Analogue harms of long processing periods

The social impact of long processing periods varies depending on the type of applicant. While in the case of a first working permit, prolonged waiting periods may lead to applicants forgoing the original job offer and other job opportunities (Yle News, 2018b; 2019c), in the case of asylum applications the consequences could be more severe. In 2022, MoI estimated that about 3000 asylum seekers from the year 2016 or earlier were still in Finland, waiting for a decision on international protection (MoI, 2022: 15). During the asylum process, state-funded reception centres offer accommodation and essential welfare and healthcare services (Reception Act section 26.1) to applicants waiting for their cases to be handled. Applicants are free to find accommodation outside of the reception centres (Reception Act section 18). However, given their limited right to work (Aliens Act 301/2004 section 79.2) and lack of state support for private accommodation, their options are scarce. Only a residence permit secures possibilities for working or studying, guarantees full access to healthcare services and enables full societal integration. In the case of a failed asylum application, the applicant can no longer use the services of the reception centre after a certain time (Reception Act section 14a), and healthcare becomes limited to urgent situations (Health Care Act, 1326/2010 section 50) and essential social welfare services (Social Welfare Act 710/1982 section 12). In addition, the applicants often cannot continue their studies or work previously carried out (Aliens Act section 73).

Harmful analogue solutions

To deal with the management crisis in Migri, multiple analogue solutions were proposed. At first, Migri was allocated additional resources, and 500 new employees were hired to process applications (Yle News, 2017). However, the authority’s operations raised doubts about the quality and, by extension, the fairness of asylum processing. Reports of investigative journalists at the beginning of the crisis indicated that caseworkers employed as a result of the emergency appointments in 2016 were given templates of negative asylum decisions (Yle News, 2017). Indeed, a comparison between the asylum decisions before and after the so-called crisis (Vanto et al, 2022) revealed that asylum caseworkers’ assessment of similar facts changed considerably. In particular, trust in the credibility of applicants’ claims decreased. Notably, there were no legal changes that could explain this interpretative shift. The considerable scope of discretion given to immigration officers allowed them to draw on legal frameworks in a flexible and instrumental manner and, under the political pressure to curb the increased number of arrivals, generated a shift in asylum decision making on a grand scale (Vanto et al, 2022).

Efforts to mitigate the crisis continued with more restrictive regulations introduced into the asylum process (Finnish Government, 2015) that limited the right to free legal aid, shortened time frames for appeal, increased the number of deportations and intensified the scrutiny of immigrants’ income (Yle News, 2020a). The multiple legal changes implemented in a brief period constituted a part of the ‘race to the bottom’, leading to undesirable consequences in the form of reduced attention to the basic human rights of asylum seekers (Pirjatanniemi et al, 2021: 229). The unprecedented situation contributed to the erroneous processing practices of Migri, shifting the burden to the administrative courts (Yle News, 2018a). The need for appeal procedures and repeated asylum processes increased, especially in the first years following the crisis, when the courts dealt with a peak in asylum cases: 8854 cases in 2017 and 6821 cases in 2018. About a third of these cases were returned to Migri for new processing (Owal Group, 2019: 13).

Concurrently, however, the number of first-time asylum seekers decisively lowered and, at the end of 2018, Migri reduced its personnel resources, leaving fewer than 40 employees to process asylum applications (Yle News, 2020b). Due to the concomitant backlog of about 20,000 cases and additional tasks connected to the rising numbers of residence permit applications, the processing times have further increased (Yle News, 2019a), and the management crisis has continued. Nevertheless, Migri’s budget allocation did not increase. Citing disagreement with the Ministry of Finance (MoF) ‘over what constitutes adequate resourcing for the agency’, Migri planned to lay off a further ten percent of its staff at the end of 2019 (Yle News, 2019b). Meanwhile, investment in information and communications technology services to develop Migri’s capacities for ADM was considered a solution to the problem.

Promise of algorithmic solutions

The right-wing government, elected just before the Migri crisis in 2015, placed digitalisation at the forefront of their political programme. The programme promoted a culture of experimentation as contributing to societal issues such as health and well-being, employment, competitiveness and growth, education and skills. Digitalisation and the dismantling of regulatory boundaries were considered essential to cultivate experimentation (Government Programme, 2015: 27). To this end, the MoE began to develop the Finnish AI strategy under the auspices of the project Era of AI (MoE, 2017). The concluding report of this project (MoE, 2019), often referred to as the Finnish Strategy on AI, promoted self-regulation among private and public actors, and encouraged an ‘AI sandbox’, a limited area for experimentation on AI and the development of new regulatory solutions (MoE, 2019: 54). Despite the neoliberal orientation of the right-wing politics, the initial AI policies often referred to the notion of ‘human-centric AI’, highlighting the trustworthiness of AI development and its ethical operation in the private as well as the public sector.

In line with these strategic goals, digitalisation and automation were considered a solution to the management crisis in Migri. First, basic customer service was transferred to chatbots, launched in May 2018 to reduce human service interactions. Then, to ensure compliance with the constitutional principle of legality, that is, that all use of administrative power requires a legal basis established by a parliamentary act, the MoI prepared a bill (HE 224/2018) concerning the use of personal data processing and automation by Migri. The idea was to fully automate a portion of the decision-making processes in straightforward cases in order to allocate the scarce human resources to cases requiring discretion and case-by-case assessment (HE 224/2018). It is important to note that the Finnish Strategy on AI refers to AI as systems ‘that are able to learn and to make decisions in almost the same manner as humans’ (MoE, 2019: 16). However, the discussion on automation relating to Migri concerned systems with pre-coded rules derived from parliamentary acts, simple calculations and unambiguous reasoning, hence with no AI component (MoJ, 2020: 63).

Regardless of the limited scope of the proposed automation, the bill did not pass constitutional review, which in the Finnish legal system occurs ex ante, during the legislative process. According to the CLC, the proposal failed to adequately address the principles of good governance or comply with GDPR rules. Eventually, the work on the first Migri bill was discontinued due to a change of government. The new left-leaning government, similarly to its predecessor, championed the possibilities of AI and encouraged experimentation and broad utilisation of AI (Government Programme, 2019). Thus, the work on legislation enabling ADM by Migri continued (HE 18/2019: 30). The goal was again to modernise and streamline Migri’s processing to mitigate the increased number of applications, while guaranteeing the constitutional requirements of good governance, in particular the right to be tried without undue delay (HE 18/2019: 29). Nevertheless, even though the updated bill delineated the scope of automation more clearly, to ensure compatibility with constitutional principles, the CLC deemed the suggested regulation too vague and therefore insufficient to fully comply with the constitutional principles of transparency, justifiability and accountability of administrative decisions. Instead, the CLC commissioned an assessment to evaluate the need to launch the legislative project (CLC, PeVL 7/2019).

The statement of the CLC effectively meant a total ban on automation in all branches of public administration until appropriate legislation could be implemented. Given that many public agencies considered automation practices mandatory for their operation (Yle News, 2019d), the preparatory process of the General Law on Automated Decision Making in Public Administration (the ADM Law) started at the end of 2019. After extensive preparatory work, the first draft for a bill on the ADM Law was published in February 2022 (MoJ, 2022). The draft leans towards relatively strict regulation and limited utilisation of ADM in public administration, allowing rule-based ADM systems in cases that do not require discretion, or when the required discretion is executed on a case-by-case basis by caseworkers prior to the decision (MoJ, 2022: 47).

The legislative process has progressed slowly. Meanwhile, Migri’s problems with high expenditures and a scarce budget have continued. Hence, at the end of 2019, in its fourth supplementary budget proposal, the Government allocated 4.7 million euros to the agency to guarantee sufficient resources for its tasks and personnel costs. The decision enabled Migri to prevent the planned reduction of its personnel. An additional 1.1 million euros were set aside for maintaining the agency’s compulsory IT systems (HE 67/2019: 37). However, the additional funds were designated for securing the continuity of the agency’s operations (HE 67/2019: 37) and thus offered no new means to solve the ongoing management crisis. Consequently, given the persistent backlog of cases, numerous applicants, especially asylum seekers, remain in a bureaucratic limbo, a state of ‘in-betweenness’ that pushes them towards surviving rather than living (Hartonen et al, 2021: 41). Next, we structure the harms connected to the Migri case within a zemiological framework, forming a basis from which to analyse the Finnish harm-reduction regime in the context of algorithmic transformation.

Analogue harms

Social harm studies (Hillyard et al, 2004), that later developed into zemiology, emerged within critical criminology, arguing that mainstream accounts of criminology fail to capture harmful conditions embedded in social structures. To achieve meaningful social change (Canning and Tombs, 2021), social harm scholars shift the focus from individuals to collective entities and socioeconomic structures, exposing how individuals and groups are harmed in the course of different social ordering practices.

Significant social harms are attributed to restrictive immigration policies (Canning and Tombs, 2021: 83); in particular to the ‘illegalisation of movement’ (Soliman, 2019). Border controls force people, especially those deemed ‘undeserving’ (Ustek-Spilda and Alastalo, 2020), to take riskier, often illegal journeys causing harms ranging from financial dependency on smugglers to an increased likelihood of physical harm or death. Those who manage to enter the country experience what Tervonen et al (2018) refer to as ‘internal everyday bordering practices’, that is, the confusing if not harmful bureaucracy of ‘endoborders’: ‘micro and macro level controls which restrict everyday interactions and civil liberties’ (Canning, 2019: 43). While some of these ‘borders’ are entrenched in legal frameworks of immigration management, others stem from defective processing procedures during which migrants are exposed to anguish and uncertainty.

Western states’ immigration policies are frequently explored in the context of neoliberalism (Joppke, 2021) and securitisation (Palander and Pellander, 2019). These trends have shaped bifurcated models of internal and external bordering practices directed at different categories of migrants, combining competitive recruitment of high-skilled migrants with restrictive attitudes towards low-skilled, family, asylum, or irregular migration (Joppke, 2021). However, the role of immigration is also dichotomous in the rhetoric of welfare nationalism (Heino and Jauhiainen, 2020). Accordingly, since the early 2000s, labour migration has been presented in Finnish governmental policies as a solution to demographic problems and concomitant labour shortage (Ollus and Alvesalo-Kuusi, 2012), and thus as beneficial, if not necessary, for the Finnish economy. Simultaneously, the restrictive governmental policies following the crisis in 2015 aimed at ‘making Finland less appealing to potential asylum seekers’ (Palander and Pellander, 2019), consequently limiting the number of arrivals through the external borders.

The restrictive regulation of the asylum process (Pirjantanniemi et al, 2021) and legal changes, such as removing humanitarian protection status from the Aliens Act and tightening the criteria for family reunification (Heino and Jauhiainen, 2020), created a hostile environment for those deemed inherently unwanted from the perspective of the Finnish welfare state. While these were undoubtedly harmful to people seeking refuge in Finland, in this paper, we focus on the role of defective processing procedures in Migri, in particular the role of long waiting periods and the increased number of erroneous decisions. The long waiting periods of immigration processes restrict applicants’ ability to fully participate in social relations and thus create the aforementioned ‘endoborders’. Beyond the regular processing periods, erroneous decisions lead to complicated appeal procedures that may span years. Uncertainty over the future and the constant sense of insecurity form the basis of what Canning (2019) calls temporal harms. These harms are connected to the prolonged lack of permanence that amplifies feelings of worry and anxiety (Hartonen et al, 2021: 42).

Temporal harms may manifest as mental health problems and psychological drama as well as cultural harms often framed as harms of misrecognition (Soliman, 2019; Pemberton, 2015: 24). In the Finnish context, research on the subjective well-being of asylum seekers conducted in one of the reception centres found that over half of the asylum seekers in the centre experienced suffering, a third were struggling, and only about ten percent considered themselves to be thriving (Hartonen et al, 2021: 41). The insecurity of the immigration process impairs applicants’ ability to formulate choices and act on these effectively. This leads to systematic dismissal of their wants, needs and concerns, amounting to the disablement of their attempt at self-actualisation and causing autonomy and relational harms (Pemberton, 2015: 29). In the case of Migri, partial automation was considered a plausible solution to the long processing periods (Migri, 2021). However, the feasibility of algorithmic decision making in public administration has been contested.

Algorithmic harms

Given the ubiquity of algorithmic transformation, attempts to use algorithmic systems to coordinate human activity have been conceptualised as a distinctive form of social ordering (Yeung, 2018). Hence, we argue that social harm scholars should pay attention to socio-techno-economic structures, which point to the technological dimension of harm (Malik et al, 2022). In official policies, the social change caused by algorithmic transformation has been couched in positive terms as driving social progress, prompting economic growth, and creating efficiency in public administration. This framing links the increased use of algorithmic technologies and datafication used by the welfare state (Dencik and Kaun, 2020) to a broader paradigm shift that has, since the 1970s, led to a transformation of welfare services. At first, the proliferation of new public management (NPM) approaches, which emphasise competition and incentivisation (Dunleavy et al, 2006, resulted in a fragmentation and decentralisation of decision-making processes (Temmes, 1998). Later, it was hoped that the turn to digital governance – a complex of changes centred around IT and information handling that preceded algorithmic transformation (referred to as post-NPM) – would reverse the ‘gains’ of the NPM era (Dunleavy et al, 2006). Optimistically, the post-NPM approach promised reintegration of governmental tasks and the creation of larger needs-oriented administrative blocks to simplify access to public services (Dunleavy et al, 2006: 478, 480). This framing aligns with the deeply-rooted Finnish optimism towards new technologies that have enabled the country’s economic progress in the past (Schienstock, 2007: 99).

In the pessimistic narratives, the gains from digitalisation and automation are irrelevant, as automation creates new power relationships leading to severe harms. For instance, in public decision making, automated systems comprehend individual cases via reductive models and thus are likely to exacerbate socially embedded bias and discrimination (Eubanks, 2017). Furthermore, the underlying problems in data quality, models of data collection or analysis may lead to false positives or false negatives (Whittaker et al, 2018; Chiusi et al, 2020) and, while accelerating decision-making processes, may also restrict avenues for redress (Citron and Pasquale, 2014), violating the fundamental rights of affected individuals (Wachter et al, 2021).

From this vantage point, ADM and the growing datafication of public administration are framed as the natural outgrowth of neoliberal managerialism, which prioritises deregulation, cost efficiency and the individualistic logic of competition (Waldman, 2019). Hence, technologies are inherently value-laden, serving particular political ends. In line with this, for example, Ustek-Spilda and Alastalo (2020), when analysing the Norwegian and Finnish practices of population registration, show that the software utilised in modern welfare services is built with the logic of capitalism, prioritising lower costs and increased profits over universal access to welfare or decommodification as a human right (Ustek-Spilda and Alastalo, 2020). In zemiological framing, the use of ADM in the public domain, especially without adequate human oversight, may exacerbate the damage already caused by privatisation, deregulation and other neoliberal policies frequently explored in social harm studies (Malik et al, 2022).

Similarly, Mann (2020) shows how ‘technology facilitates and enables the punitive, unfair, and unjust policies of the neoliberal state to be executed at great(er) scale and speed’. However, Mann argues that the centrality of technology should not overshadow the non-technological dimension. As ‘algorithmic fairness’ can never be achieved in a fundamentally unfair and unequal society (Mann, 2020: 7), technology should be decentred in discourses on social injustice. The crucial point she rejects is the capacity of technology to contribute to positive social change. In particular, she rejects the optimistic idea that harms can be mitigated by a better design of technology: without interventions into the underlying social organisation, technological innovations are deemed to perpetuate harmful conditions.

However, the deterministic framing of technology as working merely for the powerful presents an insufficient conceptualisation of human-technology relations (Wood, 2020). An abundance of progressive proposals has been made employing ex ante perspectives that centre on greater democratic oversight and deliberation about the deployment of algorithmic systems (for example, Pasquale, 2015). These approaches highlight the importance of the cross-sectional struggle over the design and deployment of algorithmic systems. In the same vein, McCarthy (2013) recognises the development of technology resulting from a political struggle over the gestalt of certain technologies to solve certain societal problems. This politically-contested process is ‘embedded within structured relations of social power’ and ‘cannot be fully closed due to the necessary ambivalence of technological objects’ (McCarthy, 2013: 477). Even if technology is a creation of the powerful, there are forces that champion alternative solutions and resist potentially oppressive technologies. These forces are sometimes strong enough to enable a more just reality to be built. However, the forms, extent and results of these struggles vary across societies. In what follows, we describe how the current approach to automation in public administration corresponds with features of Finnish society, and whether and how algorithmic and analogue harms are accounted for in the policy and legal documents.

Between vision and reality of Finnish AI policies

As in other Nordic countries, the Finnish legal system and administrative culture are marked by simplicity, transparency, equality, and avoidance of extremes (Letto-Vanamo and Tamm, 2019). This translates into a shared sense of responsibility, a relatively broad awareness of rights, and trust in public administration (Husa et al, 2007). In addition, the civil society sector is large and active (Helminen, 2019). Building on the preexisting culture of trust and citizens’ participation, the Finnish Strategy on AI (MoE, 2019: 103, 109, 111–112) underscores both the need to promote the ethical use of AI and to foster citizens’ trust and capacity to comprehend the transformation.

However, the practice differs from the ideal. The Finnish approach to automation in public administration has also been marked by tension between strong commitment to both competitiveness and good governance. The governmental work on AI, from the Government Programme of 2015 to the newest action plans under the project AI 4.0 (MoE, 2021a and 2021b), adheres surprisingly strongly to neoliberal values. Even though the change of government in 2019 drove political reality to the left with the most left-wing government elected in Finland in decades, neoliberal values became even more prevalent in the new AI policies. While the first project, Era of AI (MoE, 2017 and 2019), focused on society more broadly, the following project – AI 4.0, developed under Marin’s Government – sharpened the focus on enterprises, competitiveness, and economic growth. In the project AI 4.0, AI and other advanced digital technologies are framed as tools to accelerate systemic change of the business community (MoE, 2021a: 73). In the second report of the project AI 4.0 (MoE, 2021b), citizens and civil society are not mentioned at all. Instead, following neoliberal ethos, emphasis is placed on economic values and efficiency. As the approach considers well-being from the perspective of employment and economical gains, it diminishes the complex relations between different areas of life that contribute to the well-being of society. Thus, it fails to evaluate the crosscutting character of AI transformation and the experiences of this transformation of diverse social groups.

The initial Finnish strategies highlighted the need for active, multidisciplinary discussion between diverse societal actors on how AI can be applied in a trust-generating manner (MoE, 2019: 102, 106). Despite this ambition to involve civil society in the development of algorithmic solutions, the national Strategy on AI (MoE, 2019), as well as the following action plans (MoE, 2021a and 2021b), have been curated under the MoE. This places the emphasis strongly on the benefit to enterprises and corporations, economic growth and cutting the red tape for innovation. Although during the project Era of AI, citizens had a chance to comment and attend workshops, the project was carried out primarily by enterprise representatives, ministry advocates and academic experts.

The largest digitalisation project in the country, AuroraAI (MoF, 2020), which aims to develop human-centric public services, is being headed by the MoF. The project enables citizens to take part in the development process through open, real-time communication channels; however, the responsible stakeholders are limited to municipalities, ministry advocates and enterprises. Given the ministry’s objective ‘to stabilise general government finances, safeguard sustainable economic growth and to ensure that public services and administration are effective and efficient’ (MoF, 2022), the danger ensues that human-centric, needs-based services will be subordinated to financial policies aiming to ensure efficiency and balancing state expenditure. In addition, the lack of civil society involvement in the legislative process of the ADM Law is noticeable. Even with active participation at the consultation stage, no civil society organisations were chosen to participate in the drafting of the bill. This choice reveals the priorities in the ongoing regulatory process, as the bill establishes the grounds of future AI governance in Finland (Lepinkäinen and Malik, 2022).

This overview shows that while the current approach to automation corresponds with the Finnish socioeconomic, political and legal system as regards initial policies, a shift towards more liberal values is notable. Similarly, the incremental turn from human-centric to business-oriented frameworks could mean readjusting the interpretation of the Finnish Constitution (731/1999) (Lepinkäinen and Malik, 2022), which currently balances between liberal and social ideals on the meaning of the welfare state (Länsineva, 2012). The former emphasises the principle of individual freedom and negative protection of rights, whereas the latter highlights the promotion of justice in society, solidarity, and concomitant affirmatory actions of the government to ensure equality and social security (Länsineva, 2012: 119). Between these two ideals, reflected in the broad catalogue of constitutional rights, the Finnish constitutional system leaves space for the negotiation of welfare policies and their implementation leaning towards either social or liberal values. This includes the possibility to shift the priorities of automation in public administration.

Disconnected approaches: harm framing in Finnish official documents

The case of automation in Migri reveals contradictions between political priorities in different societal areas, as it lies at the crossroads of discussions on immigration and AI transformation in the political and legal dimensions. To foster the alleviating qualities of algorithmic technologies, these overlapping issues should not be considered in isolation. Hence, we explore how the risks and harms connected to these societal areas are accounted for in Finnish immigration policies, AI policies, the constitutional framework for automation of public administration and relevant regulatory suggestions. Our short analysis shows that the intertwining of analogue and algorithmic harms and solutions is considered inadequately in these documents, thus creating a fragmented political-legal reality.

In immigration policies, the Action Plan (MoI, 2021) as well as the Report (MoI, 2022) consider harms to people without a right of stay in Finland in a broad sense. These considerations form a rather significant part of the documents, starting from the risks following from strict border controls and attempts to evade them (MoI, 2021: 15). The Report (MoI, 2022: 15) describes how ‘the vulnerability of persons without a right of residence to exploitation and even human trafficking came to the fore’. The framing of the risks that newcomers face after arriving in Finland is often connected to different types of criminal abuse of individuals. However, the documents also acknowledge feelings of insecurity and fear (MoI, 2021: 28; MoI, 2022: 36). Different harms mentioned could lead, especially if prolonged, to mental health issues and a lessening of the quality of life altogether (MoI, 2022: 14). This is associated with risks for society, as the problems these individuals face may lead to difficulties in integrating into society (MoI, 2022: 36).

In the AI strategies, explicit mentions of risks or harms are rare. Moreover, while immigration policies consider harms and risks from the perspective of individuals who are conceptualised as active subjects, in AI policies, harms are perceived from the perspective of institutions and business life and framed as problems of efficiency due to the high costs of personnel or revenue risks (MoE, 2021a: 65). Individuals are conceptualised as customers, objects benefitting from the utilisation of modern technologies.

The Finnish Strategy on AI (MoE, 2019) contemplates the risks of an AI system strictly from a technological perspective. It examines the security of chosen technologies and the risks they pose to society, but harms for individuals are not considered. In addition, the distinction made in the report between risks ignores the possibilities of what Wood (2020) refers to as the technicity level of harms. In this account, technology has agency of its own in facilitating the unexpected and unintended and can cause an unintended impact on its users. In contrast, the Strategy on AI (MoE, 2019) only differentiates between ‘harmful AI’ (AI meant to cause harm), ‘harmful action on AI’ (weaponising the vulnerabilities of AI by a malicious actor) and ‘fallible AI’ (AI making inaccurate or biased decisions) (MoE, 2019: 109). Of these three, only the last group is considered to include possible unintended harms. However, even this last group of harms is simplified as being a consequence of insufficient training data. The strategy fails to consider the social implications of harmful AI, as it remains limited to technological risks and ways to prevent them.

The reports published in project AI 4.0 (MoE, 2021a and 2021b) take no stance at all on the risks or harms to individuals or society. Instead, taking risks is described as desirable in business life: ‘Growing and new innovative companies renew industry structures, challenge large companies to innovate and create role models for business that is flexible, risk-taking, using modern technologies and growing’ (MoE, 2021a: 64).

In legal documents framing the automation of public administration, risks and harms are often portrayed as questions of legality, constitutional rights, and freedoms. This is not surprising, as the documents adhere to legal terminology, and the perspective is constructed through requirements found in written law. The sections of the Migri bills (HE 224/2018 and HE 18/2019) considering automatisation adopt a fairly similar perspective to the AI policies, arguing that the use of ADM systems will make the institution more efficient and enhance its processes, which will also bring benefits to customers (HE 224/2018: 26; HE 18/2019: 29). The possible risks are rather implicitly equated to the risks of due process (HE 18/2019: 40–41), and restricting the use of automation is seen as potentially amplifying the risk-potential: ‘Automated decision-making improves efficiency, for example, by allowing human resources to be allocated to decision-making activities that require discretion. There is a possibility of error in decisions made by natural persons, while the possibility of human error is minimised in automatic decisions’ (HE 224/2018: 64; see also HE 224/2018: 30; HE 18/2019: 41).

The careful approach shown in the statement CLC PeVL, 7/2019, which sets a ban on the use of ADM systems in public administration, could be interpreted as being oriented towards the mitigation of the potential threat of algorithmic harms. Nevertheless, the harm reduction goal is merely implicit, as the CLC focuses on compliance with the constitutional standards of administrative processes (CLC PeVL, 7/2019: 11). The CLC statement includes no direct mention of risks and harms in the sections connected to automation. The risks and harms mentioned in the other sections of the statement predominantly relate to questions of privacy and data protection, which follows the main focus of the Government Bill 18/2019 on data processing.

Finally, the draft for the Government Bill on the ADM Law (MoJ, 2022) rarely uses the terms risk and harm without signposting the specific legal requirement on which the risk falls, or where the legal requirement for risk minimisation comes from. Importantly, the draft repeatedly relies on the notion of risks to individual rights and freedoms, and includes a reminder that the EU’s GDPR requires a risk-based approach to minimise risks in data processing.

Conversely, the immigration policies, despite the broad understanding of harms for individuals, show a rather weak understanding of automation, ADM and AI. Admittedly, automation is mentioned among the 52 actions to improve immigration management (MoI, 2021, Action 33). Yet, the goal of automation is framed from the perspective of harms to institutions, as it will be used to monitor and prevent abuses of the system. In addition, the use of automation is conceptualised as a part of the efforts to enhance the agency’s permit procedures. Automation is considered crucial to meet the objective set in the Government Programme of 2019, which is to process work-based residence in (on average) one month (MoI, 2021: 53). The Action Plan takes no stance on the question of how automation could or should be utilised in the agency’s decision making. The only mentioned risk related to automation is connected to possible irregular entries and attempts to evade entry provisions (MoI, 2021: 48), which follows the logic of the Finnish Strategy on AI (MoE, 2019) and connects risks with intentional malicious actions. Otherwise, the risks of automation or algorithmic harms are not accounted for, and the plan offers no guidance for the utilisation of algorithmic systems in the specific context of immigration matters. The following Report (MoI, 2022) does not mention AI, automatisation or digitalisation at all. It concentrates on the legal possibilities to solve the problem of irregular stay in Finland, but takes no stance on the question of algorithmic harms or solutions. This may follow from the fact that the Report was written and published at a time when the ADM Law was already under preparation. Still, the contextual understanding of the possibilities or the need for the utilisation of algorithmic technologies would be valuable.

Given the scale and unprecedented character of the 2015 immigration management crisis and the early stages of the concurrent digitalisation process, the disconnected framing of harms and solutions in these documents may not directly relate to the poor handling of immigration and asylum cases. However, more broadly, the management crisis in Migri illustrates the overlap of pressing societal issues: approaches to different forms of immigration, the future of the welfare state, and the algorithmic transformation of public tasks. This is emblematic of the embedded character of algorithmic technologies in preexisting societal structures, where algorithmic harms are only one layer in a broader stratigraphy of harms (Wood, 2020). This leads to the question of what the hierarchy of social harms is. More importantly, it also points to the cross-sectional character of AI transformation, which overlaps with racial, gender, ecological and class struggles, and should not be analysed in isolation. The disconnect between different political priorities pursued in separate policy sectors leaves a variety of harms completely unaccommodated. Bringing together legal and policy documents that regard these issues is necessary to gain a holistic view of the intertwining of possible harms and their prospective solutions. A deeper understanding of how the elements of immigration and automation policies affect each other enables solutions to become more operational. Still, the most recent immigration policy documents (MoI, 2021 and 2022), while focusing on different dimensions of analogue harms, take no stance on the question of algorithmic harms. The Action Plan considers automation in general, but the sectoral characteristics are not reflected, and even though many risks for individuals are recognised, no consideration of the impact of AI or ADM on the applicants can be found. If AI policies are handled without reference to the area to which they will ultimately be applied, the danger arises that disconnected approaches will potentially distort the perception of harms and prompt inadequate solutions.