The Memory Illusion

by Dr Julia Shaw

  These same processes make it easier to understand police officers responsible for putting innocent people in prison. Often they will have been overconfident in their ability to tell whether the suspects were guilty or not. That overconfidence may have been fuelled by several of the illusions we have described. The illusion that their police work was better than average, or the illusion that they only ever put guilty men in prison, or the illusion that their understanding of the case was deeper than other people’s. It is these natural illusions, which can easily befall us all, that have helped contribute to many miscarriages of justice.

  Forgetting we forget

  While overconfidence can be the result of memory processes, memory processes can also be the victim of overconfidence. In other words, memory causes overconfidence which causes overconfidence in memory.

  I’ll try to make that a bit clearer. Prospective memory is our ability to remember to do things – we have already touched on it in previous chapters. It is the memory that is necessary in order for us to stick to goals and do things that are important for our future. It is like the personal assistant in our brains; our internal to-do list that reminds us that we have to go to the bank, go to the supermarket, clean the house, meet Sophie for lunch at 2pm, and so on.

  It is an amazing ability. Yet, like all of our memory functions, it is far from perfect. How many times do we think to ourselves when we have an insight ‘I’ll remember that, I don’t need to write that down’, only to realise the next day that our brains are not as good at this task as our trusty Siri. Even when we have memory aids like phones or diaries at our instant disposal we overestimate our ability to remember information and may choose not to use them. This faulty appraisal process is why we forget meetings, forget to pick up packages at the post office, and on some days end up feeling like we’re not really capable of getting much done at all. We are overconfident that we will remember, and we pay the price.

  And we pay the price in more than one way. Marketing teams know this feature of our memory and they try to exploit it; companies seem increasingly keen to capitalise on our overconfidence. Many of them offer a free trial for a subscription-based service, get us signed up, and then have an automatic billing process for subsequent months. What they are doing is banking on us forgetting to unsubscribe before their fees kick in. And it works. Repeatedly. We keep falling into the trap, presumably because we think this time we will remember to unsubscribe.

  In an article published in 2010, businessmen Jeff Holman and Farhan Zaidi8 explore the economics of this prospective memory banking. They collected data on these kinds of free trials, and found that not only do many people stick with the services; they are far more likely to stick with them if the trial is longer. In their sample, retention was 28 per cent for those who were given a 3-day trial and a whopping 41 per cent for a 7-day trial group.

  According Holman and Zaidi, ‘Firms are increasing the length of their free trials, possibly to increase naive forgetting by consumers … [they] offer extended free or reduced-price trials – of lengths running in months rather than days or weeks – ostensibly to give their new customers as much time as possible to test and experience the benefits of their products, but effectively as a way to more fully capitalize on consumer forgetting.’ Clearly not all those who apply memory science have our best interests at heart, and it seems we are consistently overconfident in our prospective memory abilities.

  There is a distinction that should be made here between predicting future remembering and predicting future changes in remembering. Our subscription cancellation failure is a prospective memory example that applies to predicting future remembering – ‘I will remember to do X.’ Research has also been done into how we estimate future change in our memories – ‘I will remember all the characteristics of X.’ While predictions of future remembering are faulty, it seems predictions of future changes in memory are even worse.

  Nate Kornell from Williams College explored this issue in research published in 2011.9 He studied 430 participants in a memory-monitoring task geared around ‘judgement of learning’, where participants are asked to estimate how well they have learned something. It is generally assumed that people’s estimate of their future ability to remember will be based on how strong their memory fragment of the information in question is after learning, so they will evaluate the strength of their memories, and evaluate stronger memories as being more likely to be recalled later. We go through this process every time we study for a test or presentation – we estimate how well we have learned particular information, and then use that as a basis to decide whether we should rehearse something again. It is a type of metamemory, a way of assessing our memory skills and predicting future remembering for a particular task.

  In this particular study, Kornell asked his participants to study word pairs either once or four times. Then he asked them to estimate how well they would perform at a test administered either five minutes or one week later. When he then compared participants’ actual and estimated performances, he found a clear stability bias. For example, for estimations regarding their performance one week later, participants guessed on average that they would be able to get 9.3 word pairs correct, but when they actually came back they only accurately remembered 1.4 word pairs. According to Kornell, ‘People act as though their memories will remain stable in the future.’ This kind of research, which has been repeated in many other learning contexts, shows that even though all of us know that we can forget things, we seem to systematically underpredict how much we will forget. To make matters worse, this effect seems to increase as time delays increase – in Kornell’s study ‘The results demonstrated long-term overconfidence: Relatively modest immediate overconfidence transformed into enormous overconfidence as the test delay increased.’

  In another study demonstrating this phenomenon, published in 2004,10 Asher Koriat and his colleagues at the University of Haifa in Israel showed that participants estimated that their memory would be essentially the same at immediate recall as one year later. So, we are already bad at estimating how much we will remember in the near future, and seem to be even worse at estimating how much we will remember in the distant future.

  What can we do about our tendency to forget that we forget? Kornell has some direct advice for students: ‘If today is Friday, and you feel ready for the test you have to take on Monday, don’t take the weekend off. You might be right – maybe you are ready now. But that doesn’t mean you’ll still be ready by Monday. Indeed, you’re likely to be way overconfident.’11 And, for the rest of those everyday adult tasks we want to accomplish, ‘Don’t trust your memory. If someone asks you if you can remember something, say no. Write it down.’

  Identity crisis

  While some people are great at remembering faces, and others are great at remembering names, I am good at neither. If I ever meet you, I’m sorry in advance. I will probably introduce myself to you repeatedly on different occasions. This will almost certainly confuse you; we may have shared wonderful conversations and wine. I may even reference your own conversation or research back to you, without realising that you are its source. So, why is my memory for these kinds of interactions so terrible?

  Well, there are individual differences in the ability to recognise faces. Not just in terms of remembering them, but being able to look at them and map their features – in the way that enables you to look at a photograph of someone, and the real person in the flesh, and say ‘These two faces are the same.’ It turns out that our ability to recognise faces is actually the responsibility of a specific part of the brain, which has been named the fusiform face area. It is located approximately above your ears, relatively close to the surface of your brain.

  In 2011 Nicholas Furl and colleagues at University College London published a study about a group of people called prosopagnosics. Prosopagnosia is the inability to identify faces, and is sometimes referred to as ‘face blindness’. Furl and his team found that the fusiform face area was far l
ess active in prosopagnosics than it was in non-prosopagnosics.12

  Neurologist Oliver Sacks wrote a very successful book, first published in 1985, which took its title from a case of prosopagnosia – it was called The Man Who Mistook His Wife for a Hat.13 The title comes from a case study that Oliver worked on, where the man had severe impairment in his ability to recognise his own wife. This title initially sounds ridiculous, but one of the key features of prosopagnosia is that individuals who exhibit it have to process faces just like any other type of foreign object, and they generally seem to do so piece by piece. The rest of us have an innate ability to process faces as a whole, but prosopagnosics lack this. We think this face = Emily. They may instead think small nose, big eyes, small ears, familiar voice = Emily. Apparently about 2.5 per cent of us suffer from a face-processing deficit like this.14

  In 2009, Richard Russell and his colleagues from Harvard University15 found that there are also people at the opposite end of the spectrum. According to them, these ‘super-recognisers’ are ‘about as good at face recognition and perception as developmental prosopagnosics are bad’. What is more, this seems to be both a perceptual feature and a memory feature. Super-recognisers sometimes report being able to identify and remember faces years later.

  According to one such super-recogniser, referred to as CS, ‘It doesn’t matter how many years pass, if I’ve seen your face before I will be able to recall it.’ We currently don’t know the prevalence of this ability, or exactly how it works. However, because the term ‘super-recogniser’ has recently gone viral, the research on this ability will likely see some tremendous growth in the next few years.

  One area where this ability has obvious applications is policing. People like Josh Davis at the University of Greenwich16 work with the Metropolitan Police in London to identify and employ such super-recognisers to sift through thousands of photos and make identifications from CCTV footage. In other words, because of their ability they can look at crowds and other complex footage and find a particular face, such as that of a suspect, a task that is incredibly difficult and inefficient for people who do not have this superpower.

  There is a test, called the Cambridge Face Memory Test, released by neuroscientists Brad Duchaine and Ken Nakayama in 2004, that can help identify whether or not someone is a super-recogniser. In this test participants see a face from three different angles in the ‘study’ phase, and then have to identify it out of a line-up of three faces afterwards. This is repeated over many faces. The faces in the test line-ups get more and more similar as the test progresses, making the task more difficult. Super-recognisers are able to correctly identify most of the faces in this test, and in similar tests, and they have been responsible for some tremendous police successes. For example, super-recognisers were heavily involved in identifying people involved in the 2011 London riots, identifying significantly more culprits than facial recognition programs.

  This skill is particularly valuable, because apart from such super-recognisers, many people have a difficult time reliably matching faces and photos, as David White from the University of New South Wales and colleagues showed in a 201417 study: ‘Photo-ID is widely used in security settings, despite research showing that viewers find it very difficult to match unfamiliar faces. Here we … ask officers to compare photos to live ID-card bearers, and observe high error rates, including 14 per cent false acceptance of fraudulent photos.’ Of course, in line with our lack of insight into our own personal memory biases, most people think that they can identify whether the person standing in front of them is the same person as the one in a photo they are holding. In reality, however, it seems that facial recognition is just another area where many of us are overconfident in our abilities. For most of us, perception and memory seem to interact in a way that can make it difficult for us to solve even the most seemingly basic of identification tasks.

  In an ideal world, as well as the police having enhanced abilities to identify suspects, witnesses to crimes would also be able to describe and identify the perpetrator with ease. In such situations the police want us to make assertions with utter certainty. They do not want us to say that the offender maybe had a scar, could have had brown hair, or was between 5'7" and 6'10". The desire for confidence and clarity in identifications of this kind is understandable, but such expectations can also cause confidence judgements to become skewed.

  Our intrinsic assessment of the quality of our own memories can come into play here. It may seem an obvious thing to say, but if we think we have a good memory of a person, we generally have high confidence when asked to call on that memory. But as we have seen, just because we think we have a good memory of something does not mean we necessarily do. So, leaving aside self-assessment, how good are we actually at identifying strangers?

  It seems like a straightforward question, but it turns out that there are an almost infinite number of variables that need to qualify this question. How good are we at identifying faces? How good are we at identifying height and body shape? How good are we at identifying people with scars or deformities, people of a different ethnicity, people we only have brief exposure to, older or younger people, people we saw in dim light, people we only paid some attention to, people wearing hats?

  In 2013, Matthew Palmer and colleagues at Flinders University18 conducted a study looking at some of these complexities. They had researchers pair up and take to the streets. Researcher 1 would recruit a participant, and have them provide consent to participate. Then Researcher 2, who had been hidden, would step into view. Researcher 1 would then tell the participant to look at Researcher 2 until they stepped out of sight once more. Participants were then asked to identify Researcher 2 out of a photo line-up and to rate their confidence in the decision, with half being asked to do so immediately and half being asked to do so about a week later.

  As we might expect, participants were better when they made their line-up identification immediately – when they did this they were accurate 60 per cent of the time, and if they made it a week later they were accurate 54 per cent of the time. It may strike you that these rates are pretty low. Indeed, almost half the participants failed to correctly pick the photo of the person they just saw.

  What makes this even more disconcerting is the considerable differences in accuracy that result from some pretty basic alterations to a situation. The researchers found that while accuracy and confidence were generally aligned, overconfidence was higher in more difficult conditions. In other words, participants had disproportionately high confidence if they were allowed only a very short viewing of Researcher 2, had a longer delay before their line-up identification, and/or had to divide their attention during the task. So we seem to generally overestimate how good we are going to be at identifying perpetrators in situations where the odds are particularly against us.

  Needless to say, this is a complex topic with many amazing people doing amazingly interesting research to help combat both our built-in, and our externally facilitated, memory illusions.

  Race face

  Another factor that affects our ability to identify others is ethnicity. If you are black and witness an East Asian person committing a crime, good luck getting it right. The same is true for any combination of ethnicities – white, black, East Asian, Indian, Puerto Rican, it doesn’t matter. We are simply generally worse at identifying someone of a different ethnicity, a phenomenon known as ORB – own-race bias. These cross-race effects present a tremendous problem for the legal system, as clearly people of different ethnicities commit crimes against one another. And what is worse, the system is constantly battling what looks like everyday systematic racism.

  Perhaps we are all just racist, even if we don’t admit it. Or, perhaps there is something else going on here. ORB has been heavily studied, and one of the prevailing assumptions is that it has to do with how we remember faces.

  According to Caroline Blais and colleagues from the University of Glasgow,19 culture shapes how we look at faces. In 2008 they publ
ished research that demonstrated this using eye-tracking technology to give an indication of what a subject was looking at, and therefore presumably processing. The participants in the study were presented with photos of Western Caucasian and East Asian faces to look at. All the participants self-identified as Western Caucasian or East Asian themselves.

  They found that if the participant was Caucasian, they generally used a triangular pattern to look at the faces. They looked at eyes, mouth, nose, and then explored other parts. For East Asian participants there was a different pattern; they focused far more centrally, seemingly mostly looking at the nose. And they did this regardless of the ethnicity of the person in the photo. The researchers interpret this as a culturally derived, suggesting that ‘direct or excessive eye contact may be considered rude in East Asian cultures and this social norm might have determined gaze avoidance in East Asian observers’.

  Alternatively, different search strategies may focus on things that are typically varied within a certain culture. For example, it makes sense to spend time encoding eye colour in Western Caucasian groups, since there is significant variation, while it may not be as useful as a distinguishing feature for non-Caucasian groups. Whatever the underlying reason, this research demonstrated that participants had culturally influenced facial search strategies that could lead them astray when looking at or identifying foreign faces.

  So, inappropriate allocation of attention to particular facial features appears to be one of the main reasons for the existence of ORB, at least according to Blais and her team. Focusing on the ‘wrong’ characteristics makes it harder for us to identify and remember exactly what a person looked like, and we are more likely to do this if the person is a different race from us. This flows into other person characteristics as well. For example, in ethnicities that have less variation in hair colour or height, encoding these is hardly useful.