Blog Post

Written by
Holistic AI
Structured content
June 6th, 2022
Regulating AI: Two Approaches to the Legal Ecosystem
Scrabble letters spelling the word 'LAW'

Horizontal and Vertical Regulation

The proliferation of AI is likely to reach almost every corner of the economy, spanning and altering numerous existing sectors. This post discusses how governments should undertake the regulatory division of labour: must AI be regulated by the central government and subject to uniform rules? Or must the authority to regulate applications of AI be limited to industry regulatory bodies imposing sector-specific rules? The former approach of enacting regulations for all applications of AI is what is referred to as ‘horizontal’ regulation. The latter approach of deferring to industry regulators is known as ‘vertical’ regulation. This section provides an overview of the comparative advantages of both approaches.

Regulation can be varying degrees of vertical or horizontal with respect to the breadth of its scope and the empowered authority issuing regulations. Horizontal regulation–typically a function of central governments–would issue uniform regulations that apply to AI across industries. Vertical regulation, by contrast, crafts regulation that is sector-specific, or delegates regulative authority to the supervision of an industry body.

There are economic and political advantages to both approaches. The horizontal model is uniform and stable, whereas the vertical model is flexible. The horizontal model enables the central government to guarantee that all sectors meet the standards that are in the public and national interest. It prevents enterprises from ‘shopping around’ between sectors, ensuring that all players conform to the same standards. Horizontal regulation also gives citizens a stable set of assurances and rights in their interactions with AI–i.e. they know what to expect and will be able to tell when they are being wronged.

By contrast, the vertical model of regulation allows industry bodies to make provision for differences across sectors and to optimise regulation to meet the specific needs of the industry. Vertical regulation has the benefit of specificity, which provides greater certainty to members of the industry where horizontal regulation will necessarily be vague in order to be broad in its application. A vertical approach occasions different coordination problems: there may be a significant amount of overlap between regulations resulting in a multiplicity of responsible agents, overlap and undue burden upon relevant parties; a single party may be required to report to multiple regulators regarding the same thing; the regulators may protect their autonomy and remit, thereby restarting knowledge transfer and and enforcement across sectors.

The EU AI Act proposes a horizontal regulatory framework for the European Union, in which AI across sectors are subject to the same risk-assessment criteria and legal requirements (discussed in the following section). Per the AI Act, all low-risk algorithms are subject to transparency requirements, high-risk algorithms are subject to more strenuous compliance measures, and then a further category of algorithms are prohibited altogether. Similarly, the proposed US Algorithmic Accountability Act adopts a horizontal approach by directing the Federal Trade Commission (FTC) to mandate impact assessments of AI systems across sectors (subject to the size and reach of the enterprise).

The UK’s approach to AI regulation remains more underdeveloped on this question: whilst the UK has intimated that it favours a sectoral approach, there are indications that possibly point in the direction of horizontal or hybrid regulation. With the move away from the EU, the UK has signalled, in the digital realm, that there is an appetite to propose an alternative approach to digital regulation. This has manifested itself through the activity of individual regulators publishing and signalling in the space of AI. Although there are signals of an alternative UK approach (one that more closely approximates a horizontal approach), we read the current stand as oriented towards a sectoral stance.

The UK Data Strategy, by contrast, is more ambiguous on this point: the Strategy indicates that it will devolve some regulatory powers to existing industry regulators, but suggests too that industry regulation may be underpinned by cross-sector principles that standardise the UK’s innovation landscape. This points to the possibility of a hybrid system in which there is vertical regulation supplemented by a minimum core of horizontal regulations that ensure a coherent set of rights, duties, responsibilities, and assurances.

The UK’s National AI Strategy can be seen in terms of a (potential) horizontal approach to regulation. We read this into the text because of the sections within the strategy that relate to AI governance. In fact one of the three pillars of the text is ‘Governing AI effectively’, in which two imperatives are particularly present: stimulating innovation and enterprise, and designing a standards and regulatory regime that reflects this innovation agenda. Within this framework, drawing on our prior work, there are a number of themes to note:

  • Removing some existing regulatory burdens where there is evidence they are creating unnecessary barriers to innovation.
  • Retaining the existing sector-based approach, ensuring that individual regulators are empowered to work flexibly within their own remits to ensure AI delivers the right outcomes.
  • Introducing additional cross-sector principles or rules, specific to AI, to supplement the role of individual regulators to enable more consistency across existing regimes.

Note that these themes read as an endorsement of the sectoral approach. In fact, we have read the ecosystem as signalling that the UK will take a sector-specific approach–as such the inclusion of the cross-sector regulatory policy is to be read as an important shift.

That said, our reading also suggests that the UK can position itself for horizontal regulation. To be clear, the language of ‘the UK positioning itself’ is not a claim that there is a coherent and singular intentional force driving the agenda. Indeed, naturally, there are competing interests and forces that motivate their respective agendas, seeking to realise aims that conform to what they would envision as appropriate and adequate AI governance (whether that be in the form of self-regulation, representation, regulation (horizontal and vertical), etc.). Below we will tease out a narrative that can be thought of as implicitly (if not explicitly) leaving open the possibility of a horizontal form of UK regulation.

Regarding removing existing regulatory burdens, the reading can quite readily be one of moving towards a more principled, national level, regarding regulation. Regarding retaining the sector-based approach, a reading can be made that appeals to the imperative that the regulators should be flexible i.e. the sector regulator burdens may be in setting standards and monitoring rather than in codification of a specific regulatory requirement - something that would readily align with a horizontal approach. As such, an interesting argument can be given that the sectorial/vertical approach can serve a horizontal agenda insofar as the associated regulators of the sectoral/vertical serve as effective policy, standards and monitoring institutions.

Similarly, the UK’s AI Assurance Roadmap presents a possible interpretation that suggests a horizontal approach to regulation. The text is clear insofar as the aim is to realise mechanisms of ensuring trustworthy AI to establish trust and set the conditions for AI innovation, industry and adoption (very much in the AI as an opportunity genre). In the section that directly addresses ‘The Role of Regulators’, instruction is given regarding the need for the regulators to provide guidance and expertise, reflecting the nuances of the respective sectors. In closing the section the text reads ‘as the use of AI systems becomes more widespread across sectors, clear regulatory scope will need to be established between regulators with similar and overlapping mandates. Moreover, regulators will need to decide in what contexts assurance be directly sought by regulators (regulatory inspection) and when assurance should be delegated to assurance providers, or where assumed conformity is appropriate’ - here the reading indicates a point of indeterminacy. Although there is clearly an implicit view that the landscape will evolve through sectoral regulation (insofar as the sectoral regulators and assurance bodies will be tasked in their respective domains), it is still an open question how this will translate to the specificity of regulatory regimes with respect to specific sectors (assurance and compliance are not one and the same thing).

Existing sectors are typically already subject to their own regulatory bodies and regulations that understand the needs of their industries and are well-placed to regulate the development of AI within their sectors. However, the effects of AI are of national (and international) import; many applications of AI cut across sectors; and sector-specific is likely to be piece-meal. Therefore, it is important for regulators to adopt a combination of regulatory strategies that balance the need for industry-specific regulation and national regulatory uniformity.

Critical Assessment

Consider the choices between Horizontal and Vertical regulation. Our concern is that these regulatory choices come to be regarded as technical decisions that should be aimed towards maximising efficiency in regulation. On the contrary, we argue that these choices are fundamentally political: they must be underpinned by a normative view about the role of state intervention in the (digital) economy.

In the choice between Horizontal and Vertical regulation, the Horizontal option favours government intervention, whereas the Vertical option favours non-intervention. Horizontal intervention gives the central government the power to reduce or eliminate risks across the board by issuing universal compliance measures for the development of AI. As we discussed above, this comes at the cost of innovation, because it imposes inflexible standards across industries. In this sense, the Horizontal option is more amenable to a Responsibility-oriented approach to AI governance, because it is adept at reducing risk, even if it frustrates innovation. By contrast, the Vertical option is more amenable to an Innovation-oriented approach, because it allows industry regulators to set more adaptable standards that suit the needs of the industry–albeit at the cost of uniform safety.

In devising regulatory mechanisms, legislators must remain aware of how seemingly technical questions are in fact normative questions about how regulators ought to intervene for the purposes of public interest. This speaks to the broader question of public interest: it is the government’s imperative to serve the public interest, which entails cultivating economic innovation and securing a fair rights-respecting society for all.

Interested in our company?