A redistributive VAT on AI into a basic income prevents corporate greed and overreach.

Alternatively, the government could seek to enhance the employability of workers vis-à-vis robots by reducing employer’s NICs, and other costs of hiring them. Reducing employer’s NICs would benefit all employers, including those that were not at imminent risk of shedding jobs through automation. The government would struggle to find alternative sources of taxation to replace the very substantial revenues it derives from employer’s NICs (which comfortably exceed corporation tax receipts). It would be politically suicidal for the government to attempt to reduce the minimum wage or other employee rights.

The government might find it more politically attractive to impose a tax on robots than to abolish or reduce employer’s NICs or employment rights. A tax on robots would help it to reduce the speed of automation and safeguard its revenues, as well as raising funding for the retraining of those who are displaced and income support for them until they found alternative employment (if ever). Automation replaces labour with capital. This would increase the share of national income that goes to capital. Capital is currently taxed at a lower rate than labour income. It would be necessary to increase the tax on capital; i.e. the profits generated by robots to maintain revenue neutrality.

Governments could use some of the revenue raised to fund income support or a ‘basic income’ for those displaced, and measures to make workers more employable. Bill Gates has suggested that the government should also increase expenditure on health, social care and education, where humans have an edge over robots, to create alternative employment opportunities for those who lose their jobs. The government can also offer wage subsidies to firms in respect of those who experience particularly severe problems in becoming re-employed, and fund investment in labour intensive infrastructure projects and public services to boost employment.

But first, what would happen if robots weren’t taxed?

The fact that it is increasingly possible to replace workers with robots does not mean that firms will automatically choose to do so. Even if there were no tax on robots, firms would need to raise the necessary capital, and overcome legal and regulatory hurdles, organisational inertia and legacy systems. A firm’s propensity to automate would depend, among other things, on the relative cost and productivity of robots, the disruption this would create for the business, and the cost of discharging workers (including adverse publicity, customer reaction and the risk of strikes). One would expect the balance to shift to robots as they become cheaper to produce. This is bound to happen, since those who develop the AI that makes robots smart are likely to enjoy increasing returns to scale. (Economies of scale arise when the average cost of production of a product falls as the numbers produced rise.)

The fall in the price of smart robots will make them less expensive to purchase or lease relative to the cost of hiring workers, whose wages tend to be ‘sticky’. One would expect companies swiftly to substitute robots for workers, because this would increase their profits. In the absence of any tax or restriction on the installation of robots, companies would deploy them to the point where the expected net marginal revenues they derive from doing so were equal to their marginal costs. This is the point at which they would maximise their profits.

Why robots would begin to dominate the workplace

As happened when computers were initially introduced, robots are first being introduced as substitutes for human hands and minds. To realise their full productive potential, employers will begin to restructure their entire design, production and delivery processes and command structures to be dominated by robotic technology. Robots would increasingly begin allocating tasks to workers according to their availability and skill sets. Robots are not subject to the same constraints of limited flexibility, capability, consistency and availability that the deployment of human labour introduces into manufacturing processes and service industries. Robotic technology offers the scope of revolutionising the economy to an even greater extent than the information and communications technology (ICT) has done. ICT has affected every industry and service, and every function within them, by systemising and integrating separate departments and functions. The economy could become even more dependent on robotic technology than it has on computers.

Dividing the cake

The use of robots would make the firms that manufacture and use them more profitable. Labour’s share of the national cake would decline. The increase in the share of national income that would accrue to capital, namely the owners of the firms that deploy robots and the smaller number of workers who remain in employment, would exacerbate inequality. The firms that automate would pay more corporation tax, but the government would lose the revenues it derives from income tax and NICs from workers who lost their jobs. The government would also lose the VAT that the unemployed workers would have paid on their personal expenditure (many workers spend practically all their personal disposable income).

Corporate profits are taxed at a fixed rate of 19% (due to be reduced to 17%). There would have to be a very large increase indeed in corporation tax receipts to compensate the government for its loss of revenues.

The biggest winners from the widespread use of robots

The government may have a slim prospect of recovering significant corporation tax from the profits of firms that displace workers to deploy robots. The bulk of the profits could end up accruing to the Silicon Valley technological ‘superstar’ firms that are developing AI technology. It is likely that, as with internet technology, the companies developing AI will begin enjoying increasing returns over time; their initial investment costs may be high, but their incremental, marginal costs will be minimal. Once an AI company reaches scale, its product becomes a standard that companies that produce adjacent products must build around, even if they have better software. Firms that enjoy early economies of scale can use their cost advantage to undercut other firms and become monopolies or oligopolies.

Like the internet companies, the firms that develop AI robotics could begin enjoying ‘network effects’; in the same way that, for example, search engines, social networks and retail websites got better as their user bases grew. These firms would be able to make vast profits because they would own the rights to the software and AI used in robotic technology. They might be tempted to adopt international structures and arrangements to avoid or minimise the tax due on the profits from their patents and licence fees. Some of the big technology companies involved in the development of AI are not renowned for their fondness for paying tax.

The first step in taxing robots: decide exactly what a robot is

There is urgency about the agreement of a legislative definition of a robot because their deployment at work and in homes raises a wide range of potential legal and ethical issues that will need to be resolved. It is also an essential prerequisite for the introduction of any new taxes related to the deployment of robotic technology. The world is rapidly becoming crowded with a multitude of complex cyber-physical systems: machines or mechanisms controlled by increasingly intelligent, computer-based algorithms. But there is currently no clear and agreed definition of what constitutes a robot.

The term robot conjures up mental images of mechanical men or even humanoids like the laconic Terminator, as portrayed by Arnold Schwarzenegger in films. But, in practice, it would be challenging to identify robots by sight. As David Poole has noted, ‘A robot is not a unit equal to a human. Most are not physical robots, they’re software robots. It’s no different, really, to a spreadsheet!’

In the near future, machine learning, a higher level of intelligence, will routinely be infused into commonly used software and computer programs, not just robots. It will be very difficult to distinguish between technologies that destroy jobs and those that might enhance the productivity of existing jobs without threatening them, or create the jobs of the future in the way the personal computer did in its time.

Nature of definition of robots

Given the range and sophistication of robots likely to come into development, the definition needs to be ‘form neutral’; i.e. it should include all autonomous robots, bots and similar smart AI machines. Any proposed definition should be tested from not just from legal perspectives, but also from economic, technological and constitutional approaches. The report from the legal affairs committee of the European Parliament included proposals for a wide ranging definition of robots as ‘physical machines, equipped with sensors and interconnected so that they can gather data’. It noted that: ‘The next generation of robots will more and more capable of learning by themselves. The most high-profile ones are self-driving cars, but they also include drones, industrial robots, care robots, entertainment robots, toys, robots in farming.’ This is a useful start.

Any legislative definition of a robot would need to be flexible enough to capture not just existing notions of what constitutes such devices, but where science is driving them. The primary legislation should ideally provide ministers with the power to expand aspects of the definition by statutory instrument to keep pace with innovation. It isn’t far-fetched to say that science fiction could be a useful guide to potential risks and remedies. For example, scientists at NASA used to consult Sir Arthur C. Clarke, whose novels popularised space travel.

Whether robots should have legal identity

It is necessary to be able to assign liability and determine insurance risks when a robot malfunctions, or where issues of personal privacy arise from the use of domestic robots, or from the deployment of totally autonomous robots. These will necessitate new rules and the adaption of existing legislation to new realities. It may, for example, be necessary to require the installation of ethical principles and respect for human rights as a default setting in all AI algorithms and software.

The European Commission responded to the European Parliament’s request that it come up with a legislative proposal on civil liability for any damage caused by robots by pointing to the Machinery Directive, the General Product Safety Directive and its proposed legislation on medical devices. It believes that its new General Data Protection Regulation will be fully applicable to any kind of processing of data, including that by robots with AI. The Commission’s response smacks of egregious complacency: technological innovation in robotics is accelerating so fast that policy makers and parliamentarians, let alone the law, will find it difficult to keep up.

It has been suggested that there is a need for a specific legal status for more advanced forms of robots. Professor Luciano Floridi, professor of philosophy and ethics of information at Oxford University has suggested that a possible approach would be to adapt ancient Roman law, and attribute legal responsibility similar to that under which the owners of slaves were responsible for any damage they caused. If, however, robots were granted electronic ‘personhood’, they would become directly responsible for making restitution for any damage they cause. This could absolve the owners of the robots, who direct their activities and appropriate all the profits generated by their mechanised slaves, of their responsibilities.

Options for taxation

The government has a range of tax policy possible options. As the legal affairs committee of the European Parliament suggested, the implementation of specific taxes on robotics would necessitate the introduction of ‘corporate reporting requirements on the extent and proportion of the contribution of robotics and AI to the economic results of a company for the purpose of taxation and social security contributions’. There would almost certainly be a rush to automate and lay off workers while the proposal was being debated and enacted. The options include:

Taxing robots: It is impossible to tax robots directly. They are non-sentient inanimate objects with no independent status or legal identity. They don’t get a share of the profits they produce for their owners, or earn any income. ‘Taxing robots’ involves taxing the portion of a company’s capital invested in robots.

Increasing the corporation tax rate: This would not reduce the incentive for firms to replace workers with robots; it would hurt labour-intensive businesses and firms using robots alike, and discourage domestic and international investment in the UK. Raising the corporation tax rate would increase the incentive for firms to invest in robotics if the increase in the corporate tax rate was not accompanied by the withdrawal of capital allowances on robots, because it would increase the relative value of this relief for marginal investment.

Lump-sum taxes: These are the only kind of taxes that don’t create distortions in the economy. They would not impede automation, however, because they are payable at the same level by everyone. Being unrelated to income or consumption, they bear most heavily on the poor. They would add insult to the injury suffered by those whose livelihoods were disrupted by automation.

Taxing firms on the imputed notional income of their robots: It would, theoretically, be possible to impute a notional salary to robots that displaced human workers, equivalent to the wages that were or would have been paid to the workers they displace. In reality, the technology used in automation cannot easily be linked to specific job losses. It is not always possible to say that this robot replaced that worker (or, more probably, this robot replaced those workers). Abstracting from the difficulties involved in compliance, the company owning robots that displace workers could be required to account for and pay income tax and national insurance in respect of their notional salaries. (The company would obviously not be entitled to deduct the notional salaries in arriving at its taxable profits.) The wages of the displaced workers would be used as a reference point for assessing the company’s liability to this tax. These reference wages could be valorised by the growth in earnings or inflation on an annual basis to maintain the real yield of the tax over time.

Robot levy: Firms that lay off workers following the introduction of robots could be required to pay a levy calculated by reference to the reduction in their aggregate wage bill or the level of their expenditure on robotic technology. The levy could be applied on an incremental basis, to protect small companies. (It would be necessary to have anti-avoidance rules to prevent the artificial fragmentation of larger businesses to avoid the charge.) The assessment of such taxes could be modelled on the system in some American states, under which redundancies are tracked to assess the amount the former employers have to pay into unemployment insurance schemes.

Imposing a ‘payroll tax’ on computers: William Meisel has proposed the introduction of an automation tax: ‘If software is to take over so many jobs, why not have an income [sic] tax on software? We could perhaps think of it a payroll tax on computers.’ This would be based on the ratio of a company’s turnover to the number of its employees. Under such a tax, ‘the automation tax would increase … as the revenue-per-employee grows, making it more attractive to create jobs than to replace them with automation’. The advantage of basing the ratio on turnover, rather than profit, would reduce the risk of the manipulation of the figures. Anti-avoidance rules would be needed to prevent manipulation of the ratio by reference to which the tax would be assessed. For example, large companies could bundle their highly profitable technology intensive businesses with newly acquired low wage, labour intensive businesses.

Disallowing relief on the acquisition of robotic technology: According to HMRC’s Capital Allowances Manual, expenditure on computerised or computer aided machinery and robotics is treated as qualifying expenditure on plant or machinery for the purposes of capital allowances. This subsidy for investing in robotic technology could be withdrawn by denying firms capital allowances for expenditure on robots that have AI or are capable of being installed with AI.

Increasing the cost of robots: The government could also substantially increase the rate of VAT payable on the purchase of robotic technology, and deny firms that use this technology the ability to deduct such payments from the VAT they have to account for on the sale of their goods and services.

Increasing the rate of VAT payable on value added by robots: Firms would only have an incentive to invest in robots with AI if they were more productive than workers; i.e. if they added more value and were therefore more profitable. It would theoretically be possible to require firms to monitor the value added by the robots they deploy, and subject it to a higher rate of VAT on the sale of its goods and services. Alternatively, firms whose ratio of turnover to the number of employees was above a specified level could be required to charge a higher rate of VAT on their goods and services. Either approach would reduce or eliminate the cost or price competitive advantage that firms which automated would have over competitors that continued to employ workers. The first option, namely disaggregating the current single rate of VAT to impose a split rate, would not be permissible under the current system. It might be legally feasible after Brexit, however. It would be highly complex, costly and unpopular with businesses. It would also conflict with the government’s unstated objective of maintaining the continuity and congruency of the VAT regime with that in the EU.