Wednesday, May 18, 2022

Darwin's Logical Argument for Natural Selection

One of the things I occasionally like to do is to re-read books that had an early influence on my thinking. It is an instructive exercise. Sometimes, when you read a book early in life you are easily impressed by its ideas and arguments. Oftentimes, this because so many of them are new to you. They have, as a result, an outsized influence on your worldview. When you re-read them, you often find them less compelling. You will have learned so much in the intervening years that the ideas and arguments start to seem obvious and stale.

There are some exceptions to this trend. One example of this, for me at any rate, is Daniel Dennett’s book Darwin’s Dangerous Idea. I first read it in my late teens. I loved it at the time. I was new to debates about Darwinism, its scientific basis, and its philosophical implications. I lapped up everything Dennett had to say. Re-reading it now, I still find it compelling. To be clear, a lot of it is not as impressive as I thought at the time. For example, I used to like Dennett’s somewhat imperious and bitchy style of writing -- so critical and dismissive of his peers -- but I don’t like that so much anymore. Nevertheless, I was pleased to find that the book is still full of interesting metaphors and thought experiments: universal acid, skyhooks and cranes, the Library of Mendel, the Two-Bitser machine and so on. All of these get you to think about the world in a new way and many of them still resonate to this day.

That’s a long introduction — a mini-book review of sorts — to what is going to be a very simple post that doesn’t really have anything to do with Dennett’s book.

One of the things I re-read in Dennett’s book was the summary passage from Darwin’s Origin of Species in which Darwin sets out the logical argument for evolution by natural selection. Typical of a lot writing — particularly 19th century writing — Darwin expresses the argument in a convoluted style. Here it is in all its original glory:

If during the long course of ages and under varying conditions of life, organic beings vary at all in the several parts of their organisation, and I think this cannot be disputed; if there be, owing to the high geometrical powers of increase of each species, at some age, season, or year, a severe struggle for life, and this certainly cannot be disputed; then, considering the infinite complexity of the relations of all organic beings to each other and to their conditions of existence, causing an infinite diversity in structure, constitution, and habits, to be advantageous to them, I think it would be a most extraordinary fact if no variation ever had occurred useful to each being's own welfare, in the same way as so many variations have occurred useful to man. But if variations useful to any organic being do occur, assuredly individuals thus characterised will have the best chance of being preserved in the struggle for life; and from the strong principle of inheritance they will tend to produce offspring similarly characterised. This principle of preservation, I have called, for the sake of brevity, Natural Selection. 
(Darwin, Origin of Species, 1st Edition, pg 127)


Regular readers of this blog will know that one of my hobbies is to extract logical arguments from long prosaic summaries. Indeed, it is an exercise I often set for students in my classes. Reading through this passage, it seemed obvious to me that there is a much more straightforward and logically compelling way of expressing Darwin’s argument. I thought it might be interesting to show how to do this.

The first thing to note — which Dennett does in his book — is that the passage contains a series of ‘if…then…” statements (or conditional statements). As every first-year philosophy student knows, ‘if…then…’ statements are the building blocks of simple deductive arguments, such as:

(1) If X, then Y

(2) X

(3) Therefore, Y


Darwin’s argument consists of a chain of two “if…then…” arguments that build to his conclusion in favour of natural selection. Admittedly, some of the ‘if…then…’ statements that make up those two arguments are complex, and contain asides that are distracting, but it’s easy to see them in the text.

The first one is actually a double conditional statement contained in the first sentence. Here it is with the key bits highlighted:

If during the long course of ages and under varying conditions of life, organic beings vary at all in the several parts of their organisation, and I think this cannot be disputed; if there be, owing to the high geometrical powers of increase of each species, at some age, season, or year, a severe struggle for life, and this certainly cannot be disputed; then, considering the infinite complexity of the relations of all organic beings to each other and to their conditions of existence, causing an infinite diversity in structure, constitution, and habits, to be advantageous to them, I think it would be a most extraordinary fact if no variation ever had occurred useful to each being's own welfare, in the same way as so many variations have occurred useful to man.


To put this a bit more simply:

  • (1) If there is variation in organic beings, and if there is a severe struggle for life, then there must be some variations that are useful to surviving that struggle.

I have changed the bit after the ‘then’ in order to capture the essence of what Darwin is trying to say. If I had my druthers I would amend it even further to match modern terminology (e.g. “variations will be fitness enhancing”). The asides in the text are the claims that both of the conditions (variation and struggle) are met in reality. So the first part of Darwin’s argument, with the logical inferences filled in, works like this:

  • (1) If there is variation in organic beings, and if there is a severe struggle for life, then there must be some variations that are useful to surviving that struggle.
  • (2) There is variation in organic beings.
  • (3) There is a severe struggle for life.
  • (4) Therefore, there must be some variations that are useful to surviving that struggle (from 1, 2 and 3).

This brings us to the second part of Darwin’s argument, which occurs in the next two sentences of the quoted passage. Here they are with the relevant bits highlighted:

But if variations useful to any organic being do occur, assuredly individuals thus characterised will have the best chance of being preserved in the struggle for life; and from the strong principle of inheritance they will tend to produce offspring similarly characterised. This principle of preservation, I have called, for the sake of brevity, Natural Selection.


Okay, I highlighted a lot of that section because it is slightly less convoluted than the first sentence. But there is still a lot going on here. Tidying it up, here is what we get:

  • (5) If some variations are useful to surviving the struggle, and if there is a strong principle of inheritance, then useful variations will be preserved.
  • (6) There is a strong principle of inheritance (i.e. offspring are likely to resemble their parents) [implied not stated in the quoted passage]
  • (7) Therefore, useful variations will be preserved (from 4, 5 and 6).

And the preservation of useful variations is simply what Darwin calls ‘natural selection’.

In full, then, Darwin’s logical argument for natural selection, taken from the quoted passage, looks like this:

  • (1) If there is variation in organic beings, and if there is a severe struggle for life, then there must be some variations that are useful to surviving that struggle.
  • (2) There is variation in organic beings.
  • (3) There is a severe struggle for life.
  • (4) Therefore, there must be some variations that are useful to surviving that struggle (from 1, 2 and 3).
  • (5) If some variations are useful to surviving the struggle, and if there is a strong principle of inheritance, then useful variations will be preserved.
  • (6) There is a strong principle of inheritance (i.e. offspring are likely to resemble their parents) [implied not stated in the quoted passage]
  • (7) Therefore, useful variations will be preserved (from 4, 5 and 6).

There is a lot of detail packed into this argument. I have called it the ‘logical argument’ since no empirical evidence is adduced in the quoted passage in support of the key empirical claims (2, 3 and 6). The rest of the Origin of Species provides a lot of evidence in support of those claims. Darwin meticulously documents variation and inheritance in species and gives many examples of the struggle for life. Since Darwin’s time, the field of evolutionary biology has provided reams and reams of evidence in support of those claims, identifying, in much greater detail, the mechanisms of inheritance. In fact, one of Darwin's famous blindspots was the mechanism of inheritance: he knew it happened but didn't know why because he knew nothing about genetics. The amassing of evidence since the time of Darwin is one reason why the argument still holds up to this day.

If I were to make one amendment to the argument it would be to insist that the first premise include the phrase ‘if there is [a lot of] variation…”. Why? Because it seems obvious to me that if organisms vary only in one or two ways, an insufficient volume of variation will be produced to allow variations useful to the wide diversity of struggles for existence to arise. Fortunately, we know that there is a lot of variation in reality so this amendment is easily made.

Anyway, that's all I wanted to say in this post. I hope this logical reconstruction of Darwin's argument is of interest to some people.

Monday, April 25, 2022

Criticisms and Developments of Ethical Behaviourism

A few years ago, I developed a position I called 'ethical behaviourism' and applied it to debates about the moral status of artificial beings. Roughly, ethical behaviourism is a moral equivalent of the Turing test for artificial intelligence. It states that if an entity looks and acts like another entity with moral status, then you should act as if it has that status. More strongly, it states that the best evidence we have for knowing that another entity has moral status is behavioural. No other form of evidence (mechanical, ontological, historical) trumps the behavioural evidence.

My longest defence of this theory comes from my original article "Welcoming Robots into the Moral Community: A Defence of Ethical Behaviourism" (official; open access), but, in many ways, I prefer the subsequent defence that I wrote up for a lecture in 2019 (available here). The latter article clarifies certain points from my original article and responds to additional objections.

I have never claimed that ethical behaviourism is particularly original or insightful. Very similar positions have been developed and defended by others in the past. Nevertheless, for whatever reason, it has piqued the curiosity of other researchers.  The original paper has been cited nearly 80 times, though most of those citations are 'by the way'. More significantly, there are now several interesting and substantive critiques and developments on it available in the literature. I thought it would be worthwhile linking to some of the more significant ones here. I link to open access versions wherever possible.

If you know of other substantive engagements with the theory, please let me know.

  • "The ethics of interaction with neurorobotic agents: a case study with BabyX" by Knott, Sagar and Takac - This is possibly the most interesting paper engaging with the idea of ethical behaviourism. It is a case study of an actual artificial agent/entity. Ultimately, the authors argue that my theory does not account for the experience of people interacting with this agent, and suggest that artificial agents that mimic certain biological mechanisms are more likely to warrant the ascription of moral patiency.

  • 'Is it time for rights for robots? Moral status in artificial entities' by Vincent Müller - A critique of all proponents of moral status for robots that includes somewhat ill-tempered critique of my theory. Müller admits he is offering a 'nasty reconstruction' (something akin to a 'reductio ad absurdum') of his opponents' views. I think he misrepresents my theory on certain key points. I have corresponded with him about it, but I won't list my objections here. 

  • 'Social Good Versus Robot Well-Being: On the Principle of Procreative Beneficence and Robot Gendering' by Ryan Blake Jackson and Tom Williams - One of the throwaway claims I made in my original paper on ethical behaviourism was that, if the theory is correct, robot designers may have 'procreative' duties toward robots. Specifically, they may be obliged to follow the principle of procreative beneficence (make the best robots it is possible to make). The authors of this paper take up, and ultimately dismiss, this idea. Unlike Müller's paper, this one is a good-natured critique of my views.

  • 'How Could We Know When a Robot was a Moral Patient?' by Henry Shevlin - A useful assessment of the different criteria we could use to determine the moral patiency of a robot. Broadly sympathetic to my position but suggests that it needs to be modified to include cognitive equivalency and not just behavioural equivalency.

Another honourable mention here would be my blog post on ethical behaviourism in human-robot relationships. It summarises the core theory and applies it to a novel context.

Friday, April 8, 2022

How Can Algorithms Be Biased?

Image from Marco Verch, via Flickr

The claim that AI systems are biased is common. Perhaps the classic example is the COMPAS algorithm used to predict recidivism risk amongst prisoners. According to a widely-discussed study published in 2016, this algorithm was biased against black prisoners, giving them more false positive ‘high risk’ scores, than white prisoners. And this is just one example of a biased system. There are many more that could be mentioned, from facial recognition systems that do a poor job recognising people with darker skin tones, to employment algorithms that seem to favour male candidates over female ones.

But what does it mean to say that AI or algorithmic system is biased? Unfortunately, there is some disagreement and confusion about this in the literature. People use the term ‘bias’ to mean different things. Most notably, some people use it in a value-neutral, non-moralised sense whereas others use it in a morally-loaded pejorative sense. This can lead to a lot of talking at cross purposes. But also some people use the term to describe different types or sources of bias. A lot would be gained if we could disambiguate these different types.

So that’s what I will try to do in the remainder of this article. I will start by distinguishing between moral and non-moralised definitions of ‘bias’. I will then discuss three distinct causes of bias in AI systems, as well as how bias can arise at different stages in the developmental pipeline for AI systems. Nothing I say here is particularly original. I draw heavily from the conceptual clarifications already provided by others. All I hope is that I can cover this important topic in a succinct and clear way.

1. Moralised and Non-Moralised Forms of Bias

One of the biggest sources of confusion in the debate about algorithmic bias is the fact that people use the term ‘bias’ in moralised and non-moralised ways. You will, for example, hear people say that algorithms are ‘inherently biased’ and that ‘we want them to be biased’. This is true, in a sense, but then creates problems when people start to criticise the phenomenon of algorithmic bias in no uncertain terms. To avoid this confusion, it is crucial to distinguish between the moralised and non-moralised senses of ‘bias’.

Let’s start with the non-moralised sense. All algorithms are designed with a purpose in mind. Google’s pagerank algorithm is intended to sort webpages into a ranking that respects their usefulness or relevance to a search user. Similarly, the route-planning algorithm on Google Maps (or similar mapping services) tries to select the quickest and most convenient route between A and B. In each case, there is a set of possible outcomes among which the algorithm can select, but it favours a particular subset of outcomes because those match better with some goal or value (usefulness, speed etc).

In this sense it is true to say that most, if not all, algorithms are inherently biased. They are designed to be. They are designed to produce useful outputs and this requires that they select carefully from a set of possible outputs. This means they must be biased in favour of certain outcomes. But there is nothing necessarily morally problematic about this inherent bias. Indeed, in some cases it is morally desirable (though this depends on the purpose of the algorithm). When people talk about algorithms being biased in this sense (of favouring certain outputs over others), they are referring to bias in the non-moralised or neutral sense.

How does this contrast with the moralised sense of ‘bias’? Well, although all algorithms favour certain kinds of output, sometimes they will systematically favour outputs that have an unfair impact on a certain people or populations. A hiring algorithm that systematically favours male over female candidates would be an example of this. It reflects and reproduces gender-based inequality. If people refer to such a system as being ‘biased’ they are using the term in a moralised sense: to criticise its moral impact and, perhaps, to blame and shame those that created it. They are saying that this is a morally problematic algorithm and we need to do something about it.

There appear to two conditions that must be met in order for an algorithm to count as biased in this second, moralised, sense:

Systematic output: The algorithm must systematically (consistently and predictably) favour one population/group over another, even if it this effect is only indirect.
Unfair effect: The net effect of the algorithm must be to treat populations/groups differently for morally arbitrary or illegitimate reasons

These conditions are relatively straightforward but some clarifications may be in order. The systematicity condition is there in order to rule out algorithms that might be biased, on occasions, for purely accidental reasons, but not on a repeated basis. Furthermore, when I say that the systematic effect on one population may be ‘indirect’ what I mean is that the algorithm itself may not be overtly or obviously biased against a certain population, but nevertheless affects them disproportionately. For example, a hiring algorithm that focused on years spent in formal education might seem perfectly legitimate on its face, with no obvious bias against certain populations, but its practical effect might be rather different. It could be that certain ethnic minorities spend less time in formal education (for a variety of reasons) and hence the hiring algorithm disproportionately disfavours them.

The unfairness condition is crucial but tricky. As noted, some forms of favourable or unfavourable treatment might be morally justified. A recidivism risk score that accurately identifies those at a higher risk of repeat offending would treat a sub-population of prisoners unfavourably, but this could be morally justified. Other forms of unfavourable treatment don’t seem justified, hence the furore about the recidivism risk score that treats black prisoners differently from white prisoners. These are relatively uncontroversial cases. The problem is that there are sometimes grounds for reasonable disagreement as to whether a certain forms of favourable treatment are morally justified or not. This moral disagreement will always affect debates about algorithmic bias. This is unavoidable and, in some cases, can be welcomed: we need to be flexible in understanding the boundaries of our moral concepts in order to allow for moral progress and reform. Nevertheless, it is worth being aware of it whenever you enter a conversation about algorithmic bias. We might not always agree whether a certain algorithm is biased in the moralised sense.

One last point before we move on. In the moralised sense, a biased algorithm is harmful. It breaches a moral norm, results in unfair treatment, and possibly violates rights. But there are many forms of morally harmful AI that would not count as biased in the relevant sense. For instance, an algorithmic system for piloting an airplane might result in crashes in certain weather conditions (perhaps in a systematic way), but this would not count as a biased algorithm. Why not? Because, presumably, the crashes would affect all populations (passengers) equally. It is only when there is some unfair treatment of populations/groups that there is bias in the moralised sense.

In other words, it is important that the term ‘bias’ does not do too much heavy-lifting in debates about the ethics of AI.

2. Three Causes of Algorithmic Bias

How exactly does bias arise in algorithmic systems? Before I answer that allow me to indulge in a brief divagation.

One of the interesting things that you learn when writing about the ethics of technology is how little of it is new. Many of the basic categories and terms of debate have been set in stone for some time. This is true for much of philosophy of course, but those of use working in ‘cutting edge’ areas such as the ethics of AI sometimes like to kid ourselves that we are doing truly innovative and original work in applied ethics. This is rarely the case.

This point was struck home to me when I read Batya Friedman and Helen Nissenbaum’s paper ‘Bias in Computer Systems’. The paper was published in 1996 — a lifetime ago in technology terms — and yet it is still remarkably relevant. In it, they argue that there are three distinct causes of bias in ‘computer systems’ (a term which can covers algorithmic and AI systems too). They are:

Preexisting bias: The computer system takes on a bias that already exists in society or social institutions, or in the attitudes, beliefs and practices of the people creating it. This can be for explicit, conscious reasons or due to the operation of more implicit factors.
Technical bias: The computer system is technically constrained in some way and this results in some biased output or effect. This can arise from hardware constraints (e.g. the way in which algorithmic recommendations have to be displayed to human users on screens with limited space) or problems in software design (e.g. how you translate fuzzy human values and goals into precise quantifiable targets).
Emergent bias: Once the system is deployed, something happens that gives rise to a biased effect, either because of new knowledge or changed context of use (e.g. a system used in a culture with a very different set of values).


Friedman and Nissenbaum give several examples of such biases in action. They discuss, for example, an airline ticket booking system that was used by US travel agents in the 1980s. The system was found to be biased in favour of US airlines because it preferred connecting flights from the same carrier. On the face of it, this wasn’t an obviously problematic preference (since there was some convenience from the passenger’s perspective) but in practice it was biased because few non-US airlines offered internal US flights. Similarly, because of how the flights were displayed on screen, travel agents would almost always favour flights displayed on the first page of results (a common bias among human users). These would both be examples of technical bias (physical constraints and design choices).

3. The Bias Pipeline

Friedman and Nissenbaum’s framework is certainly still useful, but we might worry that it is a little crude. For example, their category of ‘preexisting bias’ seems to cover a wide range of different underlying causes of bias. Can we do better? Is there a more precise way to think about the causes of bias?

One possibility is the framework offered by Sina Fazelpour and David Danks in their article ‘Algorithmic Bias: Senses, Sources, Solutions’. This is a more recent entry into the literature, published in 2021, and focuses in particular on the problems that might arise from the construction and deployment of machine learning algorithms. This makes sense since a lot of the attention has shifted away from ‘computer systems’ to ‘machine learning’ and ‘AI’ (though, to be clear, I’m not sure how much of this is justified).

Fazelpour and Danks suggest that instead of thinking about general causal categories, we think instead about the process of developing, designing and deploying an algorithmic system. They call this the ‘pipeline’. It starts with the decision to use an algorithmic system to assist (or replace) a human decision-maker. At this stage you have to specify the problem that you want the system to solve. Once that is done, you have to design the system, translating the abstract human values and goals into a precise and quantifiable machine language. This phase is typically divided into two separate processes: (i) data collection and processing and (ii) modelling and validation. Then, finally, you have to deploy the system in the real world, where it starts to interact with human users and institutions.

Bias can arise at each stage in the process. To make their analytical framework more concrete they use a case study to illustrate the possible forms of bias: the construction of a ‘student success’ algorithm for use in higher education. The algorithm uses data from past students to predict the likely success of future students on various programs. Consider all the ways in which bias could enter into the pipeline for such an algorithm:

Problem specification: You have to decide what counts as ‘student success’ — i.e. what is it that you are trying to predict. If you focus on grades in the first year of a programme, you might find that this is biased against first generation students or students from minority backgrounds who might have a harder time adjusting to the demands of higher education (but might do well once they have settled down). You also face the problem that any precise quantifiable target for student success is likely to be an imperfect proxy measure for the variable you really care about. Picking one such target is likely to have unanticipated effects that may systematically disadvantage one population.
Data collection: The dataset which you rely on to train and validate your model might be biased in various ways. If the majority of previous students came from a certain ethnic group, or social class, your model is unlikely to be a good fit for those outstide of those groups. In other words, if there is some preexisting bias built into the dataset, this is likely to be reflected in the resultant algorithm. This is possibly the most widely discussed cause of algorithmic bias as it stems from the classic: ‘garbage in, garbage out’ problem.
Modelling and Validation: When you test and validate your algorithm you have to choose some performance criterion against which to validate it, i.e. something that you are going to optimise or minimise. For example, you might want to maximise predictive success (how many students are accurately predicted to go on to do well) or minimise false positive/negative errors (how many students are falsely predicted to do well/badly). The choice of performance criterion can result in a biased outcome. Indeed, this problem is that the heart of the infamous debate about the COMPAS algorithm that I mentioned at the start of this article: the designers tried to optimise predictive success and this resulted in the disparity in false positive errors.
Deployment: Once the algorithm is deployed there could be some misalignment between the user’s values and those embodied in the algorithm, or they could use it in an unexpected way, or in a context that is not a good match for the algorithm. This can result in biased outcomes. For example, imagine using an algorithm that was designed and validated in a small, elite liberal arts college in the US, in a large public university in Europe (or China). Or imagine that the algorithmic prediction is used in combination with other factors by a human decision-making committee. It is quite possible that the humans will rely more heavily on the algorithm when it confirms their pre-existing biases and will ignore it when it does not.


These are just some examples. Many more could be given. The important point, drawn from both Friedman and Nissenbaum’s framework, and the one suggested by Fazelpour and Danks, is that there can be many different (and compounding) causes of bias in algorithmic systems. It is important to be sensitive to these different causes if we want to craft effective solutions. For instance, a lot of energy has been expended in recent times on developing technical solutions to the problem of bias. These are valuable, but not always. They may not be targeted at the right cause. If the problem comes from how the algorithm is used by humans, in a particular decision-making context, then all the technical wizardry may be for naught.

Tuesday, April 5, 2022

97 - The Perils of Predictive Policing (& Automated Decision-Making)

One particularly important social institution is the police force, who are increasingly using technological tools to help efficiently and effectively deploy policing resources. I’ve covered criticisms of these tools in the past, but in this episode, my guest Daniel Susser has some novel perspectives to share on this topic, as well as some broader reflections on how humans can relate to machines in social decision-making. This one was a lot of fun and covered a lot of ground.

You can download the episode here or listen below. You can also subscribe on Apple PodcastsStitcherSpotify and other podcasting services (the RSS feed is here).

Relevant Links

Tuesday, March 29, 2022

AI and the Future of the Work Ethic

That's the title of a talk I delivered to the IEET/UMass project on the future of work. You can watch it above. I look at the history of technological displacement in work and argue that, even if widespread technological unemployment does not happen, automating technologies will make work less valuable for most workers.

I also wrote a short article summarising the key arguments from the talk for the Institute of Arts and Ideas. You can read it here. (Unfortunately, this article, like most on the IAI website, seems to be periodically paywalled; if you are interested in reading the full text, contact me and I will send it to you).

Friday, March 11, 2022

Tragic Choices and the Virtue of Techno-Responsibility Gaps (New Paper)

I have a new paper coming out in the journal Philosophy and Technology. It's about responsibility gaps and why, on some occasions, they are good thing and we shouldn't always try to plug them. More specifically, it has how one of the benefits of autonomous machines is that they enable a reduced cost form of moral delegation. More details below.

Title: Tragic Choices and the Virtue of Techno-Responsibility Gaps

Links: Official; Philpapers; Researchgate

Abstract: There is a concern that the widespread deployment of autonomous machines will open up a number of 'responsibility gaps' throughout society. Various articulations of such techno-responsibility gaps have been proposed over the years, along with several potential solutions. Most of these solutions focus on 'plugging' or 'dissolving' the gaps. This paper offers an alternative perspective. It argues that techno-responsibility gaps are, sometimes, to be welcomed and that one of the advantages of autonomous machines is that they enable us to embrace certain kinds of responsibility gap. The argument is based on the idea that human morality is often tragic. We frequently confront situations in which competing moral considerations pull in different directions and it is impossible to perfectly balance these considerations. This heightens the burden of responsibility associated with our choices. We cope with the tragedy of moral choice in different ways. Sometimes we delude ourselves into thinking the choices we make were not tragic (illusionism); sometimes we delegate the tragic choice to others (delegation); sometimes we make the choice ourselves and bear the psychological consequences (responsibilisation). Each of these strategies has its benefits and costs. One potential advantage of autonomous machines is that they enable a reduced cost form of delegation. However, we only gain the advantage of this reduced cost if we accept that some techno-responsibility gaps are virtuous.


Wednesday, February 23, 2022

What is (institutional) racism?

What is racism? In particular what is institutional (or systemic or structural) racism and how does it differ, if at all, from racism simpliciter? If you are anything like me, these are questions that will have puzzled you for some time, especially since the terminology is now ubiquitous in public debates and conversations.

Don't get me wrong. It's not that the terms mean nothing to me. I think I have an intuitive sense of what people mean when they talk about racism and institutional racism, but I sometimes feel that the terminology is used without much care and that distinct phenomena are lumped together under the same terminological heading. This bothers me and I have often wondered if some clarity could be brought to the matter.

Since philosophers are usually the ones most concerned with conceptual clarity, I decided to read up on the recent(ish) literature in the philosophy of racism to see what it has to say. As it turns out, there is a considerable degree of disagreement and confusion in the philosophical literature too. There is, of course, a strong consensus that racism is a bad thing and that different mechanisms are responsible for it, but there is inconsistency in the terms used to describe those mechanisms and the understanding of exactly what it is that is bad about it.

Not being satisfied with this state of affairs, I decided I would try to clarify the terminology for myself. The remainder of this article is my attempt to share the results of this exercise. The gist of my analysis is that there are two distinct kinds of racism -- individual and institutional (I prefer this term to 'systemic' or 'structural' for reasons outlined below) -- but they intersect and overlap in important ways because (a) individuals play key roles in institutions and (b) institutions often shape how individuals understand and act in the world.

Some people might find my analysis useful; some may not. I do not purport to offer the definitive word on how we should understand 'racism'. My main aim is to clarify things for myself so that when I use terms such as ‘racism’ or 'institutional racism' at least I understand what I am trying to say.

It is worth noting that the remainder of this article is only likely to be of interest to people that wish to have the terminology clarified, not to people with some other interest in racism and racial justice. I will not be offering a normative or historical analysis of racism, nor will I be making any overt moral or political arguments . Obviously, what I have to say is relevant to such analysis and argumentation, and I do occasionally highlight this relevance, but defending a particular moral or political view lies outside the scope of this article.

The remainder of this article proceeds as follows. First, I will defend my claim that the modern philosophical literature is contested when it comes to the definition of racism. Second, I will discuss the phenomenon of individual racism. Third I will discuss institutional racism. Fourth, and finally, I will fit it all together by explaining how the individual and institutional mechanisms overlap, and consider whether there is simply one (admittedly complex) type of racism or, rather, several distinct forms of racism.

1. The Contested Nature of 'Racism'

Although everyone agrees that racism is bad, there is a lot of disagreement among philosophers as to exactly what it is. Some philosophers are monists, suggesting that there is a single type of racism, others are pluralists, arguing that racism comes in many forms. To get a sense of the inconsistency out there, consider the following definitions of 'racism'.

Here is Naomi Zack in her book Philosophy of Race:

Racism as we will consider it in this chapter, consists of prejudice or negative beliefs about people because of their race, and discrimination or unfavorable treatment of people because of their race. 
(Jack 2018, 150)


So, according to this, there are two elements to racism and both are required - negative beliefs and discrimination. Does this imply that if you have one without the other, you don’t have racism? Zack’s subsequent discussion casts some doubt on this, but both elements are still part of her initial definition.

Consider, as an alternative, Tommie Shelby's ideological definition of racism:

Racism is fundamentally an ideology... Racism is a set of misleading beliefs and implicit attitudes about 'races' or race relations whose wide currency serves a hegemonic social function. 
(Shelby 2014, 66)


Similar to Zack, to be sure, but also different in that it covers implicit attitudes (as well as overt beliefs) and focuses on 'hegemonic social function' and not 'discrimination' (though perhaps they are the same thing).

Consider also Sally Haslanger's definition, which starts from the premise that Shelby's analysis is incomplete in that it focuses too much on beliefs and attitudes and not on the broader social forces that shape those beliefs and attitudes:

[Against Shelby] I argue that racism is better understood as a set of practices, attitudes, social meanings, and material conditions, that systemically reinforce one another.


(Haslanger 2017, 1)


In her own words, this means that racism is an 'ideological formation' and not an 'ideology'. It covers not just beliefs and attitudes, but also social practices and conceptual frameworks. This gets us closer to an idea of institutional racism insofar as it moves beyond individuals and their beliefs and practices, to social systems and their consequences.

Other philosophers take a more abstract and, one could argue, traditional approach to philosophical definition. Joshua Glasgow, for instance, tries to cut through some of the disagreement by defending a 'respect'-based definition of racism:

ψ is racist if and only if ψ is disrespectful toward members of racialized group R as Rs
(Glasgow 2009, 81)


In this definition, ψ refers to any mechanism or action that produces the relevant kind of disrespect. As such, Glasgow thinks his definition covers both individual and institutional racism. However, this attempt at abstract universalism has been criticised by others as not doing a good job in capturing the true nature of institutional racism. Andrew Pierce, for instance, has argued that disrespect is too agency-centric a notion and fails to address the fact that institutional racism is more about injustice than it is about respect.

I could go on, but I won't. Other influential definitions of racism have been offered by Jorge Garcia and Lawrence Blum. Collectively, these definitions highlight the fact that there is considerable disagreement about the best definition of racism. Is it a matter of beliefs and attitudes? Institutions and outcomes? Or all of the above?

Tommie Shelby seems to be right when he says:

...The term "racism" is so haphazardly thrown about that it is no longer clear that we all mean, even roughly, the same thing by it...This doesn't mean that the concept is no longer useful, but it does suggest that we need to clearly specify its referent before we can determine whether the relevant phenomenon is always morally problematic.
(Shelby 2002, 412)

Why is there such disagreement? Part of the problem, as Alberto Urquidez points out is that some philosophers think that it is their job to capture the 'ordinary usage' of the term. This encourages them to take a narrow and conservative view of what racism is (typically focusing on overt beliefs and actions). But this effort to capture ordinary usage is misguided because ordinary usage is contested.

What’s more, there is a deeper and obvious reason for this contestation: 'racism' is a morally loaded term. No person or institution wants to be labelled 'racist’. and hence every attempt to define it is, in part, a normative project. In attempting to define it we are trying to capture and explain a morally problematic social phenomenon.

Bearing all this in mind, in what follows I will throw my lot in with what I will call the 'racial injustice' school of thought. According to this, 'racism' is the label we use to describe a mechanism that produces a racially unjust outcome. The outcomes come in many different forms (pejorative speech acts, harsh treatment, lack of equal opportunity, etc.). The underlying mechanisms also come in many different forms but they can be usefully lumped into two main categories: individual and institutional.

Some may argue that this version of racism entails some conceptual inflation (i.e. including within the scope of ‘racism’ things that were not traditionally included within it). The philosopher Lawrence Blum is critical of this in his work on the nature of racism arguing that conceptual inflation undermines the moral function of the term ‘racism’ in our discourse. I would suggest, however, that conceptual inflation in and of itself is not a problem. Concepts often evolve and change along with society. As long as we are clear about the different mechanisms involved, and their moral significance, the conceptual inflation need not undermine an effective moral discourse about racism.

2. Individualistic Mechanisms of Racism

So my claim is that we use the term ‘racism’ to describe the different mechanisms that produce racially unjust outcomes. Though there is no perfect conceptual schema of these mechanisms, we can meaningfully talk about both individualistic and institutional mechanisms. Let’s start by considering the individualistic ones.

An individual is a single human person. This human will be defined by (or constituted by) their mind and their actions. Everything we know about human biology suggests that the brain and nervous system support our minds and we use our minds to direct our actions (speech, movement etc). It is through our actions -- what we say and what we do -- that we produce racially unjust outcomes. It is, consequently, the brain and the nervous system that constitute the mechanisms underlying individualistic forms of racism.

These mechanisms can be divided into two main sub-categories. First, there are the conscious or explicit forms of racism. These include explicit beliefs, desires, intentions and actions. A person that believes that white people are innately superior to other races, that desires the continuation or reclamation of white supremacy, that uses derogatory speech to describe those of other races, that attends rallies, harasses or physically assaults members of other races, would be engaging these overt mechanisms of racism. Second, there are the unconscious or implicit forms of racism. These include behaviours and habits that, when scrutinised, evince some racial prejudice, but, if asked, the person may well deny that they hold any explicitly racists beliefs, desires or intentions, and perhaps be shocked at the suggestion. If you clutch your wallet when walking through a neighbourhood populated by members of another race, if you are less inclined to buy from them at the market, if you are more dismissive of their achievements or likely to attribute them to luck than hard work, you may be engaging these implicit mechanisms of racism.

There are a number of complexities to contend with here. First, it is worth noting that individualistic mechanisms of racism can more or less inclined to produce racially unjust outcomes. A member of the KKK that assaults and lynches a black man is doing something that is clearly and unambiguously harmful from the perspective of racial injustice. A pub bore who spouts of theories of racial supremacy, much to the annoyance and dismissal of his fellow patrons, is probably less harmful. Similarly, people that refuse to visit a doctor from another race may, in a cumulative sense, contribute to racial injustice, but their individual actions may not seem overly harmful or problematic.

Second, there is an interesting hypothetical to consider. Imagine someone that holds explicitly racist beliefs and desires but never manifests this in their speech or behaviour (in an explicit or implicit way). Are they racist? This is, in a sense, a variation on the old puzzle  “if a tree falls in a forest but no one hears it, does it make a sound”. It may be unanswerable. It does, however, cover the widely discussed phenomenon of ‘hearts and minds’ racism. My own view is that if the racist beliefs and desires never manifest in behaviour, then it’s hard to say that the person holding them is racist. Certainly they do not contribute to racially unjust outcomes. But it’s hard to take the hypothetical seriously. If someone harbours such beliefs and desires, it’s likely that it will manifest in their behaviour, perhaps in a subtle and implicit way, at some point in time.

Third, it is worth asking the question: where do individuals get their explicitly or implicitly racist beliefs, attitudes, preferences and habits from? Surely there are other distal mechanisms at work, either cultural or biological? This sounds right. In particular, it seems plausible to suggest that cultural and social forces shape an individual’s racist beliefs and practices. To be clear, I am sure that there are deeper biological forces at work too, but I suspect these take a relatively non-specific form. So, for example, I suspect that humans are biologically predisposed to form in-groups and out-groups, but the specific information they use to code or demarcate those groups depends on their current social environment, not their genes or biology. But if that is right, then the dividing line between individual and institutional forms of racism starts to get quite blurry.

3. Institutional Mechanisms of Racism

The term ‘institutional racism’ was first used by Stokely Carmichael (aka Kwame Ture) and Charles Hamilton, in their 1967 book Black Power. They used it, specifically, to distinguish between overt and explicit forms of individual racism and a more subtle form a racism that is inherent to social norms, rules and institutions. I have already suggested that this contrast between the individual and the institution is problematic (and, to be clear, Carmichael and Hamilton did not adhere to it rigidly). Nevertheless, I think the term is useful and does describe an important phenomenon.

What is that phenomenon? It helps if we have a concrete example. Here’s one, taken from an article describing different outcomes for different racial groups in the early days of the COVID-19 pandemic (the figures cited may no longer be accurate):

… racial and ethnic disparities are being replicated in COVID19 infections and death rates. African Americans make up just 12% of the population in Washtenaw County, Michigan but have suffered a staggering 46% of COVID-19 infections. In Chicago, Illinois, African Americans account for 29% of population, but have suffered 70% of COVID-19 related deaths of those whose ethnicity is known. In Washington, Latinos represent 13% of the population, but account for 31% of the COVID-19 cases, whereas in Iowa Latinos comprise are 6% of the population but 20% of COVID-19 infections. The African American COVID-19 death rates are higher than their percentage of the population in racially segregated cities and states including Milwaukee, Wisconsin (66% of deaths, 41% of population), Illinois (43% of deaths, 28% of infections, 15% of population), and Louisiana (46% of deaths, 36% of population). These racial and ethnic disparities in COVID-19 infections and deaths are a result of historical and current practices of racism that cause disparities in exposure, susceptibility, and treatment. 
(Yearby and Mohaptra 2020, p 3 — all references removed)


The idea here is that there is a set of social outcomes — infections, serious disease and death — in which members of certain racial groups are overrepresented. Since these are bad outcomes, we can take it that they provide examples of racial injustice. But what causes those bad outcomes? It could be that there are overtly racist individuals going around infecting racial minorities and ensuring they cannot access good healthcare, but this seems implausible and, even if there were some such individuals, they are unlikely to be able to produce such outcomes by themselves. Deeper forces must be at work.

As Yearby and Mohaptra see it, the main problem is that members of racial minorities are more likely to work in low-paying manual jobs, which means they cannot work from home, which means they are more likely to be exposed to infection. They are also less likely to have health insurance and access to proper healthcare provision and live in more densely populated housing (further increasing their risk of infection). Why did this happen them? Because there was a set of social institutions that sorted them into jobs, housing and healthcare provision that made them more susceptible to the pandemic. These institutions include schools and colleges, job markets, healthcare markets and housing markets, as well as the political and legal institutions that support those other social systems. Some overtly racist people may work within those institutions, and they may keep them going, but it is likely that these institutions also operate according to habits, norms and sanctions that were set down in the past (perhaps when racism was more overt and socially acceptable) and people working within them continue to follow those habits, norms and sanctions and reproduce the same outcomes, without being overtly racist.

In short, then, institutional racism arises whenever we have a social institution or set of such institutions that sorts people into different outcome categories (educational attainment; employment; health; incarceration etc.) on the basis of race. The result of this sorting is not morally justified. These institutions may function on the basis of explicitly racist beliefs and ideologies but they also may not.

The term ‘institutional racism’ is sometimes used interchangeably with cognate terms such as ‘structural racism’ or ‘systemic racism’. Perhaps there are subtle distinctions to be made between these terms, but I have not encountered a satisfactory account of those subtle distinctions in my readings. My sense is that people use the terms synonymously. I prefer the term ‘institutional racism’ over the synonyms. Why? Because there is a rich theoretical understanding of institutions to be found in philosophy and sociology and using the term calls upon those theoretical understandings. In particular, it calls upon the different mechanisms underlying social institutions and how they can contribute to the production of racially unjust outcomes.

Seumas Miller’s article in the Stanford Encyclopedia of Philosophy is a good entry point into these theoretical literature on institutions. As he points out, institutions have four main properties:

Functions - i.e. they serve some social purpose or purposes, such as providing educational credentials or healthcare or jobs.
Structures - i.e. they have some formal structures they use to produce those outcomes. These can be tangible or intangible — buildings, roads, ICT networks, legal-bureaucratic hierarchies and, perhaps most crucially, defined roles that must be performed by human or other agents within those institutions (teachers; prisoner officers and so forth).
Cultures - i.e. the informal, sometimes tacit and unstated, attitudes and values of the institution that gets communicated and passed between people occupying institutional roles (e.g. the value of hard work; the importance of intelligence/cleverness and so on)
Sanctions - i.e. some way of policing or enforcing conformity with the institutional roles and functions.


This last feature of institutions is controversial, as Miller himself notes. Not all institutions have sanctions and some, presumably, have incentives or rewards, that perform a similar function. Still, it is probably fair to say that sanctions, either of the formal kind (legal punishment) or informal kind (moral approbation or criticism), do feature in many institutions.

What value does this account of institutions have for our understanding of racism? Well, it points to different potential causes and mechanisms of institutional racism. Some institutions have overtly racist functions (slavery being the obvious example) but many do not. They serve valid social functions but they do so in an unequal or arbitrary way. Some institutions have structures that help reproduce racist outcomes (ICT systems that are inaccessible to or fail to recognise people from a particular background). Some institutions have cultures that reinforce racial prejudices or serve racist purposes (the belief that racial minorities are less likely to be well-educated or less likely to achieve outcomes on the basis of merit). Some institutions have sanctions that affect different races differently (the tendency to be more morally critical of racial minorities). Some institutions, of course, have all of these things at once or in different combinations. These racially unjust purposes, structures, cultures and sanctions may operate in a subtle or hidden way.

Sensitivity to the complex structure of social institutions, and the different ways in which they can sort people into different outcomes along racial lines, allows us to enrich our understanding of institutional racism.

4. Fitting it All Together

To sum up, I think the term ‘racism’ can be applied to any mechanism that produces a racially unjust outcome (typically an action or event or state of affairs that affects different racial groups differently without appropriate moral justification). There are many different mechanisms that can be responsible for such outcomes and these can be grouped, loosely, into individual and institutional classes. Individual mechanisms of racism arise from an individual’s beliefs, desires, intentions, actions and so on. Some of these can be explicitly racist; some implicitly so. Institutional mechanisms of racism arise from the different properties of social institutions (their functions, structures, cultures and sanctions).

The dividing line between individual and institutional mechanisms is not clean and sharp. It is blurry and imprecise. Institutions are made up of individuals, occupying distinct institutional roles. These individuals will affect the institutional function, structure, culture and sanctions. Contrariwise, individuals imbibe many of their explicit beliefs and practices, as well as their implicit assumptions and norms, from social institutions. There is, in essence, a constant feedback loop between the individual and institutional forms of racism.

One final point, before I conclude. One thing that struck me as I wrote this piece was the sense that there may be something linguistically impoverished about the discussion of racism in the modern world. Perhaps one of the problems, hinted at previously when I referenced the work of Lawrence Blum, is that we put too much pressure on one term -- ‘racism’ -- and expect it to do too much conceptual work. A richer vocabulary might allow us to identify and reform the same moral problems, without getting tied up in linguistic debates about whether something is truly ‘racist’ or properly described as such.

In this respect, there may be some inspiration to be drawn from the feminist literature and the distinction drawn between patriarchy, sexism and misogyny. According to Kate Manne’s — now influential — account, ‘patriarchy’ is the term used to describe social institutions that favour men over women (i.e. sort the sexes/genders into different outcomes groups without moral justification); ‘sexism’ is the ideology that sustains those institutions; and ‘misogyny’ is the set of practices and habits (sanctions and incentives) that force women conform with sexist expectations. I like this conceptual division of labour and I have not found a similarly neat framework for discussing racism and racial injustice. Sure, there is talk about racist ideologies and institutional racism and racist policing, but the common use of terms like ‘racism’, ‘racial and ‘racialised’ to describe these different things, may encourage conflation and confusion.

I think the best solution to the problem might simply to be sensitive to the different mechanisms underlying racial injustice, without being overly committed to a single understanding of what truly counts as ‘racism’.