Can Smart Contracts Solve the AI Limitations Problem? Clause Thinks So

One of the problems with legal AI tools, such as the NLP/ML systems that review documents, is that they are not perfect. Lawyers are not perfect either – as parallel tests have proven – but, still, many lawyers want tools to tell them exactly what is in a contract, and very quickly. Peter Hunn, Founder of Clause, believes smart contracts could be the answer. How? Read on.

The argument goes like this: a smart contract contains coded elements that relate to the key aspects of the agreement between two parties, e.g. payment terms, dates that a payment will be made, conditional logic around how those terms may change under certain circumstances and many other key factors that a lawyer may want to find out.

If those coded elements are – as you’d expect – contained in a digital database, and can be tied directly to that particular contract, then you have a very powerful way of instantly knowing what is going on in every one of your contracts.

In fact, if you could create a dashboard that gave you visibility of all the coded aspects of all your smart contracts you’d not just be able to see what was agreed across all of them very easily – and with what one could say is ‘perfect recall’ – but as they are now ‘live’ documents with APIs connected out to the real world of many different data feeds that act as triggers to those coded elements, you could see the actual state of each contract in real time across an entire business.

In effect a corporate could gain complete visibility over its contract stack in ways that would be very hard to achieve at present using NLP/ML tools.

‘NLP is patching a problem, not solving the problem,’ says Hunn during a catch up with Artificial Lawyer in London. ‘It’s very easy to create structured data. And, if we keep it as structured data then we solve a lot of problems.’

Interestingly, Linklaters‘ legal AI team, Nakhoda, is coming at the problem of NLP in a slightly different way: make parties agree contracts on such a specific template that when you go back to review it the chances of getting an accurate picture are massively improved.

And, Hunn adds, it’s important for people to consider that smart contracts don’t need to be self-executing. 

That surprised Artificial Lawyer, as the self-execution part of smart contracts has been part of the story for some time. But, Hunn notes that the concerns of some lawyers over contracts all firing off with no human oversight are justified. So how would they work then?

The contracts would inform the parties first, if a lawyer decided they didn’t want it to execute automatically, at least in terms of performing an action that impacted something external to itself, as opposed to simply recording a change of state due to new data that has been fed into the system.

I.e. a contract is about to make a payment to another party, it pings you an email/alert/text etc, saying this. You confirm you want that to happen, and off it goes.

This is a beautifully simple human solution to the ‘smart contract run amok’ argument that some say is a risk.

But, then what is the value in smart contracts? Well, see the part above, about the contract being live and also having much of its key contractual aspects in a coded form, which allows a lawyer, company, or any user, to have total visibility of what is in that documentat least as far as key terms that have an ‘active’ element, e.g. payments, schedules of dates, conditions and the like. 

Would such coded elements contain all the subtle language contained in every clause? Nope. But, do you need that in every case? Sometimes you do want to use a legal AI tool to review every document in super detail, getting right into every clause, looking for anomalies and variances. But, sometimes you don’t, and the smart contract approach would be the quickest, most accurate way of doing things.

‘Legal AI companies do more work than is necessary. It’s a self-created problem. You already have structured data there, you just need to use it,’ Hunn concludes. 

So, there you go!

One additional thought is that, this is great for contract stacks that are made from today onwards in this way, but there will still be a huge historical tail of old contracts that will need NLP to get into them. And, many companies and law firms will keep on churning out these ‘air gapped’ traditional contracts for many years to come. This suggests that legal AI companies will be around for some time to come.

But, the fascinating thing is that from Hunn’s perspective, in the very long term such systems would eventually be less and less of use.

Interesting stuff…….(to be continued…)


  1. “many lawyers want tools to tell them exactly what is in a contract”

    So do clients

    AI should be used to rate the readability of the outpourings of lawyers.

    My lease is so bad that when I told the landlord that a demand had a wrong date in it, he couln’t find it and had to ask me where it was in the document. And he had signed it and paid a lawyer for it

  2. The “beautifully simple human solution to the ‘smart contract run amok’ argument” defeats the purpose of using a blockchain in the first place. If the parties can’t count on deterministic execution of the terms of the contract, then an off-chain enforcement and remedy solution is necessitated and reps/warranties and due diligence that were largely or completely avoidable with a deterministic smart contract reenter the contracting relationship. Pseudonymity must be given up as well. The entire dynamic collapses into a traditional contract with most of the benefits of using a blockchain lost while most of the limitations and costs remain in place. And all this new infrastructure just so uniform terms and easily specified values can be captured, distributed and preserved in machine readable form without further human intervention? Well, the easy stuff is already being done that way online. No need for “smart contracts” that are actually rather dumb to accomplish the easy stuff. As for the hard stuff for which machine readable code is not sufficiently expressive or nuanced, smart contracts aren’t going to work anyway.

    The issue isn’t AI solutions vs. smart contracts to eliminate the need for human contract review. It’s all of the above vs the complexity of modeling the real world of contractual relationships. Picking the easiest-to-reach fruit like we see all of these vendors using trivializes the scaling problem, the data rot problem and, worst of all, the very human need to differentiate rather than conform and to act based on personal experience rather than collective standards set by others.

  3. A few weeks ago we won the Global Legal Hackathon in London by turning terms and conditions into a structured data model. This concept has been around for some time, and it makes sense that someone technical would make this connection. The next step is to fully understand the complexity involved in this approach on the legal side.

    Just one example, would be tagging elements of the contract in your structured data model. The distinction between warranties, terms and innominate terms isn’t clear cut and is often argued in litigation with many perspectives. In the case of breach, major terms are conditions, minor terms are warranties. One allows termination of a contract with damages, the other only damages. That is the essence of law, in litigation lawyers will argue these points no matter how smart the contract is.

    The only way I can see this working, is if we can almost apply litigation before the contract is executed. So contract and litigation lawyers working together to form the contract into a structured data model and it then it cannot be challenged once agreed.

Comments are closed.