Narrow AI and robot lawyers

A conversation with Mischa Frankl-Duval, Research Editor

Of all the people I’m set to interview for Exponential Investor, Joshua Browder will be the youngest.

Last year, Browder launched DoNotPay.co.uk, a site that allowed its users to generate automatic appeals against parking tickets. Having saved users more than £2m in overturned fines, Browder has launched a new product – and it’s set to make an even bigger splash than the first.

His new robot lawyer responds like a (quite a business-like) human lawyer. But it’s free, fast, and learning all the time. Josh and I spoke at 2AM California time about the end of employment, machine morals, and how your car may someday challenge speeding tickets before you even know you’ve got them.

MFD: I was reading a piece in Forbes that features you, it said you were something of a night owl, which clearly still holds true. What are you working on right now?

JB: I’m trying to make improvements to requests in relation to the robot, for example, whenever a user can’t get their request handled I get a notification, and I’ve been receiving so many notifications, I’m trying to look for trends in that data to improve the website.

I was quite surprised at the responses [your robot lawyer has been getting] from the legal community. Why have they been so positive do you think?

I think in terms of what I’ve advertised the site to do, which is generate claims for parking tickets, airlines and PPI it works, and it can answer most questions in relation to those. Because it can do all that, and I don’t advertise it as any more, people aren’t quick to criticise a 19 year-old for just trying to help in those areas where I’ve created something that actually kind of works.

Where do you see the ceiling of the product? The only criticism of it I’ve really seen, has been, understandably, that it’s slightly unsophisticated in so far as it’s a narrow AI with limited applications. But how far could it go?

Well I think Artificial Intelligence, like human intelligence, requires a lot of data and learning and I think because I have so many users it’s actually quite amazing how much I can improve the website because I have models I can test on the basis of what users have asked and what users have sent me. So I’m looking forward to making it quite sophisticated, and I think it’s certainly possible.

You’ve said recently that you can see lawyers being replaced within a year. Do you stand by that?

Certain types of lawyers – obviously the app won’t be arguing in the high court – but there are entire firms of lawyers who just specialise in copying and pasting parking ticket appeals and speeding ticket appeals, and I think if a firm is just doing that, and there are many firms that are just doing that, they should be very worried. So it’s half tongue in cheek, but half serious.

I wanted to talk about machine learning – can you take me through the process, how your machine actually learns?

So there are two aspects to its machine learning: the first is its conversation skills, the second is its legal skills. In its conversation skills, I can test the model, and it can improve, for example if the model is half-unsure about a word it can collate it. SO if a user says “I got a PCF”, it will relate that to a parking ticket on the basis of previous responses, even though I haven’t specifically specified that a PCF is a parking ticket.

And on the legal side, it will improve its questioning… so if you get a parking ticket it will ask a series of questions to determine what was wrong with the ticket, and if its notices that certain appeals are more successful than others, it’ll change the basis of the questioning and prioritise the questions it should ask to get a quicker and more accurate appeal type.

So it’s a probabilistic model, if one path is more successful it’ll divert more claims that way?

Yes, exactly, so that applies both to recognising what a user has said and creating a model for the appeal type.

Which other areas do you see this being most soon applicable to?

Two areas. First, road traffic tickets (Speeding, right-hand turns and things like that). I’m also trying to experiment, to translate the site into Arabic to help with immigration claims, and I think that could be quite powerful, because at the moment, immigrants have no advocate whatsoever. Parking tickets are a first-world problem, but if I can tackle the problem of asylum, that would be really helpful, I think.

I’m quite impressed by how philanthropic a venture this is – all the press on you, understandably, has focused on the financial aspect of the site, but it’s not for profit, and you do care about the people you help, that’s quite clear.

I’m lucky that I have these skills myself, I can develop the app without paying anyone, I don’t have any shareholders to answer to, so the reward at the moment is helping people, and that’s rewarding in itself. And because I don’t have to answer to anyone, that’s been very helpful in allowing me to keep it free.

So you’re not in any rush to be answerable to shareholders?

Well, I think that to make the site sustainable in the long run, I might have to, but I what I’ve really tried to avoid is charging, which would go against the whole principle of the product.

Do you see a financial upside to this site, do you see it being profitable in the future?

I really do. I think the way it can be profitable is by building a technology, an asset, that can be useful to companies in other sectors. I envision a car – a driverless or electric car – automatically connecting with the robot lawyer, and appealing a parking ticket before the user even knows he or she got one. I think that would be useful for those car companies, so I’m trying to contact them to see if they’d be interested in some kind of commercial partnership.

Speaking of self-driving cars – clearly a subject very much in vogue at the moment – which can you see succeeding? I’ve been reading about the potential Apple Car, and, as a non-expert I just can’t for the life of me see it succeeding.

I think I’m looking to make this a mass-use product, so Ford and Tesla both have software development kits I’m experimenting with, to link my app to the car software. I think as more and more companies release these software development kits I’d quite like to do it for every car. It’s not just driverless, it’s more sort of smart cars, which are available widely today.

I’d like to talk a bit about Stanford. I had a professor who taught at Stanford, and she told me that absolutely everyone there, no matter what they were studying, was developing an app of some kind. Is that what it’s like?

Yes. It’s really amazing and humbling. I’m only fighting parking tickets, but I have a friend who’s developing an app to diagnose malaria just by pricking your skin and scanning it with a special phone attachment. That’s just one example, but everyone’s doing something.

Is the faculty encouraging?

Very much so. They have entire classes where you can do entrepreneurial activities for credit. I don’t think that’s something you’d see in the UK, which is one of the reasons I chose the US.

So, speaking of health, is there any industry you’d consider completely off-limits in terms of artificial intelligence and morals?

I think there’s a baseline concern across industries. In transportation, if an algorithm kills someone, who’s responsible? In health, there’s probably a similar concern. I don’t think it’s specific to one industry, it’s a really big issue. I’m trying to avoid that, just doing a simple AI, but I think that’s going to be one of the biggest challenges in the next few decades.

In which industries – health obviously, but people might be affronted, I’m surprised that lawyers aren’t more affronted, that what’s considered such a human profession is under threat. You saw the Future of Jobs report that came out at the end of last year. What kind of shelf life do you put on the human workforce?

This is a big worry for me, for programmers. The definition of artificial intelligence is a computer that sort of programs itself. So I think professions with judgement might last a bit longer, but within 50 years, I think, there will be widespread disruption of industries, even those that supposedly require serious thought. That’s really scary, but I think that could have a lot of positives, and free up a lot of time for people to spend their days differently.

Do you have any interest in research towards a strong AI?

Yes. A lot of my classes here are focussed on the academic side of computer science, and AI is certainly a subset that I’m particularly interested in. I think as part of my academic career AI is certainly something I’ll want to do.

And as for developing a broader, more human intelligence, are you interested in doing that yourself at some stage?

Yes.

Interesting. What do you make of Zuckerberg’s project for the year – developing his robot butler?

I think it’s kind of a cute project, but I think what Facebook is doing with regards to Messenger is much more powerful. I think that’s a symbol of how Facebook is leading AI development. I can imagine that soon we’ll all be talking to Messenger to deal with our AI requests.

What would it improve, the home, communication, what’s next?

I saw something recently about them secretly releasing a developer kit, and that developer kit allows for AI integration, so I think people will start asking people to book their cabs, their flights, and who knows what else once it expands. So I think Zuckerberg’s project should be understood in the context of him trying to lead Facebook and understand personally what it means for AI to dominate people’s lives so he can expand it in a broader sense. It’s more than just a cute project.

Another question – you’re clearly enamoured with what you’re doing, and it’s fascinating, but if you could retrain, and do something else, what do you think you would do?

That’s a good question. I’m really interested in consumer apps, for example I had an idea – I’m not pursuing it so I suppose I can tell you – where you scan a barcode in your house and then when you’re near it, your phone gives you a notification telling you a shop nearby is stocking a product you want. I think consumer tech is a big area, and it’s something I’m very interested in.

You’ve managed to stay very humble, you respond personally to all your users, and so on. So how do you respond to the “boy genius” label, how does it make you feel?

Well, you know, the press has a need to create human interest stories, and I appreciate that need, but it’s a slight exaggeration. I created a parking website. I know some of the other people on these “boy wonder” lists, and they’re really amazing. So it’s really humbling to be among them, and I’m not just saying that, it really is.

Thanks a lot Josh, best of luck.

Thank you.

Category: Artificial Intelligence

From time to time we may tell you about regulated products issued by Southbank Investment Research Limited. With these products your capital is at risk. You can lose some or all of your investment, so never risk more than you can afford to lose. Seek independent advice if you are unsure of the suitability of any investment. Southbank Investment Research Limited is authorised and regulated by the Financial Conduct Authority. FCA No 706697. https://register.fca.org.uk/.

© 2019 Southbank Investment Research Ltd. Registered in England and Wales No 9539630. VAT No GB629 7287 94.
Registered Office: 2nd Floor, Crowne House, 56-58 Southwark Street, London, SE1 1UN.

Terms and conditions | Privacy Policy | Cookie Policy | FAQ | Contact Us | Top ↑