References
— Corbett-Davies, S., Pierson, E., Feller, A., Goel, S., & Huq, A. (2017). Algorithmic decision making and the cost of fairness. In Proc. of KDD.
— Hajian, S., Domingo-Ferrer, J. (2013). A methodology for direct and indirect discrimination prevention in data mining. IEEE TKDE 25(7), 11 1445-1459.
— Hajian, S., Bonchi, F., & Castillo, C. (2016). Algorithmic bias: From discrimination discovery to fairness-aware data mining. Tutorial In Proc. KDD. ACM. Online: http://francescobonchi.com/algorithmic_bias_tutorial.html
— Kasper Lippert-Rasmussen: Born Free and Equal? A Philosophical Inquiry Into the Nature of Discrimination. Oxford University Press, 2013.
— Kleinberg, J., Mullainathan, S., & Raghavan, M. (2016). Inherent trade-offs in the fair determination of risk scores. In Proc. ITCS.
— D. Pedreschi, S. Ruggieri, F. Turini: A Study of Top-K Measures for Discrimination Discovery. SAC 2012.
— O'Neil, C. (2017). Weapons of math destruction: How big data increases inequality and threatens democracy. Broadway Books.
— Wagstaff, K. (2012). Machine learning that matters. arXiv:1206.4656. 19
9 Making quality decisions about the uses of algorithms and AI
Verónica Dahl
Simon Fraser University
9.1 Statement
How do algorithms, when exploited in different applications, affect human cognitive capabilities?
While algorithms can be very helpful as extensions of the human brain, unregulated algorithms can tend to create cognitive dissonance between our true power as humans and our perception of it, by giving us a disempowering sense of:
(a) subordination to machines: we must de facto submit to their modes of communicating, rather than vice-versa-e.g. call centers debug their defective speech recognition systems and decision trees for free at the public’s temporal expense (an instance of the “time theft” crime), often with no helpful end results and no possibility to reach a human. (b) uncertainty: decisions are often presented as unquestionable just because an impenetrable black box made them, leaving us unable to find out which criteria warranted a decision. This promotes either unwarranted blind trust or defeated resignation to uncertainty, not only among the general public but even in the very researchers developing or using the algorithms. How do algorithms have the potential to modify the way humans make decisions based on them (e.g. influence of
recommendations, personalization)?
Algorithms being morally neutral, like tools are in general, they can either be used to create a happier state of the world, or to perpetuate and deepen inequalities and injustices, placing us at greater risk of global catastrophe. In Toby Walsh’s words, the future is the product of the choices that we make today2. Given the speed at which AI advances unchecked and unregulated, we had better start foreseeing (and legally mandating) how algorithms and decisions based on, or made through them are bound to change the world, and how to best exploit them in order to gracefully adapt to the coming changes, that they may be for universal good.
Unfortunately, we are already cognitively prone to exhibit an unwarranted level of trust in algorithms when we make decisions based on them: we tend to treat information from an AI system as is it came from a trusted colleague, when in fact the system cannot even explain to us, like a colleague would, why it reached a given conclusion. A healthy skepticism is a must, as is the development of a proactive policy to address the potentially destructive consequences of algocracy (leaving decisions to machines), technological unemployment, and autonomous systems such as killer robots.
It is a fact of life that algorithms will be more and more capable of making decisions that were previously made by humans, and in general, of replacing humans, whose salary demands cannot compete with the under-regulated way in which algorithms have been allowed to replace them. But with enough foresight and legal provisions, we can revert this situation in ways that lead to humans universally getting to do work we value, and leaving the less desirable aspects of jobs to machines. As a scientific community we could gather around this and similar goals in order to prompt and help legislators in bringing the needed changes to fruition, and ensure that the results of research in AI can no longer be captured by a few powerful players to the detriment of the public that largely funded that research to begin with.
Toby Walsh (2018) Machines that think: the future of Artificial Intelligence.
Regulation should also be put in place to prevent humans in power positions from making poor or abusive decisions just because these are now made possible by technology, e.g. automated tax reassessments can now be made in blanket fashion and be sent massively, creating a mountain of appeals that remain unsolved for years, since employees can't keep up with their processing. The consequences are dire for those who need tax clearance to proceed with their lives. Many end up simply paying, from either need of clearance or attrition, the undue amounts exacted.