The reasonable coder and smart legal contracts

Where a smart legal contract needs interpretation, who do we consult?

Smart legal contracts are an emerging technology. In brief, they operate as legally enforceable contracts expressed in natural language but containing code, or supported by automated processes that perform actions. The Law Commission of England and Wales recently published its advice on smart legal contracts, concluding that they could be accommodated within the existing legal framework without statutory reform.

A core issue identified by the Law Commission was the interpretation of smart legal contracts. The phrase it used to describe the process was the “reasonable coder”. Although the phrase may sound unfamiliar, the concept of reasonableness is not new to the judicial system. The reasonable-person test is a widely accepted legal standard. Expert evidence is similarly used to provide insight into specialised technical matters.

Clifford Chance described the reasonable-coder test as an effective standard for evaluating the meaning of coded terms and whether a term performs according to the parties’ intended agreement. This brief article considers that test and, more interestingly, asks who these experts might be.

Why do we need a reasonable coder?

Consider this familiar example:

Go to the shop and buy a newspaper. If there are any eggs, get a dozen.

A human naturally interprets this as an instruction to buy one newspaper and, if eggs are available, a dozen eggs. A computer following the wording literally might instead buy a dozen newspapers.

Where smart legal contracts are coupled with autonomous action, it is necessary to consider the consequences and broader questions of contractual interpretation if something like this occurs without human input. If an action is performed on a blockchain, it may be difficult or impossible to reverse while the parties resolve what the contract was intended to do.

The Law Commission therefore proposed ascertaining meaning by reference to what the code, in the reasonable coder’s considered opinion, appeared to instruct the computer to do.

Qualifications of a reasonable coder

Unlike many traditional experts in legal proceedings, computer scientists and software engineers do not necessarily require a particular formal qualification to practise. Jack Dorsey, then CEO of Twitter, claimed that many of the best programmers were self-taught. Stack Overflow’s 2021 Developer Survey lent some support to this idea, reporting that almost 60 per cent of respondents had learned to code using online resources.

The Law Commission used “relevant code” or “knowledge of code” to mean knowledge of the programming language in question. A reasonable coder is therefore a reasonable person with knowledge and understanding of that language. Stack Overflow’s 2019 survey similarly reported that 86.8 per cent of respondents had learned a language, framework or tool without formal coursework.

The Law Commission did not prescribe how an expert should prove that knowledge. It remains an interesting question whether expertise would be demonstrated through a formal test, professional experience or, perhaps, evidence from the creators of the code themselves.

Artificial intelligence as the reasonable coder

A footnote in the Law Commission’s paper refers to work by Dr Thibault Schrepel in support of the reasonable-coder test. It also raises the prospect of artificial-intelligence systems assisting with the interpretation of smart legal contracts by supplementing experts who translate code into natural language.

The potential for bias in AI systems has been extensively explored. A paper I wrote during my degree considered the inherent difficulty of programming a theoretical “AI judge” to interpret and apply black-letter law—in that case, the tort of negligence—to an objective standard. Human behaviour permeates contractual interpretation, as the example about eggs and newspapers demonstrates. An AI system will learn from conclusions reached by humans in the past or from the people working alongside it. Although that is not necessarily a flawed approach, it complicates any attempt to achieve a genuinely objective standard of interpretation.

One relevant risk is false-consensus bias: the tendency to overestimate how widely one’s own views are shared. That bias matters when a dispute turns on whether an event falls within contractual language, which is at the heart of many contractual disputes. Research suggests that judges presented with the same scenarios can also exhibit false-consensus bias.

Conclusion

The Law Commission acknowledged that the reasonable-coder test would be a nuanced development layered on top of existing principles of contractual interpretation. It rightly pointed to the established use of expert evidence in courtrooms as a way to assess technical questions and the reasonableness of human behaviour.

False-consensus bias should remain a particular concern. A coder, expert, judge or AI system may each assume that their reading of code is more self-evident or predominant than it really is. If the reasonable-coder test is to work, courts will need to examine not only whether an expert understands the relevant programming language, but how reliably that expert distinguishes technical meaning from personal assumption.