Neuroscience-based Mind-Reading in the Law (Part One)

An image from an fMRI-based mind-reading
experiment performed by Miyawaki et al in 2008

The philosophical, scientific and legal implications of neuroscience-based mind-reading technologies has actually preoccupied me for quite some time in my “professional” life. In fact, I’ve written a couple of peer-reviewed articles about it. That said, I don’t think my own writing about the topic has been as insightful or useful as it might have been. Consequently, I thought I might start to explore some of the issues associated with it on the blog since, invariably, the more blog posts I write about something the clearer my own thinking becomes.

I’ll start off by looking at an article by Emily Murphy and Henry Greely (both of Stanford, I believe) entitled “What will be the limits of neuroscience-based mindreading in the law?”. The article appears in The Oxford Handbook of Neuroethics - which is a nice fat collection of introductory articles on all things neuroethics-y. The article provides a decent overview of the basic issues so it’s a good place to start.

Although the article itself is relatively short (as are most contributions to these Oxford handbooks), I’m going to take a couple of posts to discuss it. This is so as to really analyse some of Murphy and Greely’s more enthymematic moments. In this entry, I’ll look at the basic pro-neuroscience-based mindreading argument (what I call the NMRA) and then consider some of the problems it faces. In the next post, I’ll look at what Murphy and Greely have to say about the potential impact of mindreading on the law.

An important fact, worth flagging at the outset, is that we all engage in mind-reading everyday. Whenever I interact with another human being, I implicitly rely on an ability to decode what they are thinking by interpreting their speech, gestures and other external movements. This capacity for mind-reading is crucial to everyday life, so crucial in fact that those who lack this skill (say, autistics and young children) are noticeable for their deficit. And yet despite its obviousness, this fact is often overlooked in discussions of mind-reading (though not by Murphy and Greely, I hasten to add).

The everyday banality of mind-reading raises interesting questions in the present context. Chief among them being: why neuroscience-based mindreading? What is it about neuroscience-based technologies (such as fMRI and EEG) that makes them particularly useful when it comes to reading minds? What advantages do they confer over the more mundane sorts of mental decoding that we engage in on an ongoing basis? Answering these questions has important repercussions for how we understand the debate surrounding neuroscience-based mind-reading.


1. The Neuroscientific Mind-Reading Argument (NMRA)
Let’s look at the first question: why are neuroscience-based technologies thought to be useful for mindreading? Many people, Murphy and Greely among them, think the answer to this question lies in the implications of mind-body physicalism. The argument appears to run as follows: the mind is ultimately reducible to (or realized by) a physical entity known as the brain; neuroscience-based technologies allow us to detect the status of this physical entity; therefore, neuroscience-based technologies allow us to read minds. QED.

Attractive as this line of reasoning may be, I think it is clearly flawed. Apart from containing some important hidden premises, the relevance of neuroscience-based technologies to mind-reading is not dependent on the truth of mind-brain physicalism, nor should one construe as being so dependent simply because one is a physicalist (as I am). All that matters for the purposes of mind-reading is that there is some more-or-less reliable link between mental events and neural events. It is this reliable link, not the reducibility of the mental to the physical, that allows us to determine mental events on the basis of neural event. As far as I can tell, such a reliable link can exist even on strong forms of mind-body dualism. After all, many dualists would agree that there is a reliable correlation between the neural and the mental.

This might seem like a pernickety point, but I think it is important: there’s no sense in limiting the appeal of neuroscience-based mind-reading unnecessarily simply because one is wedded to mind-body physicalism.

This leads me to suggest the following argument for neuroscience-based mind-reading (the NMRA):

• (1) If we wish to read someone’s mind, there must be a link between observable events and mental events, such that reliably detecting an observable event of type X will help us to determine whether someone is experiencing a mental event of type Y [mental code premise].

• (2) Every neural event X1…Xn is linked to a mental event Y1…Yn in such a way that detecting the presence of X1…Xn might allow us to know that someone is experiencing Y1…Yn.

• (3) Therefore, detecting neural events might allow us to read someone’s mind.

• (4) Neuroscience-based technologies (such as fMRI and EEG) allow us to detect neural events.

• (5) Therefore, neuroscience-based technologies might allow us to read someone’s mind.

Allow me to say a few words about this argument here. First look at premise (1). This is more complex than it might first appear. For one thing, it contains within it the assumption that mental events are distinct from observable events. This seems like a credible assumption — one of the distinguishing features of mental events is that they are, first and foremost, private. But, of course, that is not to deny that they may be linked to publicly observable events, indeed, the premise explicitly demands that some such link exists. Furthermore, the premise demands that the link be a decodable one (this is a term I’m inventing), i.e. one that allows you to determine nature of the mental event on the basis of the corresponding observable event. A decodable link need not be causal, constitutive or reductive.

Now let’s look at premise (2). This proposes that there is decodable link between neural events and mental events. This premise is certainly helped by the assumption of mind-body physicalism, but as noted above it is not dependent on that assumption. Also, the premise makes no assumptions about the nature of the correspondence between mental events and neural events. Specifically, it does not stipulate that there must be a one-to-one correspondence between mental events and neural events or vice versa; it just stipulates that there must be some correspondence, this could be one-to-many or many-to-one.

This is actually an important point, one worth expanding on in a little more detail. Throughout this discussion I have spoken about the link between mental events and observable/neural events, but I have been unclear about what I mean by “mental event”. It’s probably worth clarifying this now. What I mean by “mental event” is not just an intentional state with a specific kind of content (e.g. the desire to eat ice-cream or the belief that ice-cream is good for you) but also the general category or quality of intentional state (e.g. an “unwelcome desire”, “false belief” or “malevolent intention”).

I construe the term broadly because I want to avoid implanting in the reader’s mind the belief that mind-reading requires the ability to determine the fine-grained content of person’s thoughts. While this might be ideal, oftentimes it is unnecessary. Oftentimes it is enough to know the general category or quality of mental event. For example, in the process of crime investigation, it might help the investigation to know whether a suspect is being deceptive or has a false belief, without also having to know the specific content of the deception or the belief.

Once we get beyond premise (2), I think the argument is relatively straightforward. The only concern that is worth addressing here is the claim in premise (4). Some people might argue that technologies such as fMRI or EEG do not really detect neural events at all, rather, they detect the (assumed) effects of neural events. I agree, but I’m under the impression that the link between those effects and their underlying neural causes is grounded in some pretty solid theory. The only thing I will say is that the link is a crude and imperfect one, so some important information is undoubtedly being missed.


2. Why Neuroscience-based Mind-Reading
If we accept the NMRA, we will have accepted very little. Essentially, all we will have conceded is that neuroscience technologies may help us to read other people’s minds. We won’t have accepted that neuroscience-based mind-reading is something that we ought to embrace, or ought to spend considerable resources developing. After all, I do a pretty good job reading other people’s minds by simply decoding their gestures, their words and their external behaviour. Why do I need expensive neuroscience equipment to help me in this regard? To answer that, additional arguments are needed. I’ll mention two here. The “greater accuracy” argument; and the “hard-to-fake” argument.

The first argument — unlike the NMRA itself — makes considerable concessions towards the truth of mind-body physicalism. Roughly, it maintains that neural events are somehow closer to or more representative of the underlying mental events than are the kinds of observable events we normally use to read one another’s minds. To put it another way, neural events provide more accurate information about mental events than do observable behaviours such as speech and gesture. And since we should generally prefer getting more accurate information to less accurate information, we should prefer neuroscience-based mind-reading to everyday mind-reading.

To put this in more formal terms:

• (6) In general, if our goal is to obtain information about an event or type of event, we should use the information source that is most representative of (or most accurate about) that event or type of event.

• (7) Neural events are more representative of (or more accurate about) mental events than are observable behaviours such as speech and gesture.

• (8) Therefore, if our goal is to obtain information about mental events, we should look to neural events more than observable behaviours.

I think this is a pretty weak argument. While those of a physicalist disposition will probably be more inclined to accept premise (7), I think there are good reasons for physicalists to reject it too. I think that speech and gesture are, nine times out of ten, far more accurate sources of information about a person’s mental states than are neural events. After all, that is one of the great marvels of language: it allows you to directly convey to another person what you are thinking. Now you’ll notice I said that this is true “nine times out of ten”. That’s because there are obvious cases in which speech and gesture are not good guides to what a person is thinking. This is what the hard-to-fake argument is designed to highlight.

What are the obvious cases I just alluded to? Well, they are the cases in which a person might have some inclination, desire or tendency to deceive you about the true state of their mind. In these cases, their observable behaviours will tend to mislead us or provide us with inaccurate information. These cases are particularly prevalent in the anti-terrorist or crime-investigation contexts since would-be criminals and terrorists often have strong incentive to mislead you about their true state of mind.

It should come as no surprise then to learn that these are the contexts in which neuroscience-based mind-reading are most widely touted. But why is that? The answer lies in a key assumption, namely: that neural events will be hard to fake. That is to say, a person won’t be able to control their neural events in the same way that they can control their speech and gestures, hence these events will get us closer to their true state of mind. This assumption allows us to make the following argument.

• (6) In general, if our goal is to obtain information about an event or type of event, we should use the information source that is most representative of (or most accurate about) that event or type of event.

• (9) When a person is being deceptive or has an incentive to be deceptive, neural events are likely to be more representative of (or accurate about) mental events than observable behaviours such as speech and gesture (because neural events are not as controllable as observable behaviours).

• (10) Therefore, when a person is being deceptive and our goal is to obtain about information about their mental state, we should look to neural events more than observable behaviours.

As you can see, this argument is similar to the previous one, but limits the context and provides a decent rationale for that limitation. I think this is definitely a better argument than the previous one, but it too suffers from an obvious flaw. The thing is, there’s no reason to think that neural events won’t be manipulable in the same way that speech and gesture are. Indeed, this is a something that many mind-reading tests (such as those based on EEG, or the classic polygraph test) have had to confront. Once people know how these tests work, they can often develop countermeasures that allow them to distort or corrupt the information that is being decoded by the test. At the same time, that’s not to say that countermeasures will always be easy to develop, and it may be that countermeasures are more difficult when the test is neuroscience-based. Thus, these kinds of test may still have the upper hand.


3. General Problems for Neuroscience-based Mind-reading
Now that we know some of the basic arguments in this debate, we can move on to consider the general problems that Murphy and Greely’s think affect neuroscience-based mind-reading. There are three of them.

The first is that the brain is an incredibly complex organ and so the code linking neural events to mental events is likely to be very difficult to work out. Here’s what Murphy and Greely say about this problem:

“To…associate…a particular Brain State A with a particular Mental State A, we need to be able to define both states accurately and to be confident that the detection of one indicates the presence of another…To define Brain State A in detail, though, we would need to understand the human brain and its relevant states in great detail. Such detail may well be, even in theory, impossible to obtain, both for an idealized human brain, and, for additional reasons, for any real, living human brain”

They then proceed to recite some interesting facts about the size of the brain, the number of neurons and other cells it contains, the number of connections between these neurons and the difference between idealised brains and the brains of actual people. While I accept all these points about complexity, I find the premise that motivates their presentation to be flawed. Murphy and Greely assume that we need a very fine-grained code, one that allows us to determine specific mental content from brain states. As I pointed out above, this assumption is questionable: sometimes a far more coarse-grained code will be enough (they make some gestures towards this argument later in the article, but they are not explicit enough for my liking). On these occasions, the complexity of the brain may be less of an impediment to mind-reading.

The second problem discussed by Murphy and Greely is that of neural plasticity. By neural plasticity they mean two things: (i) the fact that people’s brains can change over the course of their own lifetimes in such a way that different anatomical regions can acquire different functions; and (ii) the fact that the functional anatomy of different people’s brains can be different. That both forms of plasticity seem to exist creates many problems for proponents of neuroscience-based mind-reading. The main one being that the experimental evidence supporting (some) versions of the technology rely on correlations between the mental states and neural activity of particular individuals on particular days.

The third problem is something the authors call “the time travel problem”. It is the fact that, especially in the legal context, when we engage in mind-reading we typically don’t want to know what someone is thinking right now, instead we want to know what they were thinking at a historical moment (e.g. the time when they committed a crime, or signed a contract, or something along these lines). Murphy and Greely acknowledge that this problem might be circumvented if all we wish to do is figure out whether someone is currently lying about the state of their mind in the past, but they add that the ability to reliably distinguish a presently deceptive state from a non-deceptive state depends on how accurate someone’s present memories of their past mental states actually is. And, of course, present memories of the past may not be accurate at all.

Okay, so that brings us to the end of this post. As we have seen, the basic argument in favour of neuroscience-based mind reading (the NMRA) is more complex and resilient that we might first have thought: it does not rely on the truth of mind-body physicalism, and may not demand fine-grained mind-reading. We also saw how neuroscience-based mind reading is only likely to be useful in contexts where deception is a major problem. Finally, we looked at three general problems facing this kind of mind-reading. In the next post, we will consider how mind-reading might impact upon the law.