Implicit Bias and Scepticism

I listened to a fascinating lecture given by Jennifer Saul at the Royal Institute of Philosophy in the UK which brought up an excellent link between the psychological study of implicit bias (for which she gave numerous examples) and its implications for epistemology, in particular, scepticism. I recommend giving the talk a listen (YouTube link here) because it goes into much greater detail than I will here. I am going to focus on the fascinating link Saul made between the fact that we know we have these ingrained biases with our inability to overcome them by willpower and its implications for knowledge and rational thought.

Saul tentatively defines the term implicit bias as the collection of largely unconscious associations which people are prone to which affect how we perceive and interact with the world. The biases are not quite beliefs, they are not conscious, and yet they affect our thought processes. They are not mitigated by stated beliefs: for instance, the famous African-American civil rights activist Jesse Jackson said:

“There is nothing more painful to me at this stage in my life than to walk down the street and hear footsteps and start thinking about robbery. Then look around and see somebody White and feel relieved.” (citation)

That is implicit bias in action, and the painful experience of Jackson is not only common, it is practically universal. Unlike in Jackson's case, it is almost always sub-conscious. This presents an unfortunate sceptical problem: we have good reason to believe that our faculties are deficient when it comes to decisions that we believe are made rationally.

Jackson's example is one of a feeling produced, but others may bring out the force of the problem more clearly if they highlight how it is our conscious decision making that is affected by these ingrained biases against people of particular races, genders, sexual orientations and so forth. But the research is clear that there are pervasive and insidious cases of discrimination among self-labelled egalitarians who, try as they may to make calm, reflective decisions when choosing between biased options, still tend to make the biased on - whether it be gender, racial or even height discrimination, among countless other features to discriminate upon.

The social problem is clear in that these implicit biases produce stagnant structures of discrimination. However, there is a more philosophical problem that Saul highlights:

"I will be arguing that what we know about implicit biases shows us that we have very good reason to believe that we cannot properly trust our knowledge-seeking faculties. This does not mean that we might be mistaken about everything, or even everything in the external world (so it is weaker than traditional scepticism). But it does mean that we have good reason to believe that we are mistaken about a great deal (so it is stronger than traditional forms of scepticism). A further way in which bias-related doubt is stronger than traditional scepticism: this is doubt that demands action. With traditional scepticism, we feel perfectly fine about setting aside the doubts we have felt when we leave the philosophy seminar room. But with bias-related doubt, we don’t feel fine about this at all. We feel a need to do something to improve our epistemic situation." (citation)

I recommend giving a long and hard think to the paper she delivered (cited just above) because unfortunately, this is not a problem that exists external to us as individuals: the problem is me, the problem is you.

Laws of Nature as Patterns

An old definition of science could be that science is the study of cause and effect in nature. The idea that causality is the central idea in knowledge goes back at least to Aristotle's four cause analysis of the world and in many ways this is what people believe science does today. Instead of directly citing cause and effect, it seems to me that the intermediate notion of the "laws of nature" is interjected; science studies the laws of nature, which are about cause and effect in nature. Simple, right?

I want to propose an alternate idea based on how I view science and in particular, based on how I see modern science handling data. There's a caveat before I begin: as usual, I think much more about the canonical status of physics in science than I do of other equally legitimate branches of natural science. So what I say relates most to physics and (perhaps) the applicability slowly decreases as we move towards more qualitative sciences. Or it might not - my point is that I am making a stronger case for science as most exemplified in the hard, mathematical sciences.

Scientists do experiments and get data. This data is put often into tables or graphs and analysed, whereby models are produced which explain the correlations in the data by causal mechanisms. The model, if it is a good one, will make other predictions which can be tested to see if it is correct, and the more it passes those tests, the more credence it is given. That is the scientific method of observation, hypothesis, testing, conclusion. There are two very similar problems I see with this process: an error in data analysis and a deeper philosophical issue that Hume would have noted.

It is drilled into every data analysis/scientific statistics student that a correlation does not mean a causation and yet this explication of the scientific method clearly makes that jump. The justification is simple: eliminate as many variables as possible and the remaining correlations must be causation. That is simply mistaken and the history of science is rife with examples of deeper explanations being found of natural phenomena which destroyed the previously perceived causal mechanism.

Hume would have heartily agreed with this objection and would defiantly disagree with anyone who tried to maintain that, whilst correlation does not equal causation, a lot of correlation does, in fact, equal causation. His problem was two-fold: the assumption that the correlations of the past will hold in the future is only based on the observation that the correlations of the past have so far held true in the future. But that is circular, since in effect it says that the future is causally equivalent to the past because the past is causally equivalent to the past. But again, Hume had a deeper problem that simply the problem of induction. His biggest reason for scepticism is that causality was not a superficial relation between objects, in fact, we never really perceive causality at all, we perceive effects and infer causation. He called this "customary conjunction", or basically, correlation. For all this scepticism (and Hume unlike others branded with this title really was a sceptic), Hume did still believe in cause and effect, he simply thought it was beyond our knowledge.

I am not sure if I believe in cause and effect, but I propose right now the weaker claim that science does not study it. Science studies data to produce laws of nature which are expressed mathematically because, at bottom, the laws of nature are patterns in observable variables or parameters. In other words, the laws of nature are patterns of numbers that describe nature. Let me give an illustrating idea to stir the intuition of this proposal and consider how this relates to epistemology and the metaphysics of causality (if causality exists at all).

Think of the positive integers: 1, 2, 3, 4, ... They form a nice set whose properties can be analysed to yield fascinating mathematics. Relations can be defined on this set giving it order, taking a pair of numbers to another in the set by multiplication or addition, and so on. The tools of mathematics are about seeing mathematical structure in the integers, seeing what symmetries it might have, trying to see what patterns it has. A simple pattern in the integers is that the numbers are alternating odd, even, odd, even.

What if we tried to apply the tools of old fashioned science to the integers? It would yield fictitious language about causality to something which exists independent of causes: the fact that 2 is even is not caused by the fact that 1 is odd, even though we could conceivably speak of it that way. Many of the properties of the positive integers can be spoken of in terms of causality, but it is a fiction of our language, not a fact about the set itself.

So, I claim, it is in science. Equations like Newton's laws are mathematical statements which should not be thought of in terms of causality but in terms of patterns. It is still possible to make if-then statements: If a force is applied, then an acceleration will occur. That is a statement about what the second law predicts and codifies. It is also important to note that I am not simply saying "science produces equations that have no necessary connection to what really exists." This view is not scientific anti-realism, it is congruent with a critical realism about science.

It is important to dispel the objection that I am merely playing with words and the if-then statements are exactly equivalent to causality statements. But I refer back to the case of mathematics: it seems clear that "if you add one to an odd number you get and even number" is not equivalent to "adding one to an odd number causes an even number." The first is true, and the second uses perverted language to try and express, it seems, the first statement.

The pattern view has several advantages: for starters, it is epistemically conservative and codifies what the science actually shows rather than trying to make it jump over the correlation-causation barrier in the data. It is robust to quibbles over the meaning and nature of causality, in particular, it allows for a less stringent requirement on the necessary conjunction between a state of affairs and its antecedents; the pattern view may allow, once specified, the derivation of if-then statements, but it does not have to. Patterns can in principle be random or ordered.

It also avoids infinite causal regress problems. It seems intuitive to some (though not all) that causal chains must have a beginning (whether the causal series be per se or per accidens as Thomists would distinguish). But it does not seem to be obvious that patterns need to have a beginning: sure, the pattern of odd, even, odd, even in the positive integers has a beginning because it has a first member. But the integers have no first member because they stretch from negative infinity to infinity - and yet they still have the pattern of odd, even, odd, even. What is not definable is whether the first element was odd or even, because there is no first element.

But even if we take the positive integers, it still does not need for there to be something before the first member for the pattern to continue ad infinitum. To ask "what caused the first member of the positive integers" is a senseless question. Similarly, it may very well be the case that nature had a beginning and it is self-contained. The objection that "but it had to have had a cause, science demonstrates that!" is simply not true. The universe could be just like a set which starts at time zero and continues to infinity without  quibbling over whether there was anything at t = -1.

My Theory of Truth and Justification

My views on truth and justification (my epistemology more generally, in fact) has undergone quite a change in recent times. With my apostasy from Catholicism, I stopped having a view I had termed Christian reliabilism (sketch here), and I could have easily just moved from a peculiarly Christian version of it towards a secular form by subtracting truths that were justified on the basis of divine revelation.

Instead, I recently had my interest piqued by a pragmatic theory of truth via the physicist Sean Carroll which says that "we should always act as if A is true" implies that A is true. Whilst I previously held a correspondence theory of truth, I found it hard to justify much knowledge at all because it was difficult to see how I could justifiably access that exterior world to check if truths corresponded. But, as interesting as pragmatism was as an alternative, I strongly disliked how under-determined it was: as far as I could tell, it offered no resources for distinguishing between ontologically distinct but empirically equivalent beliefs. I could also conceive of it having difficulty with "inconvenient truths" and "useful falsehoods." Could they be smuggled in as false or true respectively when it would seem obvious that they are false? Someone who believed that free will did not exist may still hold that it is better to act as if it does, for instance, but that would imply that free will does exist (or technically, "is true").

There is an interesting feature of the definition of truth that pragmatism has which may have consequences I do not currently foresee: it is based on an ought statement. If someone believes strongly in the is-ought gap, then defining that something is true based on what ought to be done/believed is going to be a problem. Or alternatively, one can think of this definition as the beginning of a solution to the is-ought gap's related problems. It is at least conceptually odd, even if true(!), that an ought statement gives a definition of truth. If, like me, you think oughts may imply teleology, then an important question to ask would be: we ought to...if what? We ought to believe A if we want to be come billionaires? Win a Nobel prize? Colonise Mars? Have seventeen children? I would refine the imprecise definition that Carroll suggested he was partial to by adding in something along the lines of "ought to...if we want to make accurate predictions about the world" or perhaps "...of our experience of the world." There is then subsumed into the theory what we originally found to be an important part of the definition of truth, that is, that it has something to do with how the world is.

Pragmatism about truth avoids a substantial amount of scepticism which seems to follow from a correspondence theory of truth, but perhaps one of its most appealing consequences is how you can get shades of truth and falsehood. Whilst something is only true if we ought always to act as if it were true, it is relatively easy to add in provisions about partial truths by saying that they refer to propositions that we should sometimes act as if true. For instance, it is relatively easy on this view to say that classical mechanics is partially true insofar as their is a classical regime in which we should act as if classical physics is true, even though that classical regime is not the most general regime possible.

However, that under-determinism aspect of it jars at me a lot, so eventually I clicked on a compromise: we can have pragmatism as a theory of justification and keep some kind of correspondence as a theory of truth. This would say that a person is justified in believing A if they should always act as if A is true, and yet, A is only actually true if it corresponds to how the world really is. This benefits from avoiding scepticism and explaining how we can have partially justified beliefs without allowing that there are a possibly infinite number of contradictory "truths" which count as true because they are empirically equivalent.

Obviously a lot more refining needs to take place, but I think this understanding of truth, justification and their interplay may even get me out of my most begrudgingly sceptical position: scientific anti-realism. If justification via pragmatism is a reasonable position, then it seems to follow that scientific realism is a reasonable position, and as someone that aspires to work in the natural sciences, that would be great.