Making AI work in HR: Partner, Catherine Wilson gives some thoughts

In the early 19^th century, groups of Yorkshire and Lancashire textile workers, known collectively as Luddites, protested against the increased use of automation in textile manufacturing and the associated loss of jobs.  Working as an employment lawyer in the 2020s, I am beginning to understand how they felt, as barely a day goes past without some blog or article extolling the virtues of AI and our imminent demise.  Rumours of our obsolescence however, I think, are still premature.

For most of my career, IT has been seen as a threat.  Automated precedents would, we were told, render redundant  lawyers and HR professionals drafting contracts, handbooks, policies etc and certainly this type of work has become increasingly automated, a trend which seems only likely to increase with the widespread use of AI.  Chat GPT and other generative artificial intelligence chat bots can produce, at seemingly break-neck speed, contracts of employment and related documentation.

The flaws with AI and, in particular, the fact that it contains erroneous information, are well documented, but this does not make the impact of those errors any less real or problematic. We are not yet the  51^st state of the USA and the application of American legal principles in the UK context is a particular bug-bear and can lead to serious mistakes.  The fact that the chat bots can be slow to update legal changes and therefore can contain incorrect information can lead to further errors which are then reflected in the document created. Also as any legal student knows, rubbish in rubbish out, so the quality of any documentation always reflects the knowledge and quality of the questions and input of the person asking the chat bot. A first draft is always useful as a starting point, it would be a brave employer however, who simply accepted and implemented these documents without further enquiry or review. Using AI is neither a defence nor an explanation if incorrect documentation is produced and relied upon by an employer.

Beyond the drafting of simple documentation, we are beginning to see the wider application of AI in the broader HR context.  The use of AI in recruitment and so called automated decision making have been well documented but two other areas seem to me to be worthy of further comment.

Use of AI in redundancy

The first relates to the use of AI in the redundancy selection and consultation  context.  AI generated selection matrixes and scores are almost accurate to a fault even if they are subject to the same potential issues in relation to the quality of the original information fed into the process.  AI cannot, for example, create objective data where none exists.  Nor can it replace any repeated, historic failure to evaluate performance, or even carry out regular appraisals. Recorded notes are similarly becoming commonplace and I am sure are widely welcomed as an alternative to the back breaking (and often mind numbing) work of taking a physical note.   So far so good however the exponential  increase in employees’ requests for Data Subject Access Requests is not coincidental in this context. The associated increase in cost and time for employers dealing with these requests can not be ignored, although ironically AI can also assist in answering such requests and reducing this burden.

Grievances and disciplinary hearings

At the same time we have  also seen a massive increase in the use of automated AI-generated notes in grievances and disciplinary hearings. This is an alternative to the physical investigator. Again, this can save on cost and time. But there are pitfalls. Specifically, AI generated notes are so comprehensive that they can include a lot irrelevant information.  Highly accented or quieter participants can be misrepresented or even lost in their entirety. There are no hiding places with verbatim notes which can pick up non-verbal signs and tone far better than written notes which, in the absence of a shorthand writer, can never be any better than a thorough summary of the discussion. It can surely only be a matter of time before Employment Tribunals routinely listen to the recording of meetings as part of their consideration of the evidence.

What is less helpful  is a reliance on AI-derived questioning which is often much less effective than questions created by a HR specialist. Adjourned hearings to allow follow up investigations and further questioning are becoming commonplace to “fill in the gaps”.

 Of even greater concern is the reliance  on AI-generated summaries as formal investigation reports.  These can be inaccurate, or worse, and are no substitute for the HR specialist.  A useful tool, they require review and amendment if they are to be safely used as part of grievance and disciplinary  proceedings. 

Finally

A final plea. It is clear that the growth in virtual, as well as in person, Employment Tribunal  proceedings has led to a growth in regular recording.  The notes generated are clearly being used to draft orders and hearing notes.  From my experience however, these original notes are not currently being shared with the parties involved, although this could be helpful particularly to the disabled and unrepresented parties. Another good use of technology in action. 

The use of AI as discussed above only emphasises the need for the implementation of a formal AI policy and we will discuss the options in a future blog. 

Catherine Wilson is a partner at Constantine Law.