For businesses, Artificial Intelligence (AI) can represent an attractive cost saving measure. In many cases well implemented AI can improve the profitability of a business. However, the recent settlement in a case involving Uber Eats continues to illustrate that this as yet imperfect technology must be implemented with care.

1. Large language models (LLM) “train” by consuming content. A November 2023 study by Briesch, Sobania and Rothlauf demonstrated that where subsequent generations of an LLM are trained on content which it, itself, has produced, its output rapidly lost accuracy and diversity (following an initial improvement); and
2. Facial recognition software is significantly less accurate when identifying non-white faces as compared to, in particular, white male faces.

In the case of Mr P E Manjang v Uber Eats UK Ltd and others. Mr Manjang, a black man, was suspended from the app which provides Uber Eats’ drivers with access to work following failed facial recognition checks and an automated process. He alleged indirect discrimination and the case ultimately settled in his favour.

Meanwhile, in March of this year the government published guidance on Responsible AI in Recruitment. Law firms are no strangers to receiving hundreds (and sometimes thousands) of applications for training contract positions. Again, the temptation to implement AI in the recruitment process is great – in a competitive labour market where employers receive far more applications than they have allocated the resources to process, automating aspects of this process can vastly reduce the cost of finding the right candidate. But, with the Briesch et al study in mind, consider what might happen if the tool in question was allowed to learn what the business wants in its employees by looking at the current crop of employees, some of whom (and an ever-increasing proportion of whom) it will have selected, or had a hand in selecting, itself.

While the government does not have immediate plans to regulate AI, the Uber Eats case illustrates the pitfalls in the existing law which must be considered in the implementation and development of these AI tools. To avoid inadvertently discriminating, employers seeking to improve their business and recruitment processes must properly understand how these tools work, especially as more technology companies jump on the bandwagon of adding AI, machine learning and LLM technology to their products. Equally those companies who produce this technology will want to consider how their tools might be exposing their clients, and consequentially themselves, to risk.

This list is inexhaustive but shows that the considerations that an employer may have are both internal and external: considering how the manifestation of a belief impacts other employees as well as external stakeholders.

Proportionality

Even when the reason for the action complained of is the inappropriate manifestation of the belief, Higgs v Farmor’s School 2023 demonstrates that the response by the employer must be justifiable in a human rights sense as proportionate interference with the employee’s rights to achieving a legitimate aim, or the employer may face the potential of a finding of direct discrimination.

In Miller v University of Bristol 2022 it was held at the first instance that a warning for the employee’s unreasonable manifestation of his anti-Zionist beliefs would have been a proportionate response, whereas his dismissal was not and was therefore directly discriminatory.

What does this mean for employers and employees?

The current caselaw suggests a balance in favour of the employee. Grainger 5 offers little protection to employers, and even when it can be shown that disciplinary action was taken in response to an inappropriate manifestation of an employee’s belief, the action is subject to the proportionality test which at this stage seems to favour the employee.

However, this is an evolving area of the law. Cases such as Cave demonstrate that overcoming Grainger 5 is not guaranteed, and Miller may yet be subject to appeal. Given some of the factual similarities it shares with Page a different outcome is not out of the question. Nevertheless, the court is under a duty to balance the legitimate aims of employers with the human rights of employees.