Amazon is to start selling a NLP/machine learning software service that will analyse and extract information from medical records. The new service is called Amazon Comprehend Medical – but will they ever move into the legal AI space? And if they do, what would happen?
Is this just an interesting novelty for medical experts, or does it predict that one day in the future companies such as Kira Systems, Luminance, Seal Software, LawGeex…et al…will be swamped by similar, and probably far cheaper, tech tools aimed at the legal market? Because if Amazon can develop NLP for medical it can do it for legal. The big question is will it?
–
This is what Amazon says its new AI system can do for medical:
‘Amazon Comprehend Medical is a natural language processing service that makes it easy to use machine learning to extract relevant medical information from unstructured text. Using Amazon Comprehend Medical, you can quickly and accurately gather information, such as medical condition, medication, dosage, strength, and frequency from a variety of sources like doctors’ notes, clinical trial reports, and patient health records.’
The company adds that:
‘The service will ‘read’ the text and then identify and return the medical information contained within it. Comprehend medical will also highlight protected health information (PHI). There are no models to train and no ML experience is required. And, no data processed by the service is stored or used for training.’
If Amazon can do this with no model training, then it suggests that they have already done most of the hard work themselves before bringing this to market. The flip side of that is that if Amazon hasn’t developed the precise aspects of NLP you need for what you are looking for then – one can assume – it won’t find it.
That said, the $178 billion revenue per year company, probably has the spare resources to quickly fill in any gaps in its NLP that doctors report to it as this gets rolled out.
But Will They Next Do Legal?
Clearly the medical records and general medicine data sector is massive, and its global. That Amazon is going after this is not then a surprise, unless one stops to consider that Amazon started out selling books to people. Amazing to think how Big Tech can evolve so far, so rapidly.
We have also seen IBM Watson and others go after other areas of ‘healthtech’, such as using AI tools for analysing cancer scan images.
One might say: wherever there is data, there will be AI companies selling their services.
So, does this apply to the law? The legal market is clearly far smaller than the health sector. Big review projects are also limited to mostly larger commercial law firms – for now – which makes that market even smaller.
Law is highly regulated, but then so is healthcare, so that may not be an issue.
Legal language is very specific and full of professional jargon, but then, so is medical language.
Legal texts may tend to have similarities if they serve the same purpose, but variation and exceptions are often also the rule. Can a company like Amazon handle that? The answer is: if they wanted to, yes.
The leading legal AI companies have spent a long time building in pre-sets, and law firms have also worked hard – and still work hard every day – to refine their AI systems for the tasks in hand. But, Amazon has the financial resources to bring together an army of lawyers and get those provisions worked out.
The AI tech behind most of the legal AI companies is not 100% unique. Each have varying approaches to NLP, and some like Ayfie really have taken a different approach, but it seems likely that Amazon, if it wanted to, could get as good results. If you can build NLP for medical, then if you have the time, money and people with the sector expertise, you can do legal too.
So, to the question: could they do it? The answer has to be: yes.
Will they do it? Impossible to say right now. But, it’s a tantalising prospect that Big Tech could one day do this.
What Happens If They Do?
If Amazon went into legal AI review it would almost certainly drive down the prices. In fact, knowing Amazon to be a company that is obsessed with cost-cutting (check out the biography of Jeff Bezos) the pricing impact would be massive.
They’d possibly buy out a legal AI company or two also, to make sure they had the right team of people on board, though with their resources they could probably build a dream team of legal NLP experts without needing to buy whole companies (not that even paying a few hundred million dollars for the top companies would be a barrier to them.)
The incumbents would be faced with a massive challenge. But, law being the law, they would have one advantage: trust. Law firms would probably seek to keep using the services they have been using for years. Trust is such a huge part of legal services that it seems unlikely that top firms would suddenly dump Kira and Luminance to use Amazon just because it was cheaper, though some may be very tempted to do so.
Also, rival Big Tech companies may then also decide they want in. They may decide to jump-start their efforts by taking over the top legal AI companies in the market. The end result would be that there would be very few major legal AI doc review companies left over – if the top five or so decided to sell out.
That would leave some law firms with little choice but to go to Big Tech.
But, as with the law, staying independent, staying relatively small also works. Check out Wachtell Lipton! There will always likely be a number of super high quality legal AI companies that would not sell out and remain, keeping very close relationships to the leading firms.
That said, if and when legal AI review gets cheap enough for every firm on the planet to use, then Amazon may really come into its own.
Will we ever see an Amazon Comprehend Legal (as the mocked up image above explores…..) who knows? But, if Amazon can go from selling books to selling NLP, then anything is possible.
P.S. Amazon also notes that it is already working with PwC, Deloitte and Roche, to name a few, in relation to this capability. And guess what….PwC and Deloitte also work in the legal sector and use legal AI tools…..hmmmm…..interesting…..
—
UPDATE:
Just a few minutes after posting this, a legal tech expert got in contact to say the following:
‘Amazon people are looking clearly at what they want to do in law. They have been partnering with small AI/ML consultancies [here] to start to learn about legal.
Amazon are also speaking with IT teams at big law firms, but their approach is not really much more sophisticated at this stage than ‘give us your data’.
They don’t yet see the complexity and disorganised nature of the ‘data’ within big law systems. They don’t understand the internal structural impediments in law firms to re-engineering aspects of the back office functions. There is an immaturity there. It will change over time as they look harder. But they are looking.’
Interesting stuff!
ALSO:
Re. pricing, check out the rates that Amazon is charging for this:
Pricing
‘With Amazon Comprehend Medical, you pay only for what you use. You are charged based on the amount of text processed on a monthly basis.
Amazon Comprehend Medical offers a free tier covering 25k units of text (2.5M characters) for the first three months when you start using the service for both the Medical Named Entity and Relationship Extraction (NERe) API and the Protected Health Information Data Extraction and Identification (PHId) API.’
It’s not exactly analogous to the charge per document analysed that some legal AI companies use. Is this less expensive…? One would guess that it will be. But, hard to tell as yet.
What do you think will happen next?
I think this is mainly targeted at Google and it’s DeepMind medical arm.
If Google were to take a short at it, and currently we see them not. Then Amazon might. It is interesting as many of the principals are the same. And we are already working in Medical so it’s not a big leap for Amazon to come into contracts.
There was an interesting discussion of some of these questions on LinkedIn about a year ago, prompted by a post from Tom Braegelmann:
https://www.linkedin.com/feed/update/urn:li:activity:6328306423725899776
In particular, check out the comment from Itai Gurari on the potential role of Google in this space.
Thanks, Thorne, for mentioning that discussion last year.
As to the question: Will we ever see an Amazon Comprehend Legal?
Yes, very likely so, if Amazon can expect to make a lot of money with it and have millions of customers (all however more likely in medical than in legal). It will be able to do so if their product is good and easy to use.
But there is the rub: where would the legal documentary data come from? It is usually not available at one company or at one law firm, but it spread out in different places, you would need to aggregate it first, would Amazon do that, could they? Bot eceryone will give them permission. Moreover, applying the law means applying the law to facts, and facts are unwieldy, contradictory, fragmentary and can come from any walks of life as witness statements, hearsay, conclusive evidence, and they are not always available as a written document, but first have to be reviewed and then summarized in a written document. In that regard, such a tool by Amazon will not help as much. And that is the difference to the health system: whil you can have disputes in the health system what kind of a disease someone has, and can challenge a diagnosis, usually do not have contradicting parties that would fight over this, unless you have a case of medical malpractice. So, Amazon Comprehend may make more sense in health because the data intake is, as long as the doctors are competent, not disputable. Not so in law.
I do not see the messiness of legal texts as a barrier to Amazon entering into this space. After all, it is not necessarily to “understand”, say, a contract the way a human lawyer would – with all its inherent, and necessary and strategic fuzziness (http://seanmcgrath.blogspot.com/2017/03/what-is-law-part-1.html) in order to unpack text into meaningful semantic units – even if the answers are not 100% reliable. E.g. entity recognition.
Corpora of contracts are available e.g. SEC filings and elsewhere. Plus, Amazon itself would have quite a few contracts lying around:-)
I think sometimes “100% accuracy” is seen as something humans can produce when analyzing text but machines cannot. Truth is, humans make mistakes all the time 🙂 The fact that in law, the same text can be parsed by different human lawyers and yield slightly (or even highly) different interpretations, is, in fact, a vote in favor of the machines. I.e. given 1 text, N human lawyers and 1 machine, you end up with N+1 exegeses of the text.