Expunged ‘by design’: on the potential of AI to be a partial enabler in the expungement process

ABSTRACT

Expungement, which traditionally permitted reintegration into society and the turning over of a new leaf – thus giving a second chance to people who had had the misfortune to fall foul of criminal justice systems – is a legal institution that is currently being undermined. As the cyber-vetting of potential employees is a growing trend, employers readily turn to the internet in order to seek information about job candidates. In this article, the role of expungement in legal systems and society will be examined, as well as the changing reality that is undermining expungement, in order to determine how the issue of expungement should be addressed. In particular, the article discusses the possibility of expungement being incorporated in search engines (including AI chatbots) and in the actual design of AI recruitment tools in order to facilitate the re-entry of ex-offenders into the employment marketplace.

KEYWORDS:

• Artificial intelligence
• HR
• expungement
• criminal record
• employment

1. Introduction

New IT developments and data practices have brought about undreamt-of realities. Instantaneous access to data and the ability to store and analyse it have transformed not only business operations, personal and social life, but also many legal processes and institutions, be it policing, parole boards, the management of justice, court hearings, etc. While some traditional legal solutions seem to evolve and even flourish by incorporating these changes, others are being undermined by them.¹ The technological challenges faced by regulators are multispectral.² Although the changes caused by the digital revolution may be unavoidable (and indeed often commendable), new developments in certain areas can be a matter of great concern. This is particularly so in the case of the legal institution of expungement, which is currently in jeopardy. Expungement traditionally allowed people who had fallen foul of criminal justice systems to turn over a new leaf. It was thus an important means of reintegrating ex-offenders into society. The undermining of this institution can therefore have many negative consequences – not only for the individual concerned, but also for society as a whole.

In this article, we examine the institution of expungement in order to establish whether it should indeed be salvaged – i.e. whether it has any decisive merits that would call for its restitution in spite of digital developments – and, if so, to propose some means whereby at least certain elements of expungement could actually be incorporated into current social reality. In order to do this, we analyse: first, its inherent values; secondly, the hazards posed by its current environment; and, finally, we try to put forward a modest proposal that would enable a partial inclusion of the effects of expungement into current practice. Although regulatory systems, IT penetration levels, social circumstances and the current magnitude of the problem differ across States and jurisdictions, the undermining of expungement is essentially a universal development that is likely to become worse as years go by. In this article, we describe a visible overall trend and attempt to devise a means of response that would be consistent with developments in IT and data practices.

2. Expungement – or giving somebody a new chance

In a book written in 1963, Becker pointed out that one need only commit a single criminal offence in order to be labelled as a criminal.³ Generally speaking, this leads to the treatment of the person in question as an outsider or deviant ⁴ and is a kind of ‘master status’ which defines the individual.⁵ A silly mistake made in one’s teens could end up affecting one’s social status and prospects for the rest of one’s life, which would be contrary to the fact that we all change, mature and (hopefully) become more sensible, as well as being contrary to the fact that – for the most part – crime is not an inbuilt predisposition, but rather the outcome of many factors.

While crime can be a one-off thing, the consequences of status attribution are much more lasting and can have a severe impact on people’s lives. Official indications of criminality or criminal ‘branding’ bring about consequences that may well be much more severe than the punishment itself (i.e. imprisonment or a fine). It has been pointed out that in many cases, the most severe and long-lasting effect of conviction is not the punishment itself, but its collateral consequences, involving – among other things – an actual or potential loss of civil rights, parental rights and public benefits.⁶ To this we might add health deterioration, chronic unemployment and the disruption of family and community networks.⁷

Recognising the need for social rehabilitation, modern legal systems have by and large allowed for the possibility of removing the stigma of ‘criminality’, which amounts to erasing an individual’s criminal record, thus allowing him or her to turn over a new leaf. This so-called expungement is – generally speaking – a process whereby the records of a conviction (or in some countries also those of an arrest) are considered to have been erased or sealed.⁸ The understanding of this legal device differs from country to country and may be seen from a more technical perspective as being nothing more than a procedural means. It can, however, have a deeper basis and a deeper meaning.

Among other things, expungement can be seen as being an instrument of forgiveness that is embedded in the law. Forgiveness is treated as being morally good – as a virtuous thing to do.⁹ However, it is not an easy concept to explain, as its understanding varies over time and across cultures.¹⁰ Enright and Kittle claim that it is a ‘merciful act’ – a gift to someone who may not always be deserving.¹¹ Bibas for his part puts it this way: ‘Forgiveness, while acknowledging the wrongfulness of the act, separates that act from the actor and paves the way for the offender to return to the moral fold.’¹² Attention is drawn to the fact that forgiveness ‘promotes individual and community development, mental health, well-being and liberty.’¹³

As we have seen, forgiveness is important not only at the individual level, but also at the level of the whole community.¹⁴ Indeed, Mayers claims that the basic trust that makes a community possible is grounded in forgiveness.¹⁵

We can also discuss forgiveness in the context of the law, both at the level of the individual and at the level of the community. Forgiveness at the interpersonal level is easier to explain, but forgiveness by the State is also possible. A particular crime is not the only relation between a perpetrator and a victim, as harm is also suffered by the community itself.¹⁶ This allows us to think about forgiveness as being something that could be done both by the victim and the State. In criminal law systems there are examples of acts of forgiveness that – from the perspective of the perpetrator's legal situation – are relevant to both individuals and society. Forgiveness by a victim or a victim’s family can halt criminal proceedings (if they need to be initiated by the victim) and may also have an impact on the sentence. Reconciliation of the individuals involved might not only mitigate the sentence, but also even save a life. Under Islamic law (on the example of Iran), it is possible to save someone from the death penalty if the victim's family forgives the perpetrator.¹⁷ There are also examples of forgiveness by the State. Pardon may be seen as being forgiveness that is embedded in existing law, though it is not limited to that function.¹⁸ Forgiveness is a particularly conspicuous element of the restorative justice movement.¹⁹ However, expungement also has a ‘forgive nature’.²⁰ We allow people who have had convictions to move on with their lives and to live without bearing the stigma of criminality.

Undertaking attempts to mitigate the effects of the visibility of criminal records on the lives of those who have had convictions does not necessarily mean that such mitigation will be acceptable only to proponents of the utilitarian view of criminal law. It will also be coherent with the views of retributivists. The ultimate goal of the utilitarian view is a fall in the number of crimes in the future. The psychological effects of forgiveness and the effects on criminality of having a job suggest that the more forgiveness there is now, the less crime there will be in the future.²¹ Neither should proponents of retributive approaches towards criminals be opposed to forgiveness, especially when the punishment has already taken place.²² There is no doubt that the long-lasting effects of crime are collateral,²³ though such considerations are not usually included in the sentencing judge’s calculations. The significant consequences of a sentence may well extend beyond the realm of law, as information on criminality ‘[…] can drive decision-making by non-state actors when determining whether to give someone a second chance.’²⁴

Forgiveness should not be understood as a lack of reaction to crime, but rather as a shift in attitudes towards people who have had convictions, calling on the power of the State to create a new status for them in society. Martha Minow has said that ‘forgiveness involves a shift in attitudes and relationships, not merely relinquishing the authority to punish.’²⁵ Expungement could be seen as a change of attitude towards ex-offenders on the part of the State, which presents them with a ‘clean slate’, so to speak.

Moreover, expungement can arguably be seen as an official institution acknowledging the change and acknowledging a new state of affairs. Bearing in mind this permanent transition, many social constructs have incorporated this change, be it the education system adapting itself to the changing mental capacities of children, be it social security systems catering for ‘just in case’ situations, retirement pensions, etc. Interestingly, an acknowledgement of change can also be found deeply embedded in the legal institution of expungement, which traditionally permitted people to ‘shake off’ the stigma of criminality and turn over a new leaf, leaving their more or less shameful past behind them.

The possibility and ability to change – for the better, of course – is the logical reason behind the idea of the second chance. The acknowledgement of the fact that people can evolve and turn over a new leaf also has a long tradition and underlies other social constructs, e.g. re-education or parole. The example of Australia, which initially was (in part) seen as a destination for ex-convicts, speaks volumes.

Change creates a new reality which allows society to forget. Forgetting is anchored in the human condition and has been part of social reality for centuries. It greatly facilitates the giving of a ‘second chance’ as well as facilitating certain social constructs that have an inbuilt change factor. Expungement can thus be seen as a licence to forget – an official delimitation after which certain facts should not be brought up again. It is precisely this right to oblivion, however, that is currently being undermined by the ‘eternal memory’ of computer systems, which amounts to a perpetuation or artificial petrification of certain states of affairs. This creates an inhuman environment in which nothing is forgotten and – in effect – nothing is forgiven. This perpetuation of certain facts makes second chances hard to come by (to say nothing of forgiveness), as it is much easier for an internet ‘crowd’ to simply attach a label to someone and not think about taking any positive action.

We can therefore see that expungement is not merely a technical procedure for the removal of publicly available data on an individual’s previous criminal activity, but is something much more, as it acknowledges a new reality and creates a legal fiction according to which a conviction is held never to have taken place. The individual can therefore quite legally maintain that he or she has never committed a crime. While criminal systems and criminal labelling create a differentiation and distinguish between ‘law-abiding citizens’ and ‘criminals’,²⁶ expungement officially allows an individual to be re-integrated into the former category.

Looking at this institution from such a perspective, we can see that the State is hoping that the criminal records will never be made public in the future and that the individual in question will be treated by the community as being innocent of any criminal offence. According to Calvert and Bruno, while expungement statutes can differ across jurisdictions, the main rationale behind them is the same: their aim is to help in the process of the re-integration of ex-convicts into society.²⁷

Being rooted in forgiveness, expungement is an institution that aims to help the ex-offender start a new life. It entails a commitment that the sentence will not affect his or her position in society. Although this may not be fully achievable, the State should take all possible measures to fulfil that commitment. In other words, the expungement – in some legal systems, at least – may be considered to be a right of the ex-convicted that could reasonably be expected to be enforced. Given that criminal status is imposed by the State as a result of official criminal proceedings, the continued existence and resulting consequences of this status can also be removed by the State. Expungement reverses those consequences (for the individual) which are the result of official labelling imposed by the State.

3. The stigma of criminality in the new digital reality

Despite its intrinsic value, expungement is becoming more and more difficult to achieve in today’s new digital reality. People’s lives are becoming more and more visible to others and data on individuals is now much more abundant and accessible. With the growing shift towards on-line activities by people, governments and other actors, to say nothing of the growing digital presence of individuals on the internet (together with the data generated as a co-product of our reliance on various IT devices), the total volume of available information is not only rapidly multiplying, but – to a greater and greater extent – is also being taken into consideration in various social and official areas of life.

The non-availability of the effects of expungement can be quite disadvantageous as far as social life and personal life are concerned. However, one crucial problem for both individuals and society lies elsewhere – in the sphere of employment.

To begin with, abundant research shows that having a criminal record leads to a radical diminishing of one’s chances of being employed²⁸ and it has been said that a criminal record could even be seen as a kind of negative curriculum vitae or résumé.²⁹ In their recent empirical study, Sugie, Zatz and Augustine found that employers’ aversion to records is partly based on the stigma that results from the fact of having come into contact with criminal justice.³⁰

The situation in the United States can serve as a sort of reference point (or illustration), showing how the availability of data on criminal records can become a serious societal issue leading to severe stigmatisation and decreased employability³¹ – a problem that is so common and widespread that it has had to be addressed by the powers that be: in the United States, laws have had to be passed in order to make it easier for candidates with criminal records to find jobs – among other things, in certain circumstances making it illegal to ask certain questions or to discriminate on the grounds of a previous criminal record.³² It has also been said that in the United States the movement for judicial reform basically amounts to a ‘search for effective policy levers to mitigate the re-entry barriers faced by people with criminal records.’.³³ The need for re-entry is not merely based on ‘compassionate’ grounds, but also has serious societal implications: it has been noted that ‘[t]he greater the discrimination against those with prior records, the more likely the ex-offender will recidivate..³⁴

Unwanted as it may be, it is not so much the mere coming into contact with criminal justice that is the turning point in the process of stigmatisation in the eyes of the employer, but rather the availability of this information. The ways in which information about criminal records is made publicly available vary from country to country.³⁵ While in some countries personal data about offenders (or suspects) is largely anonymized, in others it is not. However, given current digital developments and the resulting problems associated with the anonymization of data or other non-disclosure policies of the State, the latter may not be able to guarantee that such sensitive information will indeed remain ‘undisclosed’.³⁶ It has been rightly noted that ‘[b]y its nature, digital information is infinitely transferable and hard to control. This openness has far-reaching effects on personal privacy, reputation and self-expression.’³⁷ Information – including untruthful information – can therefore be obtained from other sources. The shift to a greater digital sharing of information and to more advanced searching technologies can shake the very foundations of privacy and thus the State policy of non-disclosure may be rendered ineffective. This can also happen in the case of criminal records, with all the observed negative labelling consequences that lead to diminished employability.

What is also important is that more and more employers are basing their decisions not only on their reading of letters and neat curricula vitae, but also on information gained by digital vetting, or so-called cyber vetting.³⁸ This is a rapidly growing trend which is quite understandable, for why would an employer put his trust in information supplied by the candidate if he or she can check and screen applications by doing some on-line research, thus adding an extra dimension to the whole process?³⁹ According to information gained in 2017, a very large number of employers routinely use Google and other search engines to run checks on candidates.⁴⁰ Another important source of information are the social media.⁴¹ Even if requesting information about somebody’s past criminal record is legally restricted in some jurisdictions, there is no certainty that potential employers will not find out about it anyway – not only through legal sources, but also by doing online checks such as running a search for the name of the candidate in search engines such as Google, which, though forbidden in some jurisdictions, can hardly be shown not to have been undertaken.

While information sought by employers includes items that would verify a candidate’s qualifications and his or her professional ‘online persona’, they also (interestingly) include what other people are saying about the candidates and any ‘reason not to hire’ them.⁴² This rapidly increasing volume of information constitutes a good hunting ground for the verification of candidates and for making differentiations between them. Although the accessibility of information – as well as the legal framework for its use – varies from country to country (depending on the extent to which the internet is used, local culture and incumbent regulations), cyber vetting is increasingly becoming part of the process of pre-employment screening.

Employers seek and acquire information about the good and (subjectively or objectively) bad qualities of candidates, the latter perhaps being of even greater interest, as it allows employers to eliminate all potentially ‘undesirable’ members of the work force at an early stage of the recruitment process. Such information, however, may include items that could be used to discriminate against candidates on an unlawful basis and thus severely harm their chances of getting a job.

It so happens that criminal background checks in the labour market have been on the rise all over the world.⁴³ This is an important factor which influences the employer’s decision to hire someone, for it has been noted that ‘[c]riminal records are widely (but exaggeratedly) regarded as a reliable indicator of character, reliability, integrity and self- discipline’⁴⁴ – from which we may infer that the ability to acquire information on criminal records is something that is very much sought after and is HR-relevant. The discovery of information can have far-reaching consequences for those people who do have a criminal record, this obviously being one of the major ‘reasons not to hire a candidate’.

Here we would like to stress that while in certain countries there are laws that limit or otherwise structure the use of information on candidates or the use of AI (in employment amongst other things), this shield is no more a fully adequate means of response than the famous ‘right to be forgotten’ (or other similar devices). Laws structuring the use of information or the HR process can sometimes be simply bypassed by outsourcing the process abroad. Also, it is quite easy to use ‘publicly’ available information on the internet and for HR companies such ‘mining’ of information is standard procedure. As far as the ‘right to be forgotten’ is concerned, the existence of professional digital information hunters and the present state of advanced technologies mean that the hiding of certain results by the major search engines (for example) is no more than a slight inconvenience. While the ‘right to be forgotten’ requires positive action on the part of the person concerned, he or she may well be quite unaware of the existence or circulation of such information. Moreover, in cases where there are very many results, the person in question may not be in a position to find or assess the information. Another problem is that the scope of the right to be forgotten is somewhat limited and in certain cases such action may not be effective. Yet another issue is that certain web-pages (for example) are no longer maintained, and so any request to remove information from them may simply be futile. It also happens that sensitive or false information is supplied and maintained on the internet simply in order to victimise someone or extort payment in return for its removal. The ‘right to be forgotten’ may be helpful in eradicating someone’s criminal past, but – unlike expungement – it does not work ‘automatically’. If there was a request to the official register for information about a person’s criminal record, expungement would be applied automatically, as any information provided by the register would have to comply strictly with the stipulations of the expungement. If there was a past conviction that had been expunged, then the document would state that the person in question had had no previous conviction. The internet, however, works differently.

Having said all this, from a purely theoretical point of view it would seem that – given the reasons that lie behind it – expungement will begin to gain greater favour and will be in greater demand than it ever was in the non-digitalised world. One important study (from the United States) has found that the easing and expansion of the availability of expungement – even making it automatic after a certain period of time – should be a policy goal.⁴⁵

It has been noted, however, that current digital data practices greatly hinder expungement and ‘even potentially facilitate a market for expunged criminal histories.’⁴⁶ Laudable and needed as it is, the expungement procedure is increasingly seen as being a ‘paper tiger’ – or yet another unfulfilled promise of the digital age.

The law is simply unable to guarantee a fully effective ‘disappearance’ of the sentence (or criminal record) from the public sphere (in this case the internet). This is particularly the case in countries where criminal records are public and where all the data is made freely available on the web. Although efforts can be made to somehow enforce ‘disappearance’, the information might still be available somewhere or might somehow be reconstructed or inferred. This problem, however, also constitutes a considerable future threat to other systems. Information that is available on the internet can serve to de-anonymize ex-offenders, thus creating a situation which is similar to that in the United States.

In short, expungement plays an important role in social life and in the re-integration of ex-offenders into society. It could even be seen as a certain crime prevention measure. As it is a promise that is made by most modern legal systems, it has the backing of the State. It encompasses forgiveness, an acknowledgment of change and officially sanctions the act of forgetting (these last two actions being a reflection of the natural state of human reality). While the imposition of criminal branding on an offender is an acknowledgment of his or her negative actions and attributes, expungement is the acknowledgment of a change for the better and allows an individual to be officially re-integrated into law-abiding society. Given today’s ‘name and shame’ culture, social media lynching and the internet crowd mentality, expungement is needed more than ever, but – alas – is being systematically undermined.

4. The use of AI in recruitment – a new face of expungement (?)

While traditional responses and State reactions may be insufficient to uphold the promise of expungement, the continual development of IT offers new possibilities, i.e. the hope that a solution to the problem could be found in the actual search engines and AI tools that are now increasingly being used in the recruitment process. Given the growing role of search engines and AI in employment screenings and HR, we would like to propose a partial solution, namely expungement by design. As Brownsword has put it, the technologies themselves may be made to function as ‘regulatory instruments’.⁴⁷

In recent years, the use of AI in HR has been on the rise and is a trend that can be greeted not only with hope, but also with some concern.⁴⁸ The uncontrolled, ever-growing and ever-developing volume of internet data concerning individuals can have grave consequences for their future chances of gaining employment, for – as has been noted – advanced data analytics and practices have already found their way into HR.⁴⁹ A move in the direction of automated scanning and decision-taking has already been made: it has been reported, for example, that new start-ups offer advanced screening technologies that distinguish between ‘right’ and ‘wrong’ candidates by mining data on their family histories, negative attributes (such as sexual or drinking posts) and other potential risk factors.⁵⁰ With the ever-growing use of AI in most spheres of life – HR included – this vetting (and the decisions taken on its basis) can be (and increasingly is) based on or aided by the use of artificial intelligence.⁵¹ Needless to say, the depth and accuracy of such vetting and decision-making will also be rapidly increasing.

It is important to note that – if left unrestricted – the search results from search engines and AI vetting and the ensuing selection of commercially sought-after candidates are largely bound to be similar to the results of traditional vetting, as the ultimate goal (i.e. information that is of interest to the employer) is the same. As criminal records (or other signs of perceived ‘criminality’) have a highly negative influence on employability, the use of new technologies could further worsen the already precarious situation of the candidates in question (in those countries where criminal records are public), as well as hindering the anonymity system in other countries.

The use of AI in HR (understood in a wider sense) has lately been in vogue. AI-based HR products are offered on quite a wide scale and nowadays various companies readily use this kind of tool in order to ease the burdens of the recruitment process, which is not only time-consuming, but is also a drain on resources. The use of AI offers hope that the recruitment process may somehow be simplified and automated. However, the use of AI in HR has many potential shortcomings which largely reflect concerns about the use of AI in general.⁵² AI-based HR has lately had a fairly bad press, with concerns being voiced about the fact that it could lead to discriminatory outcomes (as in other areas where AI has been put to use).⁵³ With the further development of the use of AI solutions and its ever-widening scope, political concerns have been voiced and there are now signs of a movement to rein in rogue or ‘suspicious’ AI in HR.⁵⁴ In certain cases AI has been referred to as being a new ‘snake oil’ product not only promising undeliverable results, but also having a strong potential for discrimination and non-transparency.

Although AI vetting and recruitment programmes do have many drawbacks and can prove to be problematic, in this article we would like to draw attention to the possibility of using it for the greater good. In this section, we propose something that is at least a partial solution to the problem and that could/should be embedded in the AI tools used for HR. Our proposition is ‘expungement by design’. This would be consistent with the ever-growing use of AI in HR. We believe that this will not necessarily require a change in the law, but merely an adjustment of HR companies’ practices in favour of the concept of expungement. While there are other ways in which a similar effect could be achieved – for example, some States pass special laws which preclude the background checking of applicants in order to ensure a lack of discrimination on the basis of their criminal records⁵⁵ – expungement by design could arguably be more effective and consistent with those very IT developments that would ‘normally’ undermine it. Furthermore, it could be connected with or based on regulatory actions. The powers of the State, which has hitherto been unable to guarantee effective expungement, could now regulate this sphere in order to guarantee the incorporation of expungement into the AI-based recruitment process.

The term ‘expungement by design’ is inspired by the idea of ‘legal protection by design’ presented by Mireille Hildebrandt.⁵⁶ According to Hildebrandt, the idea of legal protection by design ‘[…] has arisen from the mutation of the law’s environment, which has moved from language to computation when it comes to the framing of reality. Today, much of the framing is done or prepared by data-driven machines that feed on our behavioural data.’⁵⁷

The AI tools used in the context of employment interfere with the legal promise – based on forgiveness – that old transgressions will not influence the remaining life of the ex-offender. It is precisely because these tools are designed that we propose that expungement be incorporated in them in order to be consistent with the legal fiction that a person whose criminal data has been expunged can be treated as having had no criminal record. When expungement is an individual’s right, other actors should respect the legal situation, as they too function in a reality shaped by law. In a concrete situation, actors using search engines and AI instruments for HR should respect the rules and act accordingly in a spirit of forgiveness. This will not only be morally good, but also lawful.

Expungement can affect the legal situation of the ex-offender: for example, an expunged sentence will be treated as being non-existent by the judge who imposes a punishment for a future crime. However, it should also have an impact on social life. To give but one example of a specific legal system, in the Polish criminal law literature that deals with the institution of expungement it is commonly accepted that this institution should affect both the legal realm and the social life of the ex-offender.⁵⁸ One practical way in which the impact of the criminal label on a person’s life is limited by Polish law is the non-disclosure of the accused person’s personal data in media reports, something which is now partially ineffective because of the way in which the internet operates.⁵⁹

The proposition presented here, which is expungement by design – embedded in search engines and in the HR tool itself – is a natural development of this view. It is also an attempt (at least in part) to restore the effectiveness of expungement to what it was before the digital era.⁶⁰ An expunged criminal record should not be taken into account in preparing a report on the candidate, even if the tool does find such a record on the web. Neither should the records be revealed to any person. Expunged records should be treated as being non-existent. We believe that it is right to demand that HR companies respect the legal fiction that the person under scrutiny has a clean criminal record, even if he or she has had a previous conviction that has subsequently been expunged.

In its basic form, our proposal is extremely modest, as it is limited to the criminal data which is actually expunged. However, analogous rules or improvements could be incorporated in order to promote employment among people with convictions, where conviction still has not been expunged.⁶¹

Of course, this is easier said than done. As the technical side of such expungement might prove to be somewhat tricky, we shall now present certain thoughts on the practical implementation of such a mechanism.

There are situations in which this would seem to be relatively easy. In some jurisdictions, expungement could be automatic, coming into effect ‘by the force of law’ after a certain period of time has elapsed, depending on the crime. This could be, say, three or ten years from some defined moment, e.g. the moment the penalty has been completed. If, therefore, there is information on the web about the fact that some person was released from prison after serving a sentence for a certain crime for which the expungement period is three years and such information was published ten years ago, then it should be hidden and the person should be reported as having had no previous convictions.

We now come to certain obstacles that militate against this institution being fully embedded in the AI tools that are used in HR. The most obvious issue is the fact that expungement laws differ substantially across jurisdictions. The same information can be treated as having been expunged by the force of law in one country and as being available to the public in another. HR companies which use AI tools would have to take these differences into account.

Another problem is uncertainty as to whether a conviction has been expunged or not. For example, there could be information on the web according to which a court of the first instance sentenced a person for drunk driving five years ago. Even if the institution of expungement by the force of law exists in the country in question, on the basis of such information we cannot be certain that this sentence has been expunged or not. There could have been an appeal against the sentence and the criminal proceedings could thus still be pending.

A separate issue is arrest, which – although it may be the first step towards a conviction – might just as well be a mistake, with the person in question being released from custody without any charge being brought against him. The problem of possessing a criminal record may also be connected with a wrongful conviction.⁶² Information about an arrest says nothing definite about criminal activity. It is therefore hardly ethical to take such information into account when sketching a person's profile with the aid of AI tools.

Given the devastating impact of information about a person’s supposed criminal activity, in any doubtful situation we propose that such information should not be included in the portrayal of a candidate for employment.

One should also note that the fact that the expungement could be enabled by the new mechanism does not necessarily mean that the chances for finding employment would be equal. Although a lack of information on previous contact with the criminal justice system would decidedly be worth looking for, hiring mechanisms for vetting candidates could apply other criteria that might put the job seeker in a worse position – especially if we are dealing with someone who has served a prison sentence. For example, in the standard HR vetting process, a prolonged lack of employment or unaccounted-for gaps in the curriculum (together with correspondingly fewer years of work experience) would probably worsen the chances of the candidate anyway – though not, of course, as badly as the fact of having had a conviction and having ‘done time’.

There is also the problem of assessing whether or not the expungement has actually been embedded into the search engines and the HR AI tools. In literature discussing whether it is possible to check whether algorithms comply with laws and policies, many different options are mentioned.⁶³ We believe that as far as our problem is concerned the following three ways could be the most relevant: conducting an assessment of the data sets (confirmatory assessment); human-in-the-loop (human oversight);⁶⁴ and, making algorithms open (transparency). The first is to assess samples to see whether the results of the search or the reports prepared by the HR AI tools have actually expunged the data that ought to be expunged by law. Another way is to include human beings in the process, giving them the task of looking for suspicious data. The last way is to make the algorithms open – which, however, does not necessarily mean that they will be more understandable.⁶⁵ There is a need for a proper assessment of the efficacy of such tools.⁶⁶

Another issue is that, as often as not, the check is based on limited and non-reliable data, as – for example – many people happen to have the same name, which means that at times it may be quite impossible to satisfactorily ascertain that a particular person is the ex-offender in question.⁶⁷

To summarise this section, we think that search engines and companies which devise tools for HR should be extremely cautious about information that is in any way connected with previous criminal activity. We believe that embedding expungement in those tools is ethically commendable and is congruent with the idea of expungement. Expungement by design could mitigate the negative impact of the availability of information on the internet, including that concerning former criminal activity. It would be congruent with the aims of this legal instrument.

5. Conclusions

Seen in the light of the search for employment, the awareness of a criminal record can have a deeply negative influence on an individual’s employability, leading – among other things – to his or her stigmatisation and marginalisation (as well as increasing the chances of a relapse into criminal activity). Owing to the characteristics of our increasingly digitalised world and the rapidly growing volume of information that is available online, the concomitant consequences of the gathering of such information for recruitment purposes are becoming ever more serious. This could lead to a further deterioration of the precarious situation of ex-offenders in those countries where such data is made publicly available, as well as possibly undermining privacy-based and anonymization-based systems elsewhere.

Most importantly, today’s data practices have the potential to completely nullify the legal institution of expungement. However, the possibility of embedding expungement in the actual design of AI-backed HR vetting systems holds out some hope for the future.

The use of AI in vetting and recruitment can either further worsen this trend or be used for the greater good of re-integration. As far as HR practices are concerned, when expungement is an individual’s right, other actors should respect the legal situation, as they too function in a reality shaped by law. In a concrete situation, actors using AI instruments for HR should respect the rules and act accordingly in a spirit of forgiveness. Such behaviour will not only be morally good, but also lawful. At present, the AI tools used in the context of employment could interfere with the legal promise – based on forgiveness – that old transgressions will not influence the remaining life of the ex-offender. Expungement can – at least to a certain extent – be incorporated into these AI tools in order to be consistent with the legal fiction that a person whose criminal data has been expunged can be treated as having had no criminal record. Given the inadequacy of State control of information, such a practice can at least in part restore the societal effects of the much-needed institution of expungement.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Kamil Mamak

Kamil Mamak is a philosopher and a lawyer. He is a postdoctoral researcher at the RADAR: Robophilosophy, AI ethics and Datafication research group at the University of Helsinki and an assistant professor at the Department of Criminal Law at the Jagiellonian University. He is also a Member of the Board of the Cracow Institute of Criminal Law. His newest book “Robotics, AI and Criminal Law: Crimes against Robots” will be published by Routledge in 2023.

Jadwiga Glanc

Jadwiga Glanc holds a Ph.D. in law from the University of Zurich. Her field of interest lies at the intersection of law, new technologies and regulatory issues. Her publications include “Equity crowdfunding as a regulatory question” (Zürich 2020) and other works on the changes and challenges that are being brought by the phenomenon of digitalization.

Notes

1 cf. Hin-Yan Liu and others, ‘Artificial Intelligence and Legal Disruption: A New Model for Analysis’ (2020) 12 Law, Innovation and Technology 205.

2 cf. Roger Brownsword, ‘The Shaping of Our On-Line Worlds: Getting the Regulatory Environment Right’ (2012) 20 International Journal of Law and Information Technology 249.

3 Howard S Becker, Outsiders, Studies in the Sociology of Deviance. (reprint edition, The Free Press 1966) 33.

4 Ibid 3.

5 Ibid 33.

6 Gabriel Chin, ‘The New Civil Death: Rethinking Punishment in the Era of Mass Conviction’ (2012) 160 University of Pennsylvania Law Review 1789, 1791.

7 John Bronsteen, Christopher Buccafusco and Jonathan Masur, ‘Happiness and Punishment’ (2008) 76 University of Chicago Law Review 1037/1038.

8 Cf JJ Prescott and Sonja B Starr, ‘Expungement of Criminal Convictions: An Empirical Study’ (2020) 133 Harvard Law Review 2462 <https://harvardlawreview.org/2020/06/expungement-of-criminal-convictions-an-empirical-study/> accessed 12 December 2020.

9 Charles Griswold, Forgiveness: A Philosophical Exploration (Cambridge University Press 2007).

10 David Konstan, Before Forgiveness: The Origins of a Moral Idea (Cambridge University Press 2010).

11 Robert D Enright and Bruce A Kittle, ‘Forgiveness in Psychology and Law: The Meeting of Moral Development and Restorative Justice’ (1999) 27 Fordham Urban Law Journal 1621, 1630.

12 Stephanos Bibas, ‘Forgiveness in Criminal Procedure’ [2007] Faculty Scholarship at Penn Law 329 <https://scholarship.law.upenn.edu/faculty_scholarship/920>.

13 Meg Leta Ambrose, Nicole Friess and Jill Van Matre, ‘Seeking Digital Redemption: The Future of Forgiveness in the Internet Age’ (2012) 29 Santa Clara Computer & High Technology Law Journal 99, 114. It must be noted, however, that forgiveness can bring certain risks: for example, if unlimited forgiveness is proposed, it may be seen as being a licence to harm yet again; see Julie Juola Exline and others, ‘Forgiveness and Justice: A Research Agenda for Social and Personality Psychology’ (2003) 7 Personality and Social Psychology Review 337.

14 Griswold (n 9) xiii.

15 Linda Meyer, ‘Forgiveness and Public Trust’ (2000) 27 Fordham Urban Law Journal 1515, 1515. However, we should exercise caution when shifting from interpersonal forgiveness to State forgiveness, as individuals function differently from groups; see Martha Minow, ‘Forgiveness, Law, and Justice’ (2015) 103 California Law Review 1615.

16 see e.g. Emmanuel Melissaris, ‘Posthumous “Punishment”: What May Be Done About Criminal Wrongs After the Wrongdoer’s Death?’ (2017) 11 Criminal Law and Philosophy 313.

17 Russell Powell, ‘Forgiveness in Islamic Ethics and Jurisprudence’ (2011) 4 Berkeley Journal of Middle Eastern and Islamic Law 17.

18 Ambrose, Friess and Van Matre (n 13) 129.

19 Ian Weinstein (Moderator), ‘Forgiveness in the Criminal Law’ (2000) 27 Fordham Urban Law Journal 1373.

20 Ambrose, Friess and Van Matre (n 13).

21 Steven Raphael and Rudolf Winter-Ebmer. ‘Identifying the Effect of Unemployment on Crime’ (2001) 44 The Journal of Law and Economics 259.

22 A more complicated situation arises when forgiveness is proposed even before justice is meted out Joanna North, ‘Wrongdoing and Forgiveness’ (1987) 62 Philosophy 499. This, however, is a different issue. Here we are considering the situation after the punishment has been carried out.

23 Zachary Hoskins, Beyond Punishment?: A Normative Account of the Collateral Legal Consequences of Conviction (Oxford University Press 2019).

24 Brian Murray, ‘Retributive Expungement’ (Social Science Research Network 2020) SSRN Scholarly Paper ID 3617875 8–9 <https://papers.ssrn.com/abstract=3617875> accessed 23 February 2021.

25 Minow (n 15) 1618.

26 James B Jacobs, The Eternal Criminal Record (Harvard University Press 2015) 2.

27 Clay Calvert and Jerry Bruno, ‘When Cleansing Criminal History Clashes with the First Amendment and Online Journalism: Are Expungement Statutes Irrelevant in the Digital Age’ (2010) 19 CommLaw Conspectus: Journal of Communications Law and Policy 123, 130. We must note, however, that there is a negative side to this legal device: scholars have drawn particular attention to the tension that exists between the effects of expunged criminal data on the one hand and freedom of expression on the other; see Calvert and Bruno (ibid). Ambrose et al (n 13) write about censorship in relation to ‘the right to be forgotten’, which shares many similarities with the effects of expungement.

28 Cf the study by Jakari N. Griffith, Candalyn B. Rade, and Kemi Salawu Anazodo, ‘Criminal History and Employment: An Interdisciplinary Literature Synthesis’ (2009) 38 Equality, Diversity and Inclusion: An International Journal 505.

29 Jacobs (n 26) 2.

30 Naomi F Sugie, Noah D Zatz and Dallas Augustine, ‘Employer Aversion to Criminal Records: An Experimental Study of Mechanisms’ (2020) 58 Criminology 5, 27.

31 Andrew Elmore, ‘Civil Disabilities in An Era of Diminishing Privacy: A Disability Approach for the Use of Criminal Records in Hiring’ (2014) 64 DePaul Law Review 991; Jacobs (n 26); Chin (n 6).

32 Sugie, Zatz and Augustine (n 30) 6.

33 JJ Prescott and Sonja B Starr (n 8).

34 Jacobs (n 26) 4. Another important point is that, because ‘minority populations are disproportionately represented among those with criminal records and innocent minorities are disproportionately subject to arrest’ (see Paul-Emile Kimani, ‘Beyond Title VII: Rethinking Race, Ex-Offender Status, and Employment Discrimination in the Information Age’ (2014) 100 Va. L. Rev. 893, 900), this can lead to decreased employability for minorities and their being pushed towards the edges of society.

35 Cf Alessandro Corda and Sarah E Lageson, ‘Disordered Punishment: Workaround Technologies of Criminal Records Disclosure and the Rise of a New Penal Entrepreneurialism’ (2020) 60 The British Journal of Criminology 245.

36 Eldar Haber, ‘Digital Expungement’ (Social Science Research Network 2018) SSRN Scholarly Paper ID 3214335 338 <https://papers.ssrn.com/abstract=3214335> accessed 21 December 2020.

37 Patricia Sanchez Abril, Avner Levin and Alissa Del Riego, ‘Blurred Boundaries: Social Media Privacy and the Twenty-First-Century Employee’ (2012) 49 American Business Law Journal 63, 64.

38 Jenna Jacobson and Anatoliy Gruzd, ‘Cybervetting Job Applicants on Social Media: The New Normal?’ (2020) 22 Ethics and Information Technology 175, 175.

39 Cf Brenda L. Berkelaar, ‘Cybervetting, Online Information, and Personnel Selection: New Transparency Expectations and the Emergence of a Digital Social Contract’ (2014) 28 Management Communication Quarterly 479.

40 CareerBuilder, ‘Number of Employers Using Social Media to Screen Candidates at All-Time High, Finds Latest CareerBuilder Study’ (Press Room | Career Builder, 15 June 2017) <http://press.careerbuilder.com/2017-06-15-Number-of-Employers-Using-Social-Media-to-Screen-Candidates-at-All-Time-High-Finds-Latest-CareerBuilder-Study> accessed 13 February 2021.

41 Ibid.

42 Ibid.

43 Marti Rovira, ‘The next Pandora’s Box of Criminal Background Checks’ [2020] European Journal of Criminology 1477370820977880.

44 In the United States’ context, see James B. Jacobs (n 26) 155.

45 Prescott and Starr (n 8) 2467.

46 Haber (n 36) 351.

47 Roger Brownsword, Law, Technology and Society: Re-Imagining the Regulatory Environment (Routledge 2019).

48 Tim Verhoeven (ed), Digitalisierung im Recruiting: Wie sich Recruiting durch künstliche Intelligenz, Algorithmen und Bots verändert (Gabler Verlag 2020) 114 <https://www.springer.com/de/book/9783658258849> accessed 24 February 2021.

49 Antonio Aloisi and Elena Gramano, ‘Artificial Intelligence Is Watching You at Work. Digital Surveillance, Employee Monitoring and Regulatory Issues in the EU Context’ (2019) 41 Special Issue of Comparative Labor Law & Policy Journal, ‘Automation, Artificial Intelligence and Labour Protection’, edited by Valerio De Stefano 95, 102.

50 Jacobson and Gruzd (n 50) 175.

51 Verhoeven (n 48); Ben Eubanks, Artificial Intelligence for HR: Use AI to Support and Develop a Successful Workforce (Kogan Page Ltd 2018).

52 As far as the question of differentiating people is concerned, one of the most important issues which has been noted in connection with the use of AI is the possible occurrence of bias Eubanks (n 51) 158.. The AI-enhanced employment mechanism may conceal an underlying bias, such as (for example) the built-in bias against the black community that was noted in the assessment of the risk of future ‘criminality’ Julia Angwin and others, ‘Machine Bias’ (ProPublica, 23 May 2016) <https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing?token=Kc9axO_GEcQwja43fVMKIbGDdlH9IO4Z> accessed 22 February 2021.. What is more, in a bad-case scenario, AI can become a non-accountable judge of ‘criminality’. The posts, language or pictures that people leave on the internet could unwittingly serve as a basis for making distinctions. Here, however, we are on very slippery ground, as – for example – recent research has shown that language used by Afro-Americans was wrongly perceived by AI as being more aggressive, see Maarten Sap and others, ‘The Risk of Racial Bias in Hate Speech Detection’, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (Association for Computational Linguistics 2019) <https://www.aclweb.org/anthology/P19-1163> accessed 5 March 2021. Moreover, there is also the possibility of ‘extended’ criminality profiling: for example, it would be quite feasible to screen the candidate’s family and those people with whom he or she interacts on social networks (this being based on the assumption that the social environment exerts a strong influence on us). Data acquired in this way could also lead to the de-anonymization of convictions or the establishment of links to anonymized criminal behaviour in those systems which have such anonymization. In a similar way, AI-backed information gathering and candidate scanning could not only circumvent, but actually nullify the utility of the expungement procedure.

53 Ifeoma Ajunwa, ‘An Auditing Imperative for Automated Hiring’ [2021] Harvard Journal of Law and Technology <https://par.nsf.gov/biblio/10267580-auditing-imperative-automated-hiring> accessed 9 December 2022.

54 The Associated Press, ‘U.S. Warns of Discrimination in Using Artificial Intelligence to Screen Job Candidates’ NPR (12 May 2022) <https://www.npr.org/2022/05/12/1098601458/artificial-intelligence-job-discrimination-disabilities> accessed 9 December 2022; Dinah Wisenberg Brin, ‘EU: Proposed Artificial Intelligence Law Could Affect Employers Globally’ (SHRM, 13 September 2022) <https://www.shrm.org/resourcesandtools/hr-topics/global-hr/pages/eu-proposed-artificial-intelligence-law.aspx> accessed 9 December 2022.

55 Ambrose, Friess and Van Matre (n 13).

56 Mireille Hildebrandt, ‘Legal Protection by Design: Objections and Refutation’ (2011) 5 Legisprudence 223, and ‘Saved by Design? The Case of Legal Protection by Design’ (2017) 11 NanoEthics 307; and Mireille Hildebrandt and Laura Tielemans, ‘Data Protection by Design and Technology Neutral Law’ (2013) 29 Computer Law & Security Review 509.

57 Mireille Hildebrandt, ‘Saved by Design? The Case of Legal Protection by Design’ (n 56), 307–308.

58 cf. Juliusz Makarewicz, Kodeks Karny z Komentarzem (Wyd 3, Wydawnictwo Zakładu Narodowego im Ossolińskich 1932); Leon Peiper, Komentarz Do Kodeksu Karnego: Prawa o Wykroczeniach Przepisów Wprowadzających Obie Te Ustawy Oraz Do Rozporządzenia Prezydenta R. P. o Niektórych Przestępstwach Przeciw Bezpieczeństwu Państwa z Dnia 24 Października 1934 Roku (Dz. U. Nr. 94, Poz. 851): Z Uwzględnieniem Ustawy Karnej Skarbowej, Ordynacji Podatkowej, Kodeksu Karnego Wojskowego, Ustaw Dodatkowych Orzecznictwa Sądu Najwyższego (Wyd 2, Leon Frommer 1936); Marian Cieślak, Polskie Prawo Karne: Zarys Systemowego Ujęcia (Wyd 2, popr i uzup, Wydawnictwo Naukowe PWN 1994); Radosław Krajewski, ‘Zatarcie Skazania w Prawie Karnym’ [2007] Państwo i Prawo; Włodzimierz Wróbel, ‘O Dwóch Aspektach Konstytucyjnej Zasady Domniemania Niewinności’ in Krzysztof Krajewski (ed), Nauki penalne wobec problemów współczesnej przestępczości. Księga jubileuszowa z okazji 70. rocznicy urodzin Profesora A. Gaberle (2007).

59 Kamil Mamak, ‘O Niecelowości Anonimizacji Oskarżonych w Sprawach Medialnych’ [2013] CZPKiNP; Kamil Mamak, ‘Pseudoanonimizacja Jako Zniesławienie’ in Włodzimierz Wróbel (ed), Prawo karne gospodarcze. Księga jubileuszowa Profesora Zbigniewa Ćwiąkalskiego (Krakowski Instytut Prawa Karnego Fundacja 2023).

60 Kamil Mamak, Rewolucja Cyfrowa a Prawo Karne (Krakowski Instytut Prawa Karnego Fundacja 2019).

61 AI could also be used to promote increased employment among ex-criminals. While hiring managers may be more cautious and sceptical as regards criminal records (especially if they have a higher degree of discretion and are concerned about their personal liability), this may not be so in the case of AI vetting, as the burden of liability (or personal responsibility) is to a great extent taken from the shoulders of the hiring manager and, moreover, the policies can be directly encoded into the process. In other cases, the solution could be to overlook certain minor offences or even allow the employer himself to specify the degree of ‘leniency’. The most important issue, however, is the possibility of somehow including expungement in the employment process, thus making it feasible.

62 Amy Sholsberg, Evan Mandery and Valerie West, ‘The Expungement Myth’ (2011) 75 Albany Law Review 1229.

63 cf. Rowena Rodrigues, ‘Legal and Human Rights Issues of AI: Gaps, Challenges and Vulnerabilities’ (2020) 4 Journal of Responsible Technology 100005; Maja Brkan, Monica Claes and Clara Rauchegger, ‘European Fundamental Rights and Digitalization’ (2020) 27 Maastricht Journal of European and Comparative Law 697.

64 See, also, Bettina Berendt and Sören Preibusch ‘Toward Accountable Discrimination-Aware Data Mining: The Importance of Keeping the Human in the Loop – and Under the Looking Glass’ (2017) 5 Big Data 135.

65 Rodrigues (n 63).

66 James Guszcza and others, ‘Why We Need to Audit Algorithms’ [2018] Harvard Business Review <https://hbr.org/2018/11/why-we-need-to-audit-algorithms> accessed 25 November 2022.

67 This might well be one of the biggest challenges faced by AI criminality checking systems (and, for that matter, any system that implemented expungement). One solution could be an authorized cross-check with the official register: the information from the register would then automatically override any information found elsewhere.