NOTES
AI
1Neil Postman (1992) Technopoly: The Surrender of Culture to Technology, New York, Knopf.
2See tinyurl.com/PostmanFiveLessons.
4Norbert Wiener (1950) The Human Use of Human Beings: Cybernetics and Society, New York, Houghton Mifflin.
5To watch the BBC video where this close relationship between Facebook and Cambridge Analytica is revealed, see tinyurl.com/FacebookLiedTheyKnew.
6See tinyurl.com/ZuckerbergCongress.
7See tinyurl.com/FacebookSwingsElections.
The People
2One year, I asked my travel agent to book me on a research trip from my home in Cork, to a conference in Cuba, via a meeting in Cyprus, and on to a summer school in Canberra, before returning home to Cork. My travel agent joked I was just picking places from the atlas that begin with the letter C.
3Antarctica’s emperor penguin population has been counted using machine-learning tools trained on images from Oxford University’s Penguin Watch Project. And while I don’t believe any AI conference has been held in Antarctica (yet), a technology conference called AntarctiConf 2020 was held on board a 16-day cruise to Antarctica in January 2020, just before the COVID-19 pandemic closed the world down.
4Actually 282,589,933 –1 is the largest known prime number. It would have 24,862,048 digits when written out in longhand, but this would, I imagine, make for a very boring book. It is also not any old prime number, but a very special type of prime number. It is a Mersenne prime number, since it is one less than a power of two. Such prime numbers are named after Marin Mersenne, a French friar who studied them in the early seventeenth century. While this is the largest prime number to have been discovered by mathematicians, there is no largest prime number. Over 2000 years ago, Euclid proved that, while prime numbers become rarer and rarer as numbers get bigger, prime numbers never end. And while we don’t know if 282,589,933 –1 is the largest Mersenne prime or not, we do know, thanks to the law of the excluded middle, that one of these statements must be true: either it is the largest Mersenne prime, or there is a larger such prime.
5In the interests of full disclosure, I am part of the problem, being one of those white males.
6Peter E. Hart, Niles J. Nilsson & Bertram Raphael (1968) ‘A Formal Basis for the Heuristic Determination of Minimum Cost Paths’, IEEE Transactions of Systems Science and Cybernetics, vol. 4, no. 2, pp. 100–07.
7See tinyurl.com/TimeMagazine100.
8David E. Rumelhart, Geoffrey E. Hinton & Ronald J. Williams (1986) ‘Learning Representations by Back-propagating Errors’, Nature, 323, pp. 533–36.
9In 2012, Andrew Ng, Jeff Dean and colleagues at Google’s secretive X Laboratory built one of the largest neural networks up to that time. They needed 16,000 cores across 1000 computers to train the network’s billion weights. And after three days of training on still images from random YouTube videos, the network famously taught itself to recognise cats. What else would you expect to come out of YouTube? See Q.V. Le et al. (2012) ‘Building High-level Features Using Large Scale Unsupervised Learning’, in Proceedings of the 29th International Conference on Machine Learning (ICML2012), Madison, WI, Omnipress, pp. 507–14.
10While Grace Hopper was working on a Mark II Computer at Harvard University in 1947, an error was uncovered, caused by a moth stuck in a relay. This incident is why programmers still talk about finding and removing ‘bugs’ from their programs.
11If you want an alternative, in my first book, It’s Alive! Artificial Intelligence from the Logic Piano to Killer Robots, I suggested we call Bengio, Hinton and LeCun the ‘three musketeers’ of deep learning.
12In It’s Alive!, I told the story of how I nearly drowned John McCarthy in Sydney Harbour, and how, as a result, I feared I would end up a sad footnote in a history of AI. This footnote is another attempt to do just that.
13All dollar amounts are in US dollars.
14In April 2020, SoftBank announced its first annual loss of over $16 billion in its $100-billion Vision Fund. The loss was in part due to investments in companies like Wag, a dog-walking service, and Zume, a robot pizza company.
15For more on Rand, see Chris Matthew Sciabarra (1995) Ayn Rand: The Russian Radical, University Park, Pennsylvania State University Press.
16You can read the Declaration of the Independence of Cyberspace in full at eff.org/cyberspace-independence.
The Companies
1Nick Bostrom (2014) Superintelligence: Paths, Dangers, Strategies, Oxford, Oxford University Press.
2In 2018, Facebook agreed with an independent report that found it had failed to prevent its platform from being used to ‘incite offline violence’ among its 18 million users in Myanmar. Thousands of people died in the civil unrest in Myanmar, and over 700,000 Rohingya were forced to flee for their lives to neighbouring Bangladesh.
3See tinyurl.com/KendallTestimony.
4Julia Angwin & Terry Parris Jr, ‘Facebook Lets Advertisers Exclude Users by Race’, ProPublica, 28 October 2016, propublica.org/article/facebook-lets-advertisers-exclude-users-by-race.
5‘There are only two industries that refer to their customers as “users”: illegal drugs and software,’ said Edward Tufte, American statistician and professor emeritus at Yale.
6Julia Angwin, Ariana Tobin & Madeleine Varner, ‘Facebook (Still) Letting Housing Advertisers Exclude Users by Race’, ProPublica, 21 November 2017, propublica.org/article/facebook-advertising-discrimination-housing-race-sex-national-origin.
7Jessica Livingstone (2007) Founders at Work: Stories of Startups’ Early Days, New York, Springer-Verlag.
8See https://ai.google/principles.
9See ibm.com/blogs/policy/trust-principles.
10Of course, shortly after I wrote this chapter, Lemonade appeared to have committed a serious ethical blunder. They are currently facing a class-action lawsuit for using biometric data without consent with facial-recognition software that analyses videos submitted in the claims process.
Autonomy
1My colleague Professor Genevieve Bell of ANU has whimsically asked: ‘Who is the self driving a self-driving car?’
2There are many flavours of Uber: UberX, UberPool, UberEats … By 2050, all of them will be autonomous except for one, UberChauffeur. This will be a premium service where you pay extra for a person who wears a smart uniform with a cap to sit in the driver’s seat, open the car door and carry your luggage to the kerb. They won’t drive; the Uber vehicle will still be self-driving. But you will pay extra, as in other parts of your life, for personal service that connects you with a real human being.
3See tinyurl.com/aitrends2017.
4When I started writing this chapter, Tesla had a market cap of just $400 billion. By the time I had finished the book, this had nearly doubled. I suspect that, by the time you read this, it may have doubled again.
5While Level 5 is the highest level of autonomy in the Society of Automotive Engineers’ classification scheme, there are actually six levels in total. Level 0 is, however, no automation at all.
6See tinyurl.com/ElonMuskLidar.
7See tinyurl.com/GermanEthicsCode.
8I don’t mean they literally paint the road green with a paintbrush. But if you watch a demonstration of a self-driving car on YouTube, you will often see that the video image of the open road in front of the car is shaded green.
9Edmond Awad et al. (2020) ‘Crowdsourcing Moral Machines’, Communications of the ACM, vol. 63, no. 3, pp. 48–55.
10Edmond Awad et al. (2020) ‘Universals and Variations in Moral Decisions Made in 42 Countries by 70,000 Participants’, Proceedings of the National Academy of Sciences, vol. 117, no. 5, pp. 2332–37.
11See tinyurl.com/KillerRobotPoll.
12Poseidon is a troubling name for an autonomous nuclear weapon. Poseidon is the god of the sea, earthquakes, storms and horses. He is one of the most bad-tempered, moody and greedy Olympian gods.
13See tinyurl.com/DronePapers.
14If you agree with me and the thousands of other AI researchers who have signed the open letter to the United Nations, you can sign it too: visit https://futureoflife.org/open-letter-autonomous-weapons.
15An experimental study of facial-recognition software from 16 major commercial suppliers in 2017 by the National Institute of Standards and Technology in the United States found false positive rates of over 90 per cent in some circumstances; see tinyurl.com/NISTReport2017.
16See tinyurl.com/DoDEthicalPrinciples.
Humans v. Machines
1A petabyte is 1015 bytes of information. See tinyurl.com/BrainVInternet.
2Geoffrey Jefferson (1949) ‘The Mind of Mechanical Man’, British Medical Journal, vol. 1, p. 1105.
3See Daniel Kahneman (2011) Thinking, Fast and Slow, New York, Farrar, Straus & Giroux.
4Article 13 of the Universal Declaration of Human Rights states:
(1) Everyone has the right to freedom of movement and residence within the borders of each state.
(2) Everyone has the right to leave any country, including his own, and to return to his country.
This universal right to freedom of movement has been somewhat ignored during the COVID-19 pandemic.
5Boston Dynamic’s Atlas robot was also able to open a door using the door handle, putting to rest the idea that you only had to hide beyond a closed door to escape a marauding robot.
7See stoprobotabuse.com.
8See tinyurl.com/EURobotRights.
9If you want to be impressed, check out the video on YouTube of the Guinness World Record for solving the Rubik’s cube: see tinyurl.com/fastestcube.
10Gerald Tesauro et al. (2013) ‘Analysis of Watson’s Strategies for Playing Jeopardy! ’, Journal of Artificial Intelligence Research, vol. 47, pp. 205–51.
Ethical Rules
1Howard Rheingold (1985) Tools for Thought, Englewood Cliff, New Jersey, Prentice Hall.
2I.J. Good (1965) ‘Speculations Concerning the First Ultraintelligent Machine’, Advances in Computers, vol. 6, pp. 31–88.
4Seneca the Younger, ‘XLVII: On Master and Slave’, Epistulae Morales ad Lucilium (Moral Letters to Lucilius).
5Asimov’s three laws were first published in the 1942 short story ‘Runaround’, which was later included in the 1950 collection I, Robot.
7See tinyurl.com/duckmeetscar.
8See tinyurl.com/newrobotlaws.
9Robin R. Murphy & David D. Woods (2009) ‘Beyond Asimov: The Three Laws of Responsible Robotics’, IEEE Intelligent Systems, vol. 24, no. 4, pp. 14–20.
10See tinyurl.com/RobotsDoNoHarm.
11See https://futureoflife.org/ai-principles.
12The precautionary principle has been used in numerous international laws and rules. It is, for example, the 15th principle of the United Nations’ 1992 Rio Declaration on Environment and Development.
14It would be unnecessarily rude to California and excessively generous to the European Union to recall that Oscar Wilde actually argued: ‘Imitation is the sincerest form of flattery that mediocrity can pay to greatness.’
15Apple potentially faces an even larger fine of around $15 billion for its legendary ‘double Irish’ tax avoidance strategy. In August 2016 the European Commission ruled that Apple benefited from illegal tax arrangements from 2003 to 2014. Not surprisingly, Apple appealed the decision. More surprisingly, the Irish government also appealed, telling the European Commission that it didn’t want to receive the $15 billion fine. Presumably, the Irish government preferred to keep multinational companies sweet with tax breaks rather than pay off around 5 per cent of Ireland’s national debt. The General Court of the European Union reversed the decision against Apple in July 2020. However, the European Commission has announced it will appeal this reversal. It looks like this legal saga still has a long way to run.
16Why are flight recorders called black boxes when they are bright orange?
17See tinyurl.com/ChinaAIPrinciples.
18Mark Twain, ‘The Work of Gutenberg’, The Hartford Daily Courant, 27 June 1900, p. 7.
19For a more detailed discussion, see Luciano Floridi & Josh Cowls (2019) ‘A Unified Framework of Five Principles for AI in Society’, Harvard Data Science Review, vol. 1.
20See tinyurl.com/7ethicaltraps.
21See tinyurl.com/DennettWired.
Fairness
2See tinyurl.com/NSWpolicetargets.
3‘COMPAS’ stands for Correctional Offender Management Profiling for Alternative Sanctions.
4See tinyurl.com/ProPublicaCOMPAS.
5Of course, there is an argument for the reverse, that crime causes broken homes, as crime often results in jail, and jail time often results in parents being separated from their children. But the COMPAS questionnaire presumes the reverse causal link: that broken homes cause crime.
6Julia Dressel & Hany Farid (2018) ‘The Accuracy, Fairness, and Limits of Predicting Recidivism’, Science Advances, vol. 4, no. 1, DOI: 10.1126/sciadv.aao5580.
7A linear classifier just tries to find simple linear relations between features and classes. For instance, it might decide that a person is high-risk if their number of priors less one-tenth of their age in years is greater than zero.
8Sahil Verma & Julia Rubin (2018) ‘Fairness Definitions Explained’, in FairWare ’18: Proceedings of the International Workshop on Software Fairness, Association for Computing Machinery, New York, pp. 1–7.
9Jon M. Kleinberg, Sendhil Mullainathan & Manish Raghavan (2017) ‘Inherent Trade-Offs in the Fair Determination of Risk Scores’, in 8th Innovations in Theoretical Computer Science Conference, ITCS 2017, 9–11 January 2017, Berkeley, pp. 43:1–43:23.
10False positives are also called ‘Type 1 errors’, while false negatives are called ‘Type 2 errors’.
11Stephen Hawking was warned that each equation in A Brief History of Time would halve the sales of his book. I’m hoping the same isn’t true for fractions.
12See tinyurl.com/PartnershipOnAIReport.
13Ronald C. Arkin (2010) ‘The Case for Ethical Autonomy in Unmanned Systems’, Journal of Military Ethics, vol. 9, no. 4, pp. 332–41.
14Allison Koenecke et al. (2020) ‘Racial Disparities in Automated Speech Recognition’, Proceedings of the National Academy of Sciences, vol. 117, no. 14, pp. 7684–89.
15The ‘word error rate’ is the percentage of words that needed to be added, deleted or changed to give the correct transcription of the speech.
16See tinyurl.com/FacialRecognitionArrest.
17S. Ahmed et al. (2021) ‘Examining the Potential Impact of Race Multiplier Utilization in Estimated Glomerular Filtration Rate Calculation on African-American Care Outcomes’, Journal of General Internal Medicine, vol. 36, pp. 464–71.
18Z. Obermeyer et al. (2019) ‘Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations’, Science, vol. 366, no. 6464, pp. 447–53.
19Michael J. Rosenfeld, Reuben J. Thomas & Sonia Hausen (2019) ‘Disintermediating Your Friends: How Online Dating in the United States Displaces Other Ways of Meeting’, Proceedings of the National Academy of Sciences, vol. 116, no. 36, pp. 17753–58.
20Uber does not classify its drivers as employees but as independent contractors. However, like the State of California, I don’t buy this. It appears to be largely an excuse to give them fewer rights and worse pay. This is not something a decent society should permit.
21See tinyurl.com/AmazonAITool.
22Richard Landers (2012) ‘Unfolding the IKEA Effect: Why We Love the Things We Build’, Journal of Consumer Psychology, vol. 22, no. 3, pp. 453–60.
23Alvin E. Roth & Elliot Peranson (1997) ‘The Effects of the Change in the NRMP Matching Algorithm’, Journal of the American Medical Association, vol. 278, no. 9, pp. 729–32.
Privacy
1‘Skynet’ is the name of the fictional and conscious AI system threatening the future of humanity in the Terminator films.
2See tinyurl.com/FaceInRockConcert.
3Vint Cerf has been called one of the ‘fathers of the internet’, but I’m trying to avoid that sexist pitfall here. I had the pleasure of interviewing him on stage once, so I can attest to the fact that he is indeed a nice guy. You can watch the interview and learn more about his many contributions to the technology in your life at tinyurl.com/VintCerfVideo.
4Samuel D. Warren & Louis D. Brandeis (1890) ‘The Right to Privacy’, Harvard Law Review, vol. 4, no. 5, pp. 193–220.
5Alan F. Westin (1967) Privacy and Freedom, New York, Atheneum, p. 487.
7This is, of course, an allusion to a famous sequence of instant messages that Mark Zuckerberg sent while still a student at Harvard.
Zuck: yea so if you ever need info about anyone at harvard
Zuck: just ask
Zuck: i have over 4000 emails, pictures, addresses, sns
Friend: what!? how’d you manage that one?
Zuck: people just submitted it
Zuck: i don’t know why
Zuck: they ‘trust me’
Zuck: dumb f—ks
8See tinyurl.com/IndiaFaceRecognition.
9See tinyurl.com/IndiaFaceRecognition2.
10See tinyurl.com/IndiaFaceRecognition3.
11See tinyurl.com/FaceRecognitionArrest.
12See tinyurl.com/FaceRecognitionFine.
13See tinyurl.com/MerkleyStatement.
14Yilun Wang & Michal Kosinski (2018) ‘Deep Neural Networks Are More Accurate than Humans at Detecting Sexual Orientation from Facial Images’, Journal of Personality and Social Psychology, vol. 114, no. 2, pp. 246–57.
15Michal Kosinski (2021) ‘Facial Recognition Technology Can Expose Political Orientation from Naturalistic Facial Images’, Nature Scientific Reports, vol. 11, no. 100.
16See tinyurl.com/CreepySchmidt.
17See tinyurl.com/GmailPrivacyCase.
18See tinyurl.com/GmailAppDeveloper.
19Aleecia M. McDonald & Lorrie Faith Cranor (2008) ‘The Cost of Reading Privacy Policies’, I/S: A Journal of Law and Policy for the Information Society, vol. 4, no. 3, pp. 543–68.
The Planet
1The online version of this interview was corrected after I complained about this quote, but you can read the original erroneous quote I reproduce here on the Factiva database: ‘Australia Joins World-first AI Group to Tackle Ethics, Commercialisation’, Australian Financial Review, 15 June 2020.
2Anders S.G. Andrae & Tomas Edler (2015) ‘On Global Electricity Usage of Communication Technology: Trends to 2030’, Challenges, vol. 6, pp. 117–57.
3Eric Masanet et al. (2020) ‘Recalibrating Global Data Center Energy-use Estimates’, Science, vol. 367, no. 6481, pp. 984–86.
4Lasse F. Wolff Anthony, Benjamin Kanding & Raghavendra Selvan (2020) ‘Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models’, ICML Workshop on Challenges in Deploying and Monitoring Machine Learning Systems.
5Microsoft Azure provides the compute infrastructure to OpenAI as part of Microsoft’s $1-billion investment in OpenAI. In return, Microsoft has been given an exclusive licence to GPT-3.
7Danny Hernandez & Tom B. Brown (2020) ‘Measuring the Algorithmic Efficiency of Neural Networks’, arXiv report 2005.04305.
9R. Vinuesa et al. (2020) ‘The Role of Artificial Intelligence in Achieving the Sustainable Development Goals’, Nature Communications, vol. 11, no. 1, p. 233.
The Way Ahead
1See tinyurl.com/EUAIGuidelines.
2See transparency.facebook.com.
3See tinyurl.com/FacebookNoTrust.
4See tinyurl.com/TuringRedFlag.
6With credit to Eric Schmidt, ex-CEO and chairman of Google: see tinyurl.com/CreepySchmidt.
7See tinyurl.com/TwitterMicrotargetBan.
8You can sign up for ‘Elements of AI’ at elementsofai.com.
10For the nitpicker, it is true that life expectancy in some richer countries has increased beyond three score years and ten to over four score years. And richer people within those countries tend to live a little bit longer still, due to better diet and so on. But the number of extra years alive that money can buy is still rather limited. Paul Allen’s immense wealth from co-founding Microsoft could not save him from dying from the complications of Non-Hodgkin lymphoma at just 65 years of age. Nor could the considerable wealth of Steve Jobs, co-founder of Apple, help him defeat pancreatic cancer, which killed him at just 56 years of age.