The Ethics of Algorithms: Mapping the debate | Mittelstadt, Allo, Taddeo, Wachter, Floridi
Abstract
In information societies, operations, decisions and choices previously left to humans are increasingly delegated to algorithms, which may advise, if not decide, about how data should be interpreted and what actions should be taken as a result. More and more often, algorithms mediate social processes, business transactions, governmental decisions, and how we perceive, understand, and interact among ourselves and with the environment. Gaps between the design and operation of algorithms and our understanding of their ethical implications can have severe consequences affecting individuals as well as groups and whole societies.
This paper makes three contributions to clarify the ethical importance of algorithmic mediation:
- It provides a prescriptive map to organise the debate.
- It reviews the current discussion of ethical aspects of algorithms.
- And it assesses the available literature in order to identify areas requiring further work to develop the ethics of algorithms.
Introduction
Examples with references, all cited verbatim:
- Profiling and classification algorithms determine how individuals and groups are shaped and managed (Floridi, 2012).
- Recommendation systems give users directions about when and how to exercise, what to buy, which route to take, and who to contact (Vries, 2010: 81)
- Data mining algorithms are said to show promise in helping make sense of emerging streams of behavioural data generated by the ‘Internet of Things’ (Portmess and Tower, 2014: 1).
- Online service providers continue to mediate how information is accessed with personalisation and filtering algorithms (Newell and Marabelli, 2015; Taddeo and Floridi, 2015).
- Machine learning algorithms automatically identify misleading, biased or inaccurate knowledge at the point of creation (e.g. Wikipedia’s Objective Revision Evaluation Service).
Core idea: algorithms are "inescapably value-laden (Brey and Soraker, 2009; Wiener, 1988)." (1)
How?
- Operational parameters specified by developers and users with an outcome in mind. This desired outcome implies a preference of some values over others.
- cf. Friedman and Nissenbaum, 1996; Johnson, 2006; Kraemer et al., 2011; Nakamura, 2013
- An algorithm with acceptable parameters does however not imply ethically acceptable results. Profiling algorithms may discriminate for example.
Main question: what is the "potential and ethical impact of an algorithm" (2)?
This question is hard to answer due to structural limitations that introduces a gap between algorithm design and understanding their ethical implications:
- What is the influence of human subjectivity in algorithm design and config?
- Long-term, multi-user development processes need to be tracked
- We'll probably only have a clue as problem cases arise
- Learning algorithms are not fully predictable and explainable
- Especially when they can learn online and tweak their own parameters
- Is a mistake a one-time misclassification or evidence of systemic bias?
Main question 2: how do ethical issues apply directly to algorithms rather than the technology built upon them?
Approach:
1) Define a conceptual map with six types of concerns that sufficiently describe ethical issues of algorithm design 2) Review the current research field based on that map. Results in seven main themes. 3) Pointers for future research based on the conceptual map.
Background
There is a conflation of formal mathematical definitions and popular usage of algorithms.
In popular media, algorithms might refer to any (potentially ill-defined) decision procedure, and also to any "program" that implements various algorithms and their interaction.
This article follows Hill (2015, 47), in "Hill RK (2015) What an algorithm is. Philosophy & Technology 29(1): 35–59." Hill defines an algorithm as a mathematical construct.
Def: "a finite, abstract, effective, compound control structure, imperatively given, accomplishing a given purpose under given provisions." (Hill, 2015, 47)
But algorithms need to be implemented to have effect in the world. The authors thus also include the lay interpretation of the implementation of an algorithm into a technology, configured for a particular task.
In other words, the ethical implications of algorithms can be discussed on multiple levels:
- as mathematical construct
- implementation in technologies and programs
- configuration into an application for a concrete task
The issue with the layman's interpretation of algorithm is that it can apply to any decision process, leading to a huge explosion in the amount of artefacts to be studied.
The authors focus on:
Without offering a precise definition of the class, the algorithms we are interested in here are those that make generally reliable (but subjective and not necessarily correct) decisions based upon complex rules that challenge or confound human capacities for action and comprehension.4 In other words, we are interested in algorithms whose actions are difficult for humans to predict or whose decision-making logic is difficult to explain after the fact. Algorithms that automate mundane tasks, for instance in manufacturing, are not our concern. (3)
These type of algorithms look for "actionable insights" (3) rather than exact causal relationships.
Algorithms challenge human decision-making and comprehension:
1) Sheer size of analysis (e.g. 10000 factors taken into consideration for risk stratification) that is qualitatively different than a human decision process. 2) Uncertainty introduced by algorithmic learning capacities * "Machine learning is defined by the capacity to define or modify decision-making rules autonomously." (3) * Difficulty predicting results * Difficulty explaining the decision process behind final outcomes
Map of the ethics of algorithms
Does not assume a particular ethical theory, "but rather is intended as a prescriptive framework of types of issues arising from algorithms owing to three aspects of how algorithms operate." (4)
1) "turn data into evidence for a given outcome" (4) 2) uses to "trigger and motivate an action that (on its own, or when combined with other actions) may not be ethically neutral." (4) 3) Due to the (semi-)autonomous nature of this procedure, the "apportionment of responsibility for effects of actions driven by algorithms" (4) is complicated.
This leads to six types of ethical concerns.
Three epistemic concerns. What is the quality of the evidence that motivates the made decision?:
1) Inconclusive evidence 2) Inscrutable evidence 3) Misguided evidence
Two normative concerns:
4) Unfair outcomes 5) Transformative effects
Overarching concern, relevant for all previous ones:
6) The traceability of responsibility
Inconclusive insights
statistical methods can help identify significant correlations, but these are rarely considered to be sufficient to posit the existence of a causal connection (Illari and Russo, 2014: Chapter 8), and thus may be insufficient to motivate action on the basis of knowledge of such a connection. The term actionable insight we mentioned earlier can be seen as an explicit recognition of these epistemic limitations (4)
Often, (learning) algorithms are used where no exact alternatives are available or feasible, but this limitations should be taken into account.
How does the risk of being wrong translate to "epistemic responsibilities (Miller and Record, 2013)" (4). The authors do not explain what that means, but I interpret it as such: sometimes you are responsible for bad outcomes due to neglect on epistemic grounds, i.e. you could have known better and you should have known better.
Inscrutable evidence
When data are used as (or processed to produce) evidence for a conclusion, it is reasonable to expect that the connection between the data and the conclusion should be accessible (i.e. intelligible as well as open to scrutiny and perhaps even critique). (4)
Misguided evidence
Output doesn't exceed the input, Shannon:
While Shannon’s mathematical theory of communication (Shannon and Weaver, 1998), and especially some of his information-inequalities, give a formally precise account of this fact, the informal ‘garbage in, garbage out’ principle clearly illustrates what is at stake here, namely that conclusions can only be as reliable (but also as neutral) as the data they are based on. Evaluations of the neutrality of the process, and by connection whether the evidence produced is misguided, are of course observer-dependent. (5)
Unfair outcomes
From epistemic to normative: Focus on evaluating algorithm-driven actions instead of reviewing the evidence for them.
This is actually a really clarifying point, that explains how ethical issues of algorithms often appear in the news. Cf:
An action can be found discriminatory, for example, solely from its effect on a protected class of people, even if made on the basis of conclusive, scrutable and well-founded evidence. (5)
Transformative effects
Reliance on the outcomes of algorithms can appear ethically neutral, but still cause harm. However, due to appearing ethically neutral, responsibility is not necessarily retraceable to an epistemic failure, as above. Due to the reontologizing of algorithms:
This is because algorithms can affect how we conceptualise the world, and modify its social and political organisation (cf. Floridi, 2014). Algorithmic activities, like profiling, reontologise the world by understanding and conceptualising it in new, unexpected ways, and triggering and motivating actions based on the insights it generates. (5)
Traceability
This implies that harm caused by algorithmic activity is hard to debug (i.e. to detect the harm and find its cause), but also that it is rarely straightforward to identify who should be held responsible for the harm caused (5)
Inconclusive evidence leading to unjustified actions
Continue here