journal of Criminal Law and Criminology, vol. 45, no. 4, 1955, pp. 425–31, https://pdfs.semanticscholar.org/4cd2/31dd25321a0c14a9358a93ebccb6f15d3169.pdf.
18 Ibid.
19 Bernard E. Harcourt, Against Prediction: Profiling, Policing, and Punishing in an Actuarial Age (Chicago and London: University of Chicago Press, 2007), p. 1.
20 Philip Howard, Brian Francis, Keith Soothill and Les Humphreys, OGRS 3: The Revised Offender Group Reconviction Scale, Research Summary 7/09 (London: Ministry of Justice, 2009), https://core.ac.uk/download/pdf/1556521.pdf.
21 A slight caveat here: there probably is some selection bias in this statistic. ‘Ask the audience’ was typically used in the early rounds of the game, when the questions were a lot easier. None the less, the idea of the collective opinions of a group being more accurate than those of any individual is a well-documented phenomenon. For more on this, see James Surowiecki, The Wisdom of Crowds: Why the Many Are Smarter than the Few (New York: Doubleday, 2004), p. 4.
22 Netflix Technology Blog, https://medium.com/netflix-techblog/netflix-recommendations-beyond-the-5-stars-part-2-d9b96aa399f5.
23 Shih-ho Cheng, ‘Unboxing the random forest classifier: the threshold distributions’, Airbnb Engineering and Data Science, https://medium.com/airbnb-engineering/unboxing-the-random-forest-classifier-the-threshold-distributions-22ea2bb58ea6.
24 Jon Kleinberg, Himabindu Lakkaraju, Jure Leskovec, Jens Ludwig and Sendhil Mullainathan, Human Decisions and Machine Predictions, NBER Working Paper no. 23180 (Cambridge, MA: National Bureau of Economic Research, Feb. 2017), http://www.nber.org/papers/w23180. This study actually used ‘gradient boosted decision trees’, an algorithm similar to random forests. Both combine the predictions of lots of decision trees to arrive at a decision, but the trees in the gradient-boosted method are grown sequentially, while in random forests they are grown in parallel. To set up this study, the dataset was first chopped in half. One half was used to train the algorithm, the other half was kept to one side. Once the algorithm was ready, it took cases from the half that it had never seen before to try to predict what would happen. (Without splitting the data first, your algorithm would just be a fancy look-up table).
25 Academics have spent time developing statistical techniques to deal with precisely this issue, so that you can still make a meaningful comparison between the respective predictions made by judges and algorithms. For more details on this, see Kleinberg et al., Human Decisions and Machine Predictions.
26 ‘Costs per place and costs per prisoner by individual prison’, National Offender Management Service Annual Report and Accounts 2015–16, Management Information Addendum, Ministry of Justice information release, 27 Oct. 2016, https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/563326/costs-per-place-cost-per-prisoner-2015-16.pdf.
27 Marc Santora, ‘City’s annual cost per inmate is $168,000, study finds’, New York Times, 23 Aug. 2013, http://www.nytimes.com/2013/08/24/nyregion/citys-annual-cost-per-inmate-is-nearly-168000-study-says.html; Harvard University, ‘Harvard at a glance’, https://www.harvard.edu/about-harvard/harvard-glance.
28 Luke Dormehl, The Formula: How Algorithms Solve All Our Problems … and Create More (London: W. H. Allen, 2014), p. 123.
29 Julia Angwin, Jeff Larson, Surya Mattu and Lauren Kirchner, ‘Machine bias’, ProPublica, 23 May 2016, https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing.
30 ‘Risk assessment’ questionnaire, https://www.documentcloud.org/documents/2702103-Sample-Risk-Assessment-COMPAS-CORE.html.
31 Tim Brennan, William Dieterich and Beate Ehret (Northpointe Institute), ‘Evaluating the predictive validity of the COMPAS risk and needs assessment system’, Criminal Justice and Behavior, vol. 36, no. 1, 2009, pp. 21–40, http://www.northpointeinc.com/files/publications/Criminal-Justice-Behavior-COMPAS.pdf. According to a 2018 study, the COMPAS algorithm has a similar accuracy to an
‘ensemble’ of humans. The researchers demonstrated that asking a group of 20 inexperienced individuals to predict recidivism achieved an equivalent score to the COMPAS system. It’s an interesting comparison, but it’s worth remembering that in real courts, judges don’t have a team of strangers making votes in the back room. They’re on their own. And that’s the only comparison that really counts. See Julia Dressel and Hany Farid, ‘The accuracy, fairness, and limits of predicting recidivism’, Science Advances, vol. 4, no. 1, 2018.
32 Christopher Drew Brooks v. Commonwealth, Court of Appeals of Virginia, Memorandum Opinion by Judge Rudolph Bumgardner III, 28 Jan. 2004, https://law.justia.com/cases/virginia/court-of-appeals-unpublished/2004/2540023.html.
33 ‘ACLU brief challenges constitutionality of Virginia’s sex offender risk assessment guidelines’, American Civil Liberties Union Virginia, 28 Oct. 2003, https://acluva.org/en/press-releases/aclu-brief-challenges-constitutionality-virginias-sex-offender-risk-assessment.
34 State v. Loomis, Supreme Court of Wisconsin,13 July 2016, http://caselaw.findlaw.com/wi-supreme-court/1742124.html.
35 Quotations from Richard Berk are from personal communication.
36 Angwin et al., ‘Machine bias’.
37 Global Study on Homicide 2013 (Vienna: United Nations Office on Drugs and Crime, 2014), http://www.unodc.org/documents/gsh/pdfs/2014_GLOBAL_HOMICIDE_BOOK_web.pdf.
38ACLU, ‘The war on marijuana in black and white’, June 2013, www.aclu.org/files/assets/aclu-thewaronmarijuana-ve12.pdf
39 Surprisingly, perhaps, Equivant’s stance on this is backed up by the Supreme Court of Wisconsin. After Eric Loomis was sentenced to prison for six years by a judge using the COMPAS risk-assessment tool, he appealed the ruling. The case, Loomis v. Wisconsin, claimed that the use of proprietary, closed-source risk-assessment software to determine his sentence violated his right to due process, because the defence can’t challenge the scientific validity of the score. But the Wisconsin Supreme Court ruled that a trial court’s use of an algorithmic risk assessment in sentencing did not violate the defendant’s due process rights: Supreme Court of Wisconsin, case no. 2015AP157-CR, opinion filed 13 July 2016, https://www.wicourts.gov/sc/opinion/DisplayDocument.pdf?content=pdf&seqNo=171690.
40 Lucy Ward, ‘Why are there so few female maths professors in universities?’, Guardian, 11 March 2013, https://www.theguardian.com/lifeandstyle/the-womens-blog-with-jane-martinson/2013/mar/11/women-maths-professors-uk-universities.
41 Sonja B. Starr and M. Marit Rehavi, Racial Disparity in Federal Criminal Charging and Its Sentencing Consequences, Program in Law and Economics Working Paper no. 12-002 (Ann Arbor: University of Michigan Law School, 7 May 2012), http://economics.ubc.ca/files/2013/05/pdf_paper_marit-rehavi-racial-disparity-federal-criminal.pdf.
42 David Arnold, Will Dobbie and Crystal S. Yang, Racial Bias in Bail Decisions, NBER Working Paper no. 23421 (Cambridge, MA: National Bureau of Economic Research, 2017), https://www.princeton.edu/~wdobbie/files/racialbias.pdf.
43 John J. Donohue III, Capital Punishment in Connecticut, 1973–2007: A Comprehensive Evaluation from 4686 Murders to One Execution (Stanford, CA, and Cambridge, MA: Stanford Law School and National Bureau of Economic Research, Oct. 2011), https://law.stanford.edu/wp-content/uploads/sites/default/files/publication/259986/doc/slspublic/fulltext.pdf.
44 Adam Benforado, Unfair: The New Science of Criminal Injustice (New York: Crown, 2015), p. 197.
45 Sonja B. Starr, Estimating Gender Disparities in Federal Criminal Cases, University of Michigan Law and Economics Research Paper no. 12-018 (Ann Arbor: University of Michigan Law School, 29 Aug. 2012), https://ssrn.com/abstract=2144002 or http://dx.doi.org/10.2139/ssrn.2144002.
46 David B. Mustard, ‘Racial, ethnic, and gender disparities in sentencing: evidence from the US federal courts’, Journal of Law and Economics, vol. 44, no. 2, April 2001, pp. 285–314, http://people.terry.uga.edu/mustard/sentencing.pdf.
47 Daniel Kahneman, Thinking, Fast and Slow (New York: Farrar, Straus & Giroux, 2011), p. 44.
48 Chris Guthrie, Jeffrey J. Rachlinski and Andrew J. Wistrich, Blinking on the Bench: How Judges Decide Cases, paper no. 917 (New York: Cornell University Law Faculty, 2007), http://scholarship.law.cornell.edu/facpub/917.
49 Kahneman, Thinking, Fast and Slow, p. 13.
50 Ibid., p. 415.
51 Dhami and Ayton, ‘Bailing and jailing the fast and frugal way’.
52 Brian Wansink, Robert J. Kent and Stephen J. Hoch, ‘An anchoring and adjustment model of purchase quantity decisions’, Journal of Marketing Research, vol. 35, 1998, pp. 71–81, http://foodpsychology.cornell.edu/sites/default/files/unmanaged_files/Anchoring-JMR-1998.pdf.
53 Mollie Marti and Roselle Wissler, ‘Be careful what you ask for: the effect of anchors on personal injury damages awards’, Journal of Experimental Psychology: Applied, vol. 6, no. 2, 2000, pp. 91–103.
54 Birte Englich and Thomas Mussweiler, ‘Sentencing under uncertainty: anchoring effects in the courtroom’, Journal of Applied Social Psychology, vol. 31, no. 7, 2001, pp. 1535–51, http://onlinelibrary.wiley.com/doi/10.1111/j.1559-1816.2001.tb02687.x.
55 Birte Englich, Thomas Mussweiler and Fritz Strack, ‘Playing dice with criminal sentences: the influence of irrelevant anchors on experts’ judicial decision making’, Personality and Social Psychology Bulletin, vol. 32, 2006, pp. 188–200, https://www.researchgate.net/publication/7389517_Playing_Dice_With_Criminal_Sentences_The_Influence_of_Irrelevant_Anchors_on_Experts%27_Judicial_Decision_Making?enrichId=rgreq-f2fedfeb71aa83f8fad80cc24df3254d-XXX&enrichSource=Y292ZXJQYWdlOzczODk1MTc7QVM6MTAzODIzNjIwMTgyMDIyQDE0MDE3NjQ4ODgzMTA%3D&el=1_x_3&_esc=publicationCoverPdf.
56 Ibid.
57 Ibid.
58 Mandeep K. Dhami, Ian K. Belton, Elizabeth Merrall, Andrew McGrath and Sheila Bird, ‘Sentencing in doses: is individualized justice a myth?’, under review. Kindly shared through personal communication with Mandeep Dhami.
59 Ibid.
60 Adam N. Glynn and Maya Sen, ‘Identifying judicial empathy: does having daughters cause judges to rule for women’s issues?’, American Journal of Political Science, vol. 59, no. 1, 2015, pp. 37–54, https://scholar.harvard.edu/files/msen/files/daughters.pdf.
61 Shai Danziger, Jonathan Levav and Liora Avnaim-Pesso, ‘Extraneous factors in judicial decisions’, Proceedings of the National Academy of Sciences of the United States of America, vol. 108, no. 17, 2011, pp. 6889–92, http://www.pnas.org/content/108/17/6889.
62 Keren Weinshall-Margel and John Shapard, ‘Overlooked factors in the analysis of parole decisions’, Proceedings of the National Academy of Sciences of the United States of America, vol. 108, no. 42, 2011, E833, http://www.pnas.org/content/108/42/E833.long.
63 Uri Simonsohn and Francesca Gino, ‘Daily horizons: evidence of narrow bracketing in judgment from 9,000 MBA-admission interviews’, Psychological Science, vol. 24, no. 2, 2013, pp. 219–24, https://ssrn.com/abstract=2070623.
64 Lawrence E. Williams and John A. Bargh, ‘Experiencing physical warmth promotes interpersonal warmth’, Science, vol. 322, no. 5901, pp. 606–607, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2737341/.
Medicine
1 Richard M. Levenson, Elizabeth A. Krupinski, Victor M. Navarro and Edward A. Wasserman. ‘Pigeons (Columba livia) as trainable observers of pathology and radiology breast cancer images’, PLOSOne, 18 Nov. 2015, http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0141357.
2 ‘Hippocrates’ daughter as a dragon kills a knight, in “The Travels of Sir John Mandeville”’, British Library Online Gallery, 26 March 2009, http://www.bl.uk/onlinegallery/onlineex/illmanus/harlmanucoll/h/011hrl000003954u00008v00.html.
3 Eleni Tsiompanou, ‘Hippocrates: timeless still’, JLL Bulletin: Commentaries on the History of Treatment Evaluation (Oxford and Edinburgh: James Lind Library, 2012), http://www.jameslindlibrary.org/articles/hippocrates-timeless-still/.
4 David K. Osborne, ‘Hippocrates: father of medicine’, Greek Medicine.net,
2015, http://www.greekmedicine.net/whos_who/Hippocrates.html.
5 Richard Colgan, ‘Is there room for art in evidence-based medicine?’, AMA Journal of Ethics, Virtual Mentor 13: 1, Jan. 2011, pp. 52–4, http://journalofethics.ama-assn.org/2011/01/msoc1-1101.html.
6 Joseph Needham, Science and Civilization in China, vol. 6, Biology and Biological Technology, part VI, Medicine, ed. Nathan Sivin (Cambridge: Cambridge University Press, 2004), p. 143, https://monoskop.org/images/1/16/Needham_Joseph_Science_and_Civilisation_in_China_Vol_6-6_Biology_and_Biological_Technology_Medicine.pdf.
7 ‘Ignaz Semmelweis’, Brought to Life: Exploring the History of Medicine (London: Science Museum n.d.), http://broughttolife.sciencemuseum.org.uk/broughttolife/people/ignazsemmelweis.
8 Quotations from Andy Beck are from personal communication.
9 Joann G. Elmore, Gary M. Longton, Patricia A. Carney, Berta M. Geller, Tracy Onega, Anna N. A. Tosteson, Heidi D. Nelson, Margaret S. Pepe, Kimberly H. Allison, Stuart J. Schnitt, Frances P. O’Malley and Donald L. Weaver, ‘Diagnostic concordance among pathologists interpreting breast biopsy specimens’, Journal of the American Medical Association, vol. 313, no. 11, 17 March 2015, 1122–32, https://jamanetwork.com/journals/jama/fullarticle/2203798.
10 Ibid.
11 The name ‘neural network’ came about as an analogy with what happens in the brain. There, billions of neurons are connected to one another in a gigantic network. Each neuron listens to its connections and sends out a signal whenever it picks up on another neuron being excited. The signal then excites some other neurons that are listening to it.
A neural network is a much simpler and more orderly version of the brain. Its (artificial) neurons are structured in layers, and all the neurons in each layer listen to all the neurons in the previous layer. In our dog example the very first layer is the individual pixels in the image. Then there are several layers with thousands of neurons in them, and a final layer with only a single neuron in it that outputs the probability that the image fed in is a dog.