Personalised Criminal Law


This post is part of a special series including contributions to the OBLB Annual Conference 2022 on ‘Personalized Law—Law by Algorithm’, held in Oxford on 16 June 2022. This post comes from Rebecca Williams, who participated on the panel on ‘Personalised Law’.


In Personalised Law, Ben-Shahar and Porat claim that their proposal is different from old forms of legal contextualisation. Obviously, it is indeed different in the method of personalisation, ie the use of Artificial Intelligence and Machine Learning as opposed to human judgment, but the authors argue that it is also different in four other ways. These are:

  • that personalised law is more intensive in terms of the number of factors used for personalisation;
  • that it is more extensive in terms of how many commands can be subject to personalisation;
  • that the factors taken into account are more likely to be internal personal differences as opposed to external, environmental differences, and
  • that the personalisation can be more continuous, permitting (though not necessitating) an unbroken spectrum of variation as opposed to categorisation or classification. This variation can take place either across cases or even within a particular person over the course of time.

It is worth examining some of these potential differences through the lens of criminal law, not least because, as the authors themselves note, it is an area in which we do already have a greater degree of personalisation than can be found elsewhere. The law discussed below is that of England and Wales, and of course the precise rules in other jurisdictions will vary. Nonetheless, it provides a useful case study to compare to Ben-Shahar and Porat’s proposals, and some of the underlying principles may well have equivalents in other systems.

Both the first and third potential differences the authors claim  relate to the factors used to personalise the law. And it is here, as the authors note, that sentencing in criminal law already comes close to what they propose. But it is not just sentencing but also the substantive criminal law that does this, for example via the defence of diminished responsibility. It is true that this defence is still binary; either a defendant was acting under diminished responsibility, or they were not; it only applies in cases of homicide, and it is only at the subsequent sentencing stage that the variation becomes a more continuous spectrum. Nonetheless, it does still provide an example where internal factors specific to the defendant open that spectrum of variance through the application of the substantive law.

Similarly, the rules of intoxication operate differently based on the internal characteristics of the defendant. Someone with diabetes can become voluntarily intoxicated through ingestion of insulin not followed by food. And yet the rules applicable to that individual in the law of England and Wales are not the usual Majewski rules, but are instead the specific rules of Bailey, which recognised that for diabetics, maintenance of sobriety is effectively a positive burden which the defendant Bailey had omitted to discharge. He was thus treated more leniently than those who positively choose to become recreationally intoxicated. It is even theoretically possible, though extremely difficult in fact, to argue using the case of Tandy that an alcoholic may not count as being voluntarily intoxicated if the consumption of a particular drink at a particular time was the result of compulsion rather than choice. It may well be that with the ability to analyse more data about these defendants we could fine tune these rules even further, and that we could consider other genetic or physiological factors in doing so, but that would be more of a difference in the extent to which we personalise the rules, rather than a difference in kind.

Similarly, the authors variously discuss the potential to personalise the rules relating to alcohol purchase and driving, but this is an area where we already do have a large degree of personalisation. It is true that the legal blood-alcohol limit is the same for all drivers, and that with more data we might be able to vary that limit, but it is already the case that what different people can drink and yet remain under the legal limit will not be the same. Indeed, the Metropolitan Police website specifically states that ‘[t]here is no way to know how much you can drink and stay under the limit, since it can depend on your weight, age, metabolism, the amount of food you’ve eaten and other factors.’ If, therefore, there were to be variation not only in what people can drink while remaining legal but also the level at which they would cease to be legal, this too seems more like a difference in extent than in kind.

Now of course it could be argued that the personalisation we do have is much cruder than the authors suggest it could be, and it is stratified or binary, rather than continuous. Could we not move to a more continuous and accurate version of diminished responsibility, for example, even as a matter of substantive law, not just sentencing? Even if we could overcome concerns about the accuracy of the data we would use and our inability fully to remove matters of discrimination (on which one may not be quite as sanguine as the authors), there is a wider point that personalised law is not, ultimately, the same as personalised medicine for at least four reasons:

  • In personalised medicine there is always specific consent to the relevant treatment, something which does not apply when a defendant is dealt with by the general criminal law.
  • In medicine, decisions are always taken in the best interests of the patient. There may be questions relating to the cost-effectiveness of funding a particular treatment, but if treatment is given, there is never an aim to do other than improve that patient’s health.
  • In medicine there is usually a specific, individualising test, for example to determine the DNA profile of a patient’s tumour. There is thus a stage at which the treatment is genuinely personalised by testing to see whether the patient does indeed fall into the category of those predicted to respond in a particular way to treatment.
  • Finally, in medicine there is no real discretion about the outcome: an oncologist does not choose the fact that certain tumours will typically respond to certain drugs while others will not. The oncologist’s choice is simply whether to give the drug to those who will, in fact, benefit from it.

Where, by contrast, data is used to predict the likely extent to which a person’s responsibility was diminished, following which the compulsorily applied law chooses to treat that person differently in terms of conviction or sentencing in order to achieve a series of aims, some of them defendant-focused while others are focused on goals relating to society or, indeed, the victim, none of these four features are present.

In fact, however, to a large extent in criminal law it is possible to avoid worrying about personalising objective rules simply by avoiding objective rules altogether. English law has long debated the correct meaning of ‘recklessness’, and for a while, in the 80s and 90s the law defined recklessness objectively, meaning that the defendant could be considered reckless where the reasonable person would have foreseen harm. But then, as a result of various problems with this objective approach, among them its inability to deal with defendants who were incapable of acting as reasonable people, in 2003 the House of Lords in G&R opted to make the standard entirely subjective. Thus, the defendant will not now be liable for criminal damage unless they personally foresaw that damage to another’s property would result. In other words, the standard is entirely personalised.

But that does not, conversely, mean that capacity will always be relevant in criminal law. English law has long had a defence to murder which was originally known as provocation. To plead this defence the defendant had originally to show that they had acted as a reasonable person in responding to provocation and the only characteristics which would be admissible for consideration in deciding that would be age and gender. Then, in 2000, in the case of Smith, the House of Lords expanded this list of characteristics so that anything which rendered D’s response more excusable could be taken into account. This, however, completely undermined the basis of the defence. One might think of it as the ‘there but for the grace of God defence’; we allow people to plead it when none of us can truly swear that, faced with the same provocation, we would have been able to restrain ourselves and would therefore have acted differently. But if the argument is not that any one of us could have done the same, but rather that the defendant could not be expected to have done anything differently, then the defence simply merges with diminished responsibility as discussed above. For this, as well as other reasons, in the 2009 Coroners and Justice Act the government therefore replaced the old common law defence of provocation with a new statutory defence of loss of self control (LOSC), which goes back to allowing the defence only on the basis that ‘a person of D’s sex and age, with a normal degree of tolerance and self-restraint and in the circumstances of D might have reacted in the same or in a similar way to D’. And, arguably, age and sex should have been taken out too. Here, then, is an example of a case where even before the law could personalise as accurately as may now be possible, it chose not to do so because that would be to undermine the basis of the defence. Its conceptual basis involved a reference to universality, without which it collapsed into a different defence entirely.

And this in turn goes to the authors’ point that their conception of personalisation is different because it can be more extensive in terms of the commands to which it applies. As they themselves note, and as the example of Smith shows, the claim cannot be that personalisation can extend to all areas of law because there will be circumstances in which personalisation would defeat the purpose of the law. This means we need to think carefully about why we might want to personalise; Horst Eidenmüller, Timothy Endicott and Karen Yeung have also addressed this point, noting that certain goals can be achieved through personalisation while others cannot. The kind of personalisation at issue is also important here. While the authors’ discussion of driving and alcohol focuses on variation by capacity, their discussion of rules on intestacy focus more on variations in preference. And this is a crucial distinction in the criminal context, because while, as demonstrated, variation by capacity has always been central to criminal law (there being neither a retributive nor a deterrent purpose to holding someone to a standard they physically cannot attain), there is every purpose in holding a person to a standard they may not wish to attain. Indeed, one might argue that that is the whole purpose of criminal law.


The immediate conclusion is therefore that in criminal law at least, where we might want personalisation, we often already have it, and where we do not have it, that may well be for good reason.

But precisely because of this, the potential to personalise gives us, as is so often the case with the intersection of law and technology, the opportunity to view law in a new light and establish what it is we wish to achieve. If Endicott and Yeung are right that the purpose of the law of succession is to exercise a normative power to dispose of property, then they are also right that ‘computationally personalised... succession’ would ‘be an illegitimate departure from the rule of law.’ But if Ben-Shahar and Porat are correct that the purpose of the law of succession is to satisfy the preferences of the property holder, then personalisation would be acceptable.

If an objection to the personalisation of law is the potential breach of the rule of law’s requirement to treat like cases alike and different cases differently, then the option of personalisation requires us to decide and articulate the axes along which we regard cases as being alike or different. Thus, for example, to consider two sets of rules discussed above, in the case of drink-driving we wish to keep the rule the same (legal blood alcohol level), but this inevitably varies the burden it carries (people can metabolise alcohol differently). Conversely, the contrast between the Majewski rules for voluntary, recreational intoxication and Bailey for diabetic intoxication is an attempt to equalise the burden between those with and without diabetes through varying the applicable rules. The principal potential for personalisation is therefore perhaps not its own novelty, but rather its ability to offer us a fascinating lens through which we can thus see our existing laws more clearly.

Rebecca Williams is a Professor of Public Law and Criminal Law at the University of Oxford.


With the support of