• Home
  • Latest
  • Podcasts
  • Mishcon Academy: Digital Sessions podcast - Machine Learning and AI

Mishcon Academy: Digital Sessions podcast - Machine Learning and AI

Posted on 10 November 2020

The Mishcon Academy Digital Sessions.  Conversations on the legal topics affecting businesses and individuals today. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

In this episode, what’s the difference between artificial intelligence and machine learning?  How can the data they leverage add value to an organisation and what are the key things to consider when adopting this technology? 

Hello and welcome to the Mishcon Academy Digital Sessions podcast.  I am Tom Grogan, Head of MDRxTECH, Mishcon de Reya’s digital transformation business, and I am joined today by my colleague Dr Alastair Moore, our Head of Analytics and Machine Learning.

So, Alastair, a pretty big question to start us off with, what is artificial intelligence?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

Hi Tom, thank you for having me.  So, artificial intelligence, or AI, it describes a system’s ability to perform something that we think might be intelligent.  So, the things that we might associate with human-like behaviours through an intelligent interpretation of an environment or some data and often it’s an aspirational terms so a lot of the time as soon as we know how something’s working, we sort of discard the terms so, you know, is the DLR train an AI because it navigates around a track without a driver?  You know, not really but in the sense that it, you know, it’s still using systems that sense their environment and adapt in a meaningful way. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

Okay, that’s helpful, thank you.  And what’s the difference between AI and machine learning?  We often hear them used interchangeably. 

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

So machine learning or ML, you know, is a more recent approach to trying to create AI systems and specifically what we mean is, the system is taught or it learns from data.  So, in a traditional system or a computer programme, we normally start with a set of inputs and then a set of rules and the rules themselves transform the inputs into outputs, whereas in machine learning we sort of swap it around and what we do is we collect lots of inputs and outputs, what we refer to as the data, and then the learning is automatically working out how to map between the inputs and outputs, the sort of learning the rules. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

I am always incredibly jealous of your job title, Head of Analytics and Machine Learning.  What sort of things do you get up to on a day-to-day basis, both within MDRxTECH and beyond?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

So, I think it’s not as glamorous as it might sound, you know, if the title was someone that looks at stuff in data bases, probably more of an appropriate characterisation of what we do day-to-day but the team I work with does sort of four things principally.  So, the first thing we do is analytics, we look at the data that our clients or the firm’s already collected so, for example, the data that systems are collecting as part of a business doing its ordinary operations, we analyse it, this is the analytics part of the title, you know, what did we do, when did we do it?  We use this to diagnose what has happened in the past and we use it to influence business decisions.  So everything from operational efficiencies to reporting on how the law is practiced.  So for example, we recently submitted a response to a call for evidence for judicial review in the UK and the report included an analyses of ten years of judgements showing how the cases change over time, what Government departments have been involved and by and large analytics function looks backwards into the past so we are looking at what has happened to derive insights that might help us frame what we do in the future. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

And how about forward looking?  How do we use data to map our path moving forwards?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

So the second thing we do really is this kind of idea of business process improvement that, you know, me and the team I work with, we review and advise on how business processes and the systems collect data, how should we use it?  You know, if it’s collected here, can we use it elsewhere?  And mostly this is like you said, forward looking, you know what’s the best way to do this?  How can we make it better?  We look at products, we evaluate different features, we decide how best to support the various business functions.  So for example, recently we’ve been working with one of our MDR Lab companies, Ping, to automate time recording and allow our lawyers to focus on a client’s problems rather than administration. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

Excellent.  And one of the key distinguishing features of MDRxTECH is clearly our engineering function, our ability to actually build stuff.  In what way do we find that that skillset is beneficial to clients?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

Well, you know, the practice and the activities we do in the firm means that we build up a fair degree of expertise in these types of systems so, the way we can use this to help clients, it might be, you know, general advice on how a system collects and processes data and how that fits with data protection law, you know, secondly, it might be sort of subject matters specific analysis regarding the law, I just mentioned the judicial review but, for example, you know we were working in real estate or finance or food and health and safety and then it might be more detailed engineering advice on the advantage or disadvantages of a specific system so, we actually build the systems in-house, we have a degree of expertise in them and we’ve actually started building systems like machine learning engines for clients on specific projects. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

And research clearly forms a key part of what we do, the rate of progress in these technologies is so rapid, they’re not new, we’ve had things that we might refer to as AI or machine learning for nearly a hundred years but I think we are really seeing now that they are defusing into the real economy and improving significantly in performance.  How do you go about making sure we are on top of the latest trends?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

So yes, staying on top of it’s quite difficult and is a sort of key component of the work we do because it means figuring out what’s working, what’s working best and also trying to adapt the research base practically so, thinking about practical applications in industrial uses and how they might vary so, we read papers, we publish papers, you know, we work with PhD level students on research, we work with several research organisations and we work with partners and collaborators so, a good example of that would be vLex Justis analysing legal data for example.  And it’s not just about machine learning, we are working with other data technologies like graph databases or, you know, a blockchain or distributed ledgers that you might have heard about. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

So, as you’ve just said, the technology is improving all of the time and the rate of progress we are seeing is really significant.  What do you think is the most significant advance you’ve seen in the last couple of years?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

In the last couple of years there’s been huge improvements in natural language processing or NLP and that’s the ability of computers to recognise and understand text or words.  In general, over the last ten years the advances have been around bigger and bigger models or what’s referred to as deeper models that are able to store a greater number of correlations over larger and larger datasets.  Some of these datasets are now web scale, they are huge and they have lots of general information in them and therefore the models can start to be used in fairly general circumstances, you know, it’s become so popular that it’s hit the press and algorithms like GPT-3 have made the editorial of The Times so you know that something is going on.  One of the big advances which we are still trying to understand, or realise in industry, is called transfer learning.  This is where we take on a machine learning model that’s trained on one large dataset, for the sake of argument, and then it gets retrained on a different, possibly related, task but using a much smaller dataset and it’s interesting because that means that the notion of big data or the requirement to have lots and lots of data, at least for the end task, starts to disappear so you can start to get things working even when there’s little data or it’s very difficult to get hold of or it’s expensive to collect. 

In NLP I also think that you are starting to see a convergence of several different technologies.  So graph databases, in terms of fixed sets of relations and language models, trained on huge datasets and thirdly, weak learning sets of rules, rules that might not be right all of the time but convey a sort of general idea of a set of relations and over the next few years we are likely to see a convergence of these things that mean that it’s much easier to pull out and use information from a traditional database and also create much more reliable language models.

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

Great.  And how are we seeing organisations leverage their data and how can we help them?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

So, AI, you know the umbrella term, artificial intelligence, has been on the radar for most companies across all industries but it’s difficult for you to know exactly where to start, you often need help navigating the different technologies while having the confidence that they are going to maintain their legal and regulatory compliance, you know, and upholding your company’s digital ethics in some sense.  You know, when I teach technology on the Master’s or the MBA programme at UCL School of Management, I start by getting managers and executives to break the problem down, you know, what are the inverts, what re the outputs, what data are you trying to collect, where is it stored?  One of the main management challenges is starting to think about these systems probabilistically.  How wrong could the answer be?  How frequently could you get the answer wrong but still on average it provides utility?  Most management education isn’t really geared to this and so starting to be able to think creatively is often useful in trying to identify where machine learning opportunities lie within your business. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

And what are the first steps often for these companies looking to adopt AI in their practises?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

So often the first thing that needs doing is an awful lot of looking at data so, identifying, curating, improving the quality of datasets before you can get anywhere near anything that’s kind of interesting.  So, you know, what data do we have?  Are we allowed to use it for this purpose?  Who else can we get data from or who can we share it with?  Often there are lots of knotty problems around mixing and matching and aggregating datasets because it can be then used to create value for a business.  So getting comfortable from this from a governance perspective and a legal perspective is something that we can definitely help with.  Often, for example, we just spoke about transfer learning, you want to be able to take a machine learning engine trained on that data and use it on this problem or you want to be able to share data between various sets of parties and agree how and when the data can be used and for what purpose.  Again, this is an area where MDRxTECH can provide expertise and create real value.  Things like setting up data trusts, auditing processor agreements, thinking about the bias and discrimination with regard to your obligations under the law, you know, maybe we’ll say a little bit more about this in a minute. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

So Alastair, what sort of thing should organisations be thinking about when they consider the implementation of AI?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

There are several different things that you need to think about.  Commercially, you need to start by identifying where it’s going to create value.  Technically, you need to think about what appropriate systems to use, how do they work, how can you explain how they work within your organisation?  Legally, compliant by design, how do the prevailing legal and regulatory frameworks apply to the system you are creating?  And ethically, does it really do what you want?  Is it aligned with your organisation’s values?  You want a degree of comfort that the new system is going to achieve its aims and not create new problems that you are unaware of. 

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

So having set out that approach, can you give us a little bit more detail on the commercial and technical elements?

Dr Alastair Moore, Head of Analytics and Machine Learning

Mishcon de Reya

Firstly, you know, can you create more customers?  Technically there’s things like lead scoring, market intelligence, optimising the marketing funnel.  Secondly, it’s serving your customers better, you know, by creating more of what they want so that’s things like recommendation engines or optimising a sales funnel and thinking about maximising lifetime value.  And thirdly, you can think about how to serve those customers more efficiently so you can improve your processes through automation or probably on the more experimental end of the spectrum, you can think about things like workforce optimisation or process prioritisation.  And technically speaking, you have to audit what data you’ve got so, you know, what data’s available, what’s the security processes around the data, what’s the provenance of the data, where did it come from, am I allowed to do what I want with it?  Actually, you know, how might it be possible to achieve these things that I want to do?  And the MDRxTECH team can start by helping you frame those problems but we can also go through all the way to the details of the engineering based on the experience we have.  Maybe I’ll flip this round, maybe I’ll ask you a question, what do you think are the most important things to consider when you are thinking about implementing AI from a legal perspective?

Tom Grogan, Head of MDRxTECH

Mishcon de Reya

I mean legally, data protection compliance is absolutely key, fairness, transparency, accuracy, data minimisation, the exercise of data subject rights but you’ve got to consider a number of things, for example, discrimination under the Equality Act, you’ve only got to look at the recent successful judicial review of the Home Office’s Immigration algorithm that was challenged on the grounds that it created a digital hostile environment.  Navigating this environment can be hard and the MDRxTECH team work with clients all the time to help ensure their implementations are compliant by design in that respect.  Ethically, it’s really interesting as well, it’s so important for all organisations, public or private sector, that they uphold good data ethics in all of their implementations and practices, there are a number of useful resources that they can use to help with this, the European Commission, their high level expert group on artificial intelligence has produced an AI Ethics Guideline document, the ICO in the UK has published an AI auditing framework and the UK Centre for Data Ethics and Innovation published a data ethics framework back in 2018.  All of these things are great but in reality, organisations wishing to actually ensure they uphold good data ethics, need to look in quite granular detail at their own processes and see whether their proposed implementations are actually appropriate in this respect.  MDRxTECH, of course, are on hand to help them conduct this analysis. 

Well, for now, let’s wrap up here.  Thanks so much to Dr Alastair Moore for joining me in this Mishcon Academy Digital Sessions podcast.  I’m Tom Grogan and in the next episode, my colleagues Jo Rickards, Paul Noble and Matthew Ewens will give us an insight into HMRC’s approach to tax investigations and the powers they have to conduct them.  We expect to see far more tax investigations in the near future given that the Government will be under pressure to maximise tax receipts in the next few years having supported the economy through the Covid pandemic. 

The Digital Sessions are a series of online events, videos and podcasts, all available on Mischon.com and if you have any questions you’d like answered or suggestions for what you’d like to have us to cover, please do let us know at digitalsessions@mishcon.com.  Until next time, take care. 

The Mishcon Academy Digital Sessions.  To access advice for businesses that is regularly updated please visit mishcon.com.

Mishcon Academy: Digital Sessions are a series of online events, videos and podcasts looking at the biggest issues faced by businesses and individuals today.

In this Mishcon Academy: Digital Session podcast, Head of MDRxTech Tom Grogan and Head of Analytics and Machine Learning Alastair Moore, discuss artificial intelligence (AI) and machine learning, the ways in which they can leverage data to add value to organisations, and the key things to consider when adopting them.

How can we help you?
Help

How can we help you?

Subscribe: I'd like to keep in touch

If your enquiry is urgent please call +44 20 3321 7000

Crisis Hotline

COVID-19 Enquiry

I'm a client

I'm looking for advice

Something else