The Future of Everything: The Science of Prediction

by David Orrell

  8. Galton 1886. Only a distribution is given, so I randomly generated individual samples that are consistent with the original data.

  9. Galton 1888. For the correlation to equal the slope of the line, the standard deviation of the scatter in the two data sets should be the same, so a small correction is required here.

  10. Human bones from the Classic Maya period, for example, show a divergence in height between the poor and wealthy classes, caused by increasing inequality in the society. Wright 2004, p. 100.

  11. Galton 1878. Quoted from Gillham 2001, p. 217.

  12. Galton 1891.

  13. Galton 1865.

  14. Quoted from Gillham, p. 207.

  15. Plato’s Republic, book 5, line 1929. Written 360 B.C. Translated by Benjamin Jowett. Available online at http://classics.mit.edu/.

  16. Ibid., line 1961.

  17. Ibid., line 1959.

  18. Churchill predicted that “the breeding of human beings and the shaping of human nature” would be common by 1982. Churchill 1932. In Alberta, almost 3000 forced sterilizations were carried out between 1928 and 1972, under the Sexual Sterilization Act. Since then, the province has made over $800 million in compensation payments. Van Kampen 2005, p. 41.

  19. Quoted from Ridley 2003, p. 185.

  20. In plants, the egg cells are fertilized by pollen. Pea plants are self-fertilizing, so to control this, Mendel occasionally “castrated” his plants by cutting off the pollen-producing stamens.

  21. Mendel’s theory was so little known that it was independently rediscovered in 1900 by three separate people—Hugo de Vries, Carl Correns, and Erich von Tschermak—none of whom was aware of his work.

  22. Morgan et al. 1915.

  23. Matthews 2000. Tjio and Levan (1956) were the investigators who found that the total number of human chromosomes was forty-six.Painter had used the method of sectioning cells. During sectioning, chromosomes can be cut in half and then identified on an adjacent section as a separate chromosome.

  24. Schrödinger 1944.

  25. Watson and Crick 1953.

  26. Maddox 2002.

  27. Nirenberg 1963. The correspondence between the genetic code and the resulting sequence of amino acids in the protein is similar to that between the digital representation of letters in a computer and the resulting text.

  Computers work in 0s and 1s, or base 2. In genes, each letter is specified by a three-digit string in base 4 (denoted A, C, G, T rather than 0, 1, 2, 3), which can represent up to sixty-four different amino acids. Since there are only twenty such acids, there is some redundancy in the genetic code—more than one string can code for the same thing.

  28. Shapiro 1999.

  29. Dawkins 1976.

  30. As the biotech CEO Michael West put it, “The dream of biologists is to have the sequence of DNA, the programming code of life, and to be able to edit it the way you can a document on a word processor.” Quoted in McKibben 2003, p. 12.

  31. The sequence was complete in 1995, but publication did not occur until the next year. The yeast genome was divided up among about a hundred laboratories around the world.

  32. Blair’s comment is strange, since during the 1997 campaign, when asked his chances of victory, he replied: “I have never been in the business of making predictions, I am not now, and I never will be in that business.” Ormerod 2000, p. 79.

  33. The International Human Genome Sequencing Consortium 2001.

  34. Hammock and Young 2005. See also Mattick 2005.

  35. In fact, the difference between apes and humans may have less to do with genes, per se, than with the rate at which our bodies develop.

  Adult humans resemble juvenile or even fetal apes. Our slower pace of development gives our brains greater size, more plasticity, and the ability to learn. Gould 1977, p. 365.

  36. Weston and Hood 2004.

  37. Van Kampen 2005, p. 85.

  38. Van Kampen 2005, pp. 208–17. Myriad offers tests for a number of conditions, including hereditary melanoma: see www.myriad.com. They were assisted in their search for the BRCA genes by the detailed genealogical tables maintained by local Mormon families. In the Canadian province of British Columbia, the patent is effectively ignored. See http://www.bccancer.bc.ca.

  39. Gonzalez et al. 2005.

  40. Eto et al. 1989. APOE is also associated with the development of late-onset Alzheimer’s disease.

  41. The International HapMap Consortium is searching for sets of genes that tend to vary together in the human population, and may be related to the development of certain diseases. For a discussion, see Goldstein and Cavalleri 2005.

  42. Willich et al. 2006.

  43. A famous study of British civil servants showed that the strongest predictor for heart disease was job status, apparently because low-status workers had less control over their jobs and experienced more stress.Marmot et al. 1991.

  44. For a study of the relationship between longevity and social networks, see Giles et al. 2005.

  45. Alter and Eny 2005.

  46. Kollar and Fisher 1980.

  47. Yamamoto 1985; Glaser et al. 1987.

  48. Manning 2002.

  49. Ackerman 2001, p. 111.

  50. Not quite that simple for people with an “intersex” condition. See Dennis 2004.

  51. The Committee for the Prevention of Jewish Genetic Disease in the U.S. now routinely checks schoolchildren’s blood for the genetic mutations that cause Mendelian diseases like Tay-Sachs (which destroys the nervous system and usually kills by the age of five). The dangerous form of this gene is nine times more prevalent in Ashkenazi Jews than in the general population. If a child is born with two copies, he develops the disease, so carriers of the gene want to avoid marrying other carriers.See Van Kampen 2005 for a discussion of population genetics programs.

  52. Moore 2002, p. 208. See also p. 3.

  53. Part of the problem is that in a large number of studies with different groups of people, only those that show a positive correlation get reported. If there are enough studies, at least one is bound to show a positive result. See also Haga et al. 2003.

  54. It is important to separate inheritance from genetics. The former often includes many socio-economic factors—we usually inherit a family and social structure as well as genes. An embryo’s development is also influenced by non-genetic factors, such as the environment in the womb, as discussed in the text. Finally, as we will discuss later, it is possible for traits to be the result of certain genes, without being able to compute that trait from a knowledge of the DNA alone. Note that some DNA, known as mitochondrial DNA, is inherited from the mother, while in men the Y-chromosome is inherited from the father. This means that certain traits may be inherited from the mother, others from the father. For example, the mitochondrial DNA seems to be associated with aging. See Martin and Loeb 2004.

  55. See Hubbard and Wald 1999.

  56. See http://mededu.med.uottawa.ca/oisb/media/pic_news_article.jpg. One such program is the Virtual Cell project at Harvard: http://vcp.med.harvard.edu/research.html. See also Gibbs 2001.

  57. The sensitivity of DNA to its immediate cellular environment is apparent in cloning experiments, such as those that produced the ewe Dolly in 1996 and are now done on a fairly routine basis with mice and other animals. In these experiments, a nucleus from the animal to be cloned is injected into an egg cell whose own nucleus has been removed. The resulting cell, placed in the womb of a third animal, therefore contains both the DNA and a large number of proteins. The variability in these proteins, which control among other things the DNA-reading process, is what makes cloning difficult. Dolly, for example, was the single survivor of 277 cloned embryos. It also means that animals cloned from the same DNA are not perfect copies of each other. The closest thing to this are identical twins, who split after conception from the same fertilized egg cell. Owing to small random differences in their development, even they are not really identical.

  58. Heath et al. 2003.

  59. The figur
e was obtained using the open source program Cytoscape, with the sample yeast data. See http://www.cytoscape.org.

  60. Ramsey et al. 2005 presents one computational tool. As discussed in that paper, it is also possible to obtain a quick estimate of the stochastic variability without doing the detailed calculations.

  61. A result is that more than one set of values for the parameters can give the same model output. This problem is termed nonuniqueness or underdetermination. See Oreskes et al. 1994.

  62. Lewontin 2001, p. 128. See also Edelman 1987.

  63. Saunders et al. 1998.

  64. Zhang et al. 2005.

  65. The Institute for Systems Biology was set up by the biologist and inventor Leroy Hood, whose company, Applied Biosystems, supplied the “fluorescent” sequencing technology that helped make the Human Genome Project feasible. Information about the ISB is available at http://www.systemsbiology.org.

  66. Over 1,500 papers have been written on the galactose network, dating back to the 1950s. See, for example, Verma et al. 2003, Acar et al. 2005. Our model is described in De Atauri et al. 2004. Pedro de Atauri is a biologist, Stephen Ramsey was the astrophysicist, and I was the resident mathematician. The model was run on a forty-six-node Intel dual-processor cluster computer (IBM). The theoretical analysis of the feedback loops is given in Orrell et al. 2006. The stochastic noise generated by a biological system, which results in the spread of responses, can be estimated using the same drift techniques used for model error in weather forecasting. The stochastic perturbations play the same role as the random atmospheric effects that create weather-prediction errors.

  The estimation technique allows us to understand how noise is controlled. Details are in Orrell and Bolouri 2004; Orrell et al. 2005.

  67. The laboratory experiments were performed by J. J. Smith, M. Marelli, and T. W. Petersen in J. Aitchison’s group at the ISB. See Ramsey et al. 2006.

  68. The model reactions were only parameterizations of complex processes within the cell. The parameters were underdetermined by the experimental data, but at the same time, they had to be carefully balanced to make the model work. Small random perturbations to parameters would upset this balance, even if the perturbations were far smaller than the uncertainty in the parameters themselves. The sensitivity is a natural result of the combination of feedback loops. Another example is the Daisyworld system in Appendix III, which is robust to external perturbations but sensitive to internal modelling errors. It follows from these that a model of a biological or an ecological system doesn’t always need to be robust against random changes in each of its kinetic parameters (at least in terms of exact “point” predictions), and conversely that parameter sensitivity in the model does not necessarily imply sensitivity in the true system. Of course, in nature individual cells often vary significantly, but still manage to function (Morohashi et al. 2002). Similarly, different versions of Daisyworld could be constructed, each of which is homeostatic, but also sensitive to changes in the parameters.

  69. As Antoine Danchin wrote, “With the evolution of ever more complex animals, numerous regulatory mechanisms have appeared. Gradually they have combined to ensure the stability of the environment surrounding the cells within the organism, rather than leaving them subject to major environmental changes. These protection mechanisms seem to be necessary because the enzymes that catalyze the cells’ chemical reactions are highly sensitive to their physicochemical environment. This sensitivity is an effective selection procedure that explains how the organism as a whole can normally maintain the local environment of its organs and cells within narrow limits.” Danchin 2002, p. 312. The models’ sensitivity to parameters is like the mirror image of the organism’s sensitivity to its external environment—the more sensitive and fine-tuned the control system, the more easily a model of it is disturbed by internal model error. The human eye exists in many different variations and is in many ways robust, but this does not mean you would trust a doctor to randomly alter some aspect of its function in your own body. In each person a delicate balance is maintained, in what Heraclitus described as “a backward-turning adjustment” or “a harmony of opposite tensions” (depending on translation, see note below).

  70. Heraclitus, fragment 51. An alternative translation reads: “by being at variance it agrees with itself, a backward-turning adjustment like that of the bow or the lyre.” Guthrie 1968, p. 439. Heraclitus rejected the Pythagorean division of phenomena into two classes, one good, the other evil, and argued instead that “Good and evil are one.” Cornford 1969, p. 84.

  71. Danchin 2002, p. 109.

  72. As Evelyn Fox Keller wrote, the gene “is itself part and parcel of processes defined and brought into existence by the action of a complex self-regulating dynamical system in which, and for which, the inherited DNA provides the crucial and absolutely indispensable raw material, but no more than that.” Keller 2000, p. 71.

  73. Slonim et al. 2000. One problem is that cancer cells are constantly mutating, so they become resistant to chemotherapies. Statistical models can be helpful in understanding this process.

  74. One example of this approach is Genentech’s drug Herceptin, developed to target the 25 percent or so of breast cancer cases where the HER2 gene is over-expressed in the tumour. Individual patients also vary greatly in their response to other compounds, such as antipsychotic drugs, and this may be correlated to genetic differences. Goldstein and Cavalleri 2005. As another example, the enzyme denoted EGFR, which stimulates cell division, is over-expressed in about 80 percent of lung cancers. Drugs exist to block its activity, but their effectiveness depends strongly on the patient, and is highest in the 10 percent of patients who have a particular form of EGFR. Anonymous 2002, 2004.

  75. For example, Sciona in Boulder, Colorado, will sell you a “DNA, diet and lifestyle assessment test” which assesses your health risks based on a questionnaire and a DNA sample from a cheek swab (http://www.sciona.com/). Their test was withdrawn from stores in the U.K. in 2002 after Genewatch UK protested that it was misleading (www.genewatch.org). Another company working in this area to develop tests is Craig Venter’s Celera Genomics. See also the Center of Excellence for Nutritional Genomics, University of California at Davis, http://nutrigenomics.ucdavis.edu/. The potential for new medicines is a subject of debate. Sir David Weatherall of the Royal Society says “Personalised medicines show promise but they have undoubtedly been over-hyped” (Anonymous 2005g). Cancer treatment is one area that has so far shown real benefits.

  76. Quoted in Horgan 1996, p. 125. Gould was diagnosed with abdominal mesothelioma, a rare and serious form of cancer, in July 1982.

  Median mortality was eight months after diagnosis. He died twenty years later, of another form of cancer. He is the author of many popular books, including The Structure of Evolutionary Theory (2002). Another example of how biotechnology can be used to fight cancer is that new diagnostic tests, based on a blood sample, can detect some forms of mesothelioma at an early stage (Cullen 2005).

  77. Quoted in Abraham 2005. The Economist magazine, for example, predicts that we will soon have “knowledge about how genes interact with each other to generate different personal characteristics. Eventually, it might be possible to engineer a true ‘designer baby’—one whose adult looks, and possibly mental characteristics, were chosen by its parents . . .” (Anonymous 2000). As the science writer John Horgan put it, “Designer-baby predictions . . . are laughable, given the track record of behavioral genetics.” Horgan 2004.

  78. Quoted in Koestler 1968, p. 245.

  79. Haga et al. 2003.

  80. Quoted in Lewontin 2001, p. 207.

  81. Biological reductionism is, of course, still around. In the early 1990s, the U.S. government launched the Federal Violence Initiative, which aimed to detect potentially violent people by testing levels of the neurotransmitter serotonin. People with low serotonin are somewhat more likely to show violent or impulsive behaviour, and somewhat more likely to die from accidents, murder, or suicide
. Drugs such as Prozac are believed to work by affecting serotonin levels in the brain. But serotonin levels are also affected by one’s perceived societal status: a decrease in rank tends to lower them. It would make as much sense to stamp out crime by arresting anyone who looks a bit down. The initiative was fortunately squashed in 1992, after its lead organizer made some remarks comparing youths in America’s ghettos with monkeys in the jungle. Moore 2002, p. 226.

  82. Quoted in Bernstein 1998, p. 216.

  6 ⊳ BULLS AND BEARS

  PREDICTING THE ECONOMY

  1. Mackay 1852.

  2. Jonathan Swift, The South Sea Project, 1721.

  3. Freese 2003, p. 40.

  4. Quoted in Bernstein 1998, p. 124.

  5. Ibid., p. 141.

  6. Ibid., p. 160. Quetelet demonstrated the use of the normal distribution to fit human data by using published data on the chest measurements of 5,738 Scottish soldiers.

  7. Orléan 2001.

  8. Any such valuation of environmental services will of course be highly subjective and political, but it’s better than valuing them at zero, which has led, for example, to the loss of rainforest to agricultural production. In the Brazilian Amazon, about 25,000 square kilometres is lost per year. Nearly half of that occurs in Mato Grosso state, where the world’s largest soy farms are located. Anonymous 2005; Anonymous 2005a.

  9. I used to pass Jeremy Bentham in the hallway from time to time on my way to teach classes at University College London. He died in 1832, but one of the terms of his will was that his mummified body would be displayed in a glass case, his head replaced with a wax copy. Perhaps he saw it as a way of maximizing his own utility, even after his death.

  10. Jevons 1888. See also McCusker 2004, Nicholls 1998. For weather forecasting based on sunspots, see http://www.weatheraction.com/.

  11. Galilei 1967, p. 59.

  12. Smith 1776.

  13. McMurdy 2004.

  14. Bachelier 1900.

  15. Quoted in Bernstein 1998, p. 200.

  16. In Cootner 1964. See also Roberts 1959.