When writing an article on the implications of AI on the legal costs world my immediate focus was to test the present AI knowledge of the field. As such I thought it prudent to seek Chat GPT’s assistance in writing its own article with a limited instruction of ‘write me an article of 1000-1500 words on the impact of AI on legal costs work in the UK and how this will lead to a swifter automation within the field’. While sceptical, after setting up an account and ironically confirming that I was myself not a robot, Chat GPT produced the following article. Simply put the level of precision, albeit broad as opposed to in-depth, was impressive and highlighted the realistic possibility that AI will be prominent in legal costs.
As known to everyone within the legal costs field and those who utilise the services of legal costs firms, it is clear that a large amount of the work is data entry & management. It therefore follows that there is a clear and obvious platform whereby AI could assist in increasing the efficiency of the initial workings required but the question remains to what degree of reliance can be placed on such systems.
At present while a number of factors inputted into modern day drafting software are set in stone (i.e. date and time) there are still a number requirements which need a specialist to ensure these are appropriately set, with the primary example being phasing and content.
In respect of content of the description of work undertaken, while it is assumed that AI would be able to prepare concise summaries of key text, it is questioned whether in doing so critical data would be removed. The specifics of entries as contained within the Bill of Costs are vital to ensure key phrasing is utilised to highlight work of a progressive nature and it is unclear whether at present such precision would be possible via automation.
While the capabilities of AI will undoubtedly exist to identify key words to enable for items to be phased for the purpose of a Precedent H, it is questioned when an item is multifaceted how such software would cope in splitting an item into its required elements. While AI learning is famed for its ability to grow and recognise patterns to enable such practices to be documented, I wanted to investigate how in practice this would likely work. When reverting back to Chat GPT its response was that it would utilise a formulaic equation as to the amount of words within the prose in question and based on this word count would apply the percentages accordingly to the selected key words.
We can therefore see that automation within the field is possible however we must evaluate the reality of this outcome which could be achieved via AI. While it would be entirely possible to apply the above formulae to the individual file note, this would require the fee earner to ensure that their written prose is a precise reflection of the work undertaken without ambiguity or confusion. For example, if an entry read “discussions relating to future trial ideas / concerns” and was undertaken at a pre-issue stage, AI would likely phase this within a Trial phase which is of course incorrect. As such vetting of these file notes would fall to the practitioner who is seeking to outsource such work leading to time wasted rather than saved.
This example however is reflective of a more subjective area of legal costs being the interpretation of time spent or work undertaken. We must however appreciate that the rules themselves surrounding the progression of fixed costs are seeking to remove said subjectivity and is very much an area of legal costs which could be assisted by way of AI automation.
The latest fixed costs regime as set out in accordance with CPR 45 is essentially bound to three main factors 1) complexity 2) value of the claim and 3) the stage of proceedings whereby the claim concluded. Those familiar with the fixed costs regime will be aware that there are some areas of additional consideration to be applied (such as alternative dispute resolution fees or approval for settlement for a child) but in principle there is now limited objectivity, aside from complexity, to be accounted for, with items 1 and 2 being primarily unchallengeable. It is the opinion of the writer that this area of costs, with a case management system which allowed one to track the progression of proceedings alongside a form of assessment to govern complexity, could very quickly automate the level of costs calculable. That being said, as has already been found on historic fixed costs cases, satellite litigation will always develop in some form so such automation would not resolve these issues.
Whether AI is utilised in a more complex area of dismantling information or to act as an automatic calculator based on specific check listed factors it must be understood the risks in automation and AI in law is real.
A failure to properly spot check or understand the inner workings of AI within your systems will undoubtedly lead to serious errors as evidenced in Harber v Commissioners for His Majesty’s Revenue and Customs [2023] UKFTT 1007 (TC) whereby fictitious case law was generated by AI software. Ultimately the signatory will be the one responsible for any such errors and it is not envisaged that opposition parties nor the Court will take malfunctioning AI as a reasonable excuse for failing to fully conduct checks on systems put in place.
At present the future of AI in legal costs will be a fine balancing act. With reference to the words of Bill Gates in his co-authored book ‘The Road Ahead’.
‘The first rule of any technology used in a business is that automation applied to an efficient operation will magnify the efficiency. The second is that automation applied to an inefficient operation will magnify the inefficiency.’
It will therefore remain to be seen whether the latest AI innovation and automation will fall into the former or latter rule but it is this writer’s opinion that if kept simplistic and prepared as an assistive rather than fully automatic tool the use of AI will be a success when combined with the knowledge and expertise of cost professionals to monitor and maintain the accuracy required.