Artificial Intelligence, Costs and Proportionality

Artificial intelligence is usually presented as a means of reducing legal costs. The proposition has an obvious attraction. If research, drafting, document analysis and administration can be undertaken more quickly, fewer hours should be required and litigation should become cheaper. There is some truth in that. There is also a danger of treating the number of minutes spent producing something as though it were the only question which costs law has ever asked.

Costs have never depended solely upon speed. On assessment the court is concerned with what work was reasonably undertaken, the amount reasonably incurred and, where applicable, proportionality. Fixed recoverable costs add a different economic structure again: the amount recoverable may bear little relationship to the precise time actually spent. Artificial intelligence therefore does not produce one simple “AI discount”. It alters the factual circumstances in which familiar costs questions have to be answered.

The first change is obvious. AI can reduce the time required for some tasks. If a chronology can be produced and checked in one hour rather than built manually in four, the recorded time may fall substantially. That creates pressure on work still charged purely by reference to time. A client is unlikely to be impressed by an explanation that three hours have been charged because that is how long the task used to take before the software was purchased.

But reduced production cost creates another problem: abundance. When an additional draft, research route, chronology, comparison or schedule can be produced almost instantly, the natural brake imposed by expense disappears. Lawyers can do more simply because more has become cheap. That does not make the additional work reasonable. Ten unnecessary documents do not acquire virtue because each took only six minutes to produce.

Checking complicates the arithmetic further. AI may save substantial time producing a first version and require careful human work before the result can safely be used. In some cases that review will still produce a large net saving. In others the apparent efficiency may be largely consumed by checking. The relevant question is not whether AI was involved, but whether the overall method was reasonable for the task. Using a machine to correct grammar is one thing. Using one to identify authorities, analyse evidence or produce court-facing material plainly requires more substantial scrutiny.

There will also be arguments about non-use. As particular forms of technology become ordinary, a paying party may contend that a task was performed inefficiently because readily available tools could have done part of it faster. That argument cannot be accepted merely by naming a product. The existence of technology does not establish that it was suitable for the information, reliable for the task or available within a properly governed professional environment. Equally, there will come a point with some technologies at which refusing to use ordinary efficient methods becomes difficult to justify. Costs practice has encountered that transition before.

Fixed recoverable costs present the issue from the opposite direction. There, increased efficiency does not necessarily reduce recovery because the recoverable figure is fixed by the rules rather than generated from the firm’s time ledger. The commercial incentive is therefore powerful: reduce the cost of production while preserving the fixed recovery. That is not an abuse. It is one of the economic characteristics of a fixed-cost regime. It does, however, increase the importance of designing the work so that automation removes routine labour without removing professional control.

AI may also alter proportionality more subtly by changing what parties consider worth doing. An elaborate line of enquiry which would once have been abandoned because it required twenty hours may suddenly cost almost nothing to generate. That does not answer whether it was necessary or proportionate to pursue it. Technology lowers the price of excess as well as the price of useful work.

The costs consequences of AI will therefore emerge incrementally rather than through a single rule that machine-assisted work is cheaper. Assessment will still require somebody to ask what was done, why it was done, whether the method was sensible, what checking was required and what value the work produced. Those questions are familiar. The interesting development is that the relationship between time and value is becoming less comfortable.

Costs law has never been hostile to efficiency. It is hostile to unreasonable expenditure. Artificial intelligence may make the distinction between those two things considerably more important.