O'Reilly Radar Podcast - O'Reilly Media Podcast

Latest Episodes:

Ame Elliott on making security usable and delightful Jun 16, 2016

Show notes

The O'Reilly Radar Podcast: UX for security, architectural inspirations, and problem finding over problem solving.

This week's episode is a cross-post from the O'Reilly Design Podcast. O'Reilly's Mary Treseler chats with Ame Elliott, design director at Simply Secure. They talk about security and privacy design, with a focus on the end user experience, and how to give designers a voice in changing the shape of a product and getting the right values out in the world. Elliott also talks about how architecture inspires her work and why problem finding is a better approach than problem solving.

Here are a few highlights from their chat:

Problem finding

What makes architecture interesting are some properties of what are called 'wicked problems.' A professor in the architecture department at UC Berkeley before my time had a whole lot of things to say about why defining the problem is really congruent with solving it. What that means is by the time you completely write an exhaustive, functional specification for something, you're describing the solution in such a way that it makes a universe of one.

There's a lot of really great thinking around in the built environment, "No one has put a building on this particular site." This design problem is a universe of one. I think that there's a bunch of systematic things around ways in which knowledge we use versus what's specific and particular to this problem that's really pretty interesting. By 'problem finding' I mean that it's less about coming up with a right answer and more about bringing multiple voices into the question—let's work together as a group to find a problem and define the problem, and then work together on making that better.

Technical depth and UX

Being explicitly collaborative, I think, has shaped me in some pretty clear ways, too. I could also go further back in my background, when I was a research scientist. Collaboration is really important. The same impulse that drew me to some abstract technology projects like image processing and machine learning—it's the same impulse that draws me to security and privacy. I see there being a really exciting tension between technical depth and user experience, and how you get the right team together to move forward.

Make security usable and delightful

Privacy and security are tightly interrelated. Privacy or confidentiality is one technical goal of security. There are other technical goals of security—integrity, non-reputability, and other kinds of things. Coming at this from a human-centered design perspective, I'm a UX designer, I care about what end users experience, and privacy feels like the quality that people are looking for in an interaction.

It's less about what's tougher. There's plenty of tough to go around. Really, what I would like to see is designers working together with some of the fantastically talented cryptographers to make security usable and delightful so that end users can experience privacy. In order to do that, there's a real need to help users understand how privacy and security aren't necessarily the same. There can be opportunities for new interactions, new product messages to make it clear to end users who is accessing their data and to what purpose. That could be everything from privacy being a feature that a cloud service company promotes, to a secure system for end-to-end encryption in a messaging application, for example. ... What I would like to see is a new class of interfaces that give people confidence and give people power in how their data is accessed and used.

Getting the right values out in the world

One of the things that really influenced me in my journey toward working on security and privacy was Mike Monteiro's talk at Webstock in 2013 called, "How designers are destroying the world." It's a provocative title, but I think it was pretty eye-opening for me. He used an example of ways in which the users of Facebook can make decisions that have drastic, real-world consequences to people's lives. That was pretty eye-opening to me to think, 'Hey, these aren't just pixels on a screen. There are people behind these systems, and where designers are making questionable choices, there can be drastic consequences.'

Surely, we in a way see this now in a Simply Secure context, where we're looking at things like human rights violations, and globally everything from groups that are working to report evidence of atrocities and sexual violence to the international criminal court, and all the way down to activists and journalists who are trying to make sure that their communications are protected so they can participate in some of the systems that help people get information about the world around them.

Beyond that I think designers do have a responsibility. User experience is critical. I think that design leadership is the piece for unlocking that so designers feel they really have a voice and an agency in changing the shape of a product and getting the right values out in the world.

Ben Lorica on the emergence of intelligent, real-time data applications Jun 02, 2016

Show notes

The O'Reilly Radar Podcast: Emerging themes in the data space.

This week, O'Reilly's Mac Slocum chats with Ben Lorica, O'Reilly's chief data scientist and host of the O'Reilly Data Show Podcast. Lorica talks about emerging themes in the data space, from machine learning to deep learning to artificial intelligence, and how those technologies relate to one another and how they're fueling real-time data applications. Lorica also talks about how the concept of a data center is evolving, the importance of open source big data components, and the rise in interest of big data ethics.

Here are a few highlights:

Human-in-the-loop recommendations

Stitch Fix is a company that I like to talk about. They use machine learning recommendations. This is a company that basically recommends clothing and fashion apparel to women. They use machine learning to generate a series of recommendations but then, human fashion experts actually take those recommendations and filter them further. In many ways it's the true example of augmentation. That humans are always in the loop of the decision making process.

Deep learning inspiration

The developments in deep learning have inspired ideas from other parts of machine learning as well. ... People have realized it's really a sequence of steps in the pipeline, and in each step, you get better and better representation of your data culminating in some kind of predictive task. I think people have realized that if they can automate some of these machine learning pipelines in the way that deep learning does, then they can provide alternative approaches. In fact, one of things that has happened a lot recently is that people will use deep learning particularly for the feature engineering, the feature representation step in the machine learning task and then apply another algorithm in the end to do an actual prediction.

There's a group out of UC Berkeley, AmpLab, that produced Apache Spark. They recently built a machine learning pipeline on top of Spark. Some of the examples that they ship with are pipelines that you normally associate with deep learning, such as images and speech index. They built a series of primitives that you can understand—you can understand how each of these primitive components works—and then you just piece them together in the pipeline. Then they optimize the pipeline for you, so in many ways, they mimic what a deep learning architecture does, but maybe they provide more transparency because you know exactly what's happening in each step of this pipeline.

Learning to build a cake

Deep learning people—and Yann LeCun in particular—have this distinction where, if you think of machine learning as playing a role in AI there are really three parts to it. Unsupervised learning, supervised learning and re-enforcement learning. He talks about unsupervised learning as being the cake, supervised learning being the icing on the cake, and reinforcement learning as being the cherry on the cake. He says the problem is that we don't know how to build the cake. There's a lot of unresolved problems in unsupervised learning.

Structuring the unstructured

At the end of the day, I think AI, and machine learning in general, will require the ability to basically do feature extraction intelligence. In technical terms, if you think of machine learning as discovering some kind of functional mapping one space to another—here's an image, map it into a category—then what you are really talking about is a function that requires variables, so these variables are features. One of the areas I'm excited about is people who are about to take unstructured information like text or images and turn it into structured information. Because once you go from unstructured information to structured information, the structured information can be used as features in machine learning algorithms.

There's a company that came out of Stanford's Deep Dive project called Lattice.io. It's doing interesting things in this area, where they are taking text and imaging and extracting structured information from these unstructured data sources. Basically, human-level accuracy, but, obviously, since they are doing it using computers, they can scale as machines scale. I think this will unlock a lot of data sources that normally people would not use for predictive purposes.

The rise of the mini data center

The other interesting thing I've noticed over the past few months is the notion of a data center. What is a data center? Well, a data center is a huge warehouse near a hydroelectric plant, right? It's the usual notion. But as some of our environments generate more data—think of a self-driving car, or a smart building, or an airplane—once they are generating lots and lots of data, you could consider them mini data centers in many ways, right? Some of these platforms need to look ahead into the future, where they have to be simpler. Simple enough so you can stick a mini data center inside a car. Slim down your big data architecture enough so they can stick it somewhere so you don't have to rely too much on network communication to do all of your data crunching. I think that's an interesting concept; there are companies that are already deciding their architectures around this notion that there will be a proliferation of these data centers, so to speak.

Scalable and sustainable approaches to product design May 19, 2016

Show notes

The O'Reilly Radar Podcast: Ben Yoskovitz on a bottom-up approach to building products and the importance of poking holes in the reality distortion field.

This week, we're featuring a special crossover podcast from our O'Reilly Design Podcast. O'Reilly's Mary Treseler chats with investor, entrepreneur, and former VP of product, Ben Yoskovitz. Yoskovitz talks about product design strategy and the benefits of lean approaches, where product teams tend to fall down, and why a bottom-up approach to product design is more successful—and more scalable—than a top-down approach.

Here are some highlights from their discussion:

Experiment and test

I think we all appreciate that it is certainly getting easier to build stuff. At the end of the day, most of the risk for most companies is not can we build it. The risk is, market risk. Will anybody pay for this thing? Does anybody care?

Can we build this piece of software or build this feature, usually the answer is yes. Most of us are not building rockets to go to Mars, right? That part is pretty straightforward. Where I think product teams do struggle is with, well, how much should I build? When do I stop building and test? Historically, that hasn't been something that people have done.

...You have to build things like experiments and tests. That's hard for people to wrap their brains around, but also for companies culturally to understand.

Bottom-up approach

I believe product teams are successful when the ideas are coming from everybody. I think the days of a senior whatever—I don't want to use titles because I don't want to throw anybody under the bus who might have this title. The VP of product or director of product or CEO says, here's what we have to build. I know exactly what to build. Go build it. All I need you to do is, don't think, just execute.

That's very hard to do. I don't think in the long run that's a scalable model. What's more successful, and that freedom that you described, is, no, we all have ideas. All of them are valid as long as we don't put a giant bet on any one of them. Let's test 10 ideas and see which handful are actually material.

That's more of a bottom-up approach to building product, than a top-down approach.

Reality distortion fields

Where the risk comes is when that reality distortion field gets so strong to the point where you're deluding yourself. That's when, as an entrepreneur, you're running a 100 miles an hour. You crash and you burn and you die. I actually think that's where intellectual honesty comes in. Frankly, we say this a lot about the Lean Analytics book, which is, it's not about exclusively using data. It's not about being so wholly data-driven that you ignore your gut or insights or anything else. It's just about poking holes in that reality distortion field, so you don't crash, burn, and die.

Another way of thinking about this is ego. Entrepreneurs need ego in order to survive. I believe that to be true. You have so much, when you're ego is so big or so strong or so overpowering that you stop listening to other people. You stop recognizing when you're making mistakes. Then, you're going to fail. There's a balance between the 'we're all liars' reality distortion field and intellectual honesty, which I believe is so important for entrepreneurs because it's just so easy to delude ourselves into believing things that are ultimately not true.

Marc Warner on AI's fundamental shift: Supplemental thinking May 05, 2016

Show notes

The O'Reilly Radar Podcast: The short-term and long-term future of artificial intelligence.

In this episode, I chat with Marc Warner, CEO of ASI, a data science and business analytics consultancy and training organization in London. We talk about artificial intelligence, speculating about the future and looking at current real-world business applications of AI. We also talk about a survey Warner recently conducted with data science companies in London, where he uncovered a data scientist skills cap.

Here are some highlights from our chat:

The last problem we ever solve

If we create a general intelligence in a safe and beneficial manner, the gains to humanity could be absolutely enormous; it really could be the last problem we ever have to solve. After that, we spin up our AGI in the cloud and basically everything else is taken care of. Having said that, if we mess up this transition somehow and things go badly, then it could end up being literally the last problem we ever solve and terrible things happen. Hopefully, the actions that we can take now have the ability to influence the probability of either outcome.

AI in the short term

We see a really interesting cross section of what people are actually doing with machine learning and artificial intelligence right now. It's really exciting. Conventionally, software's been used for decades in the form of expert systems where humans would code in explicit sets of rules, and nowadays you just don't need to provide those rules; the machine learning algorithms can pick out the understanding themselves from the data, and it's actually much, much more effective.

For things like custom analytics or recommendation or speech recognition or image tagging, all of these are being used across more and more domains. One of the things that this enables is a continuing personalization of services, so things like smart personal assistants or personalized health care are all going to be in the relatively short-term future of using these tools in the commercial environment.

AI's fundamental shift: Supplemental thinking

If we look at a human as sort of an abstract information processing unit, there are a few things we do: we receive information, we store it, we process it, and then we transmit it. With the advent of the printing press, suddenly we had the ability to store and transmit information with high fidelity and at scale, but in all of human history, pretty much all the processing of information that's ever been done has been done inside a human's head. It's only now, for the first time ever, that we can supplement this ability to think, to process information with machines, and that seems to me to be quite a fundamental shift.

Winning at AI

Probably the winners of this shift are those that are thinking today about which parts of their business are most vulnerable. Alongside this, obviously investing in things like data collection and fairly sensible ideas around using open source tools and cloud infrastructure. It means that they're giving themselves the smallest amount of fixed costs to move on any new trends that come out. I guess this is sort of fundamentally tied to my previous answer, in that if you don't think you can predict the exact specifics of what a change is going to be, you want to make yourself as agile as possible to deal with it in the moment.

Marc Warner will be speaking more about artificial intelligence and the future of data science at Strata + Hadoop World London 2016.

Designing with code and computation Apr 21, 2016

Show notes

The O'Reilly Radar Podcast: Scott Murray on creative coding, data visualization, and STEAM.

This week, O'Reilly's Mary Treseler chats with designer, creative coder, and artist Scott Murray about coding and computation in design, his book Interactive Data Visualization for the Web and his new book coming out soon Creative Coding and Data Visualization with p5.js.

Here are some highlights from their chat:

Design, code, and computation

I use to call myself a code artist but I really struggled with that title, and I think other people in my position feel similarly—those who are doing data visualization, or generative art, computationally based art or design work. I was kind of uncomfortable with the word art. I don't show work in galleries, I don't participate in the traditional art economy, so maybe I'm not an artist but more of a designer. I get excited about design as a problem-solving process. I'm totally a process geek; I get excited about design systems and consistency, so thinking about rules, and values, and data flowing through those rules and how they get expressed. I think, for me, working with code and computation is really kind of a natural fit.

Creative coding approach to programming

I differentiate between what I call coding or creative coding and programming. We're calling this course "Programming for Designers," but it is not going to be a computer science-y approach to programming. This is going to be a creative coding approach to programming, which is to say that the philosophy I'm bringing to this is, 'You figure out how to communicate to the computer to get it to do what you want'—that's pretty different from, 'You figure out the most efficient way of solving a particular problem.'

P5.js

We're going to use this new tool called P5 or P5.js. P5 is this open source, free programming language. You can download it from processing.org; it's based on Java. ... The reason why is it's a tool intended for beginners, and intended for artists and designers. A lot of the language it uses is language that designers are already familiar with. If you're trying to draw a shape you set the 'fill,' and you set the 'stroke,' and you set the 'stroke weight'; you can have red, green, blue values; hue saturation and brightness values, you can have alpha transparency values. It's language that you're already familiar with in terms of thinking of visual properties.

Design and visualization

Visualization is a natural fit for designers because it's leveraging all the visual communication skills, all the problem solving skills that we've already practiced. It's just in a more specific domain.

The STEAM evolution

My sense [of the state of design education] within my own little niche is that things are improving, and they're improving in an exciting way. That's in the sense that we have this whole STEM discussion: sciences, technology, engineering, math. That's really valuable, but that's evolved into a STEAM discussion, where we inset A for art in the middle. The art also represents design in this case, so creative coding fits into that. Getting students not necessarily even into computer science, but getting exposure to these coding skills earlier—that's really exciting.

Using technology to design across the senses Apr 07, 2016

Show notes

The O'Reilly Radar Podcast: Designing a framework to shape how humans experience technology in the physical world.

In this week's episode of the Radar Podcast, O'Reilly's Mac Slocum chats with Christine Park, senior product designer at Basis, and John Alderman, director of Supereverywhere. They talk about multi-modal design, which is an approach to design that takes into consideration the physical senses and the role they play in the user experience, and they also chat about how multi-modal design applies to the Web.

Here are a few highlights:

Christine: A lot of the focus on human factors in design that influences technology is really based on human factors that were developed for industrial design, how we use physical objects in environments. We're entering a new stage in design where we have to think about how we actually use physical information. Multi-modal design is asking the questions, "How do we experience information in our environment? What does it take? What are the limitations? What are our strengths at observing and experiencing physical information? How should that and could that shape the way we experience technology?"

John: The Web, like almost every other media, is multi-modal. It's just that it's multi-modal within a very restricted set of senses and modes. Someone sitting butt on a chair facing a computer screen only has a few modes that they're really engaging with, so design has reflected that. It's stayed pretty static. One of the things that we're trying to do is expand that set of modes, because as people move out into the world, their senses operate in very different ways. ... The Web is a part of a lot of different experiences already, it's just not recognizable as being a browser-based experience. I think that's one of the interesting things—what will the new frameworks for Internet usage become?

John: When photography came along, it freed painters to be non-pictorial. It opened it up. Painting didn't go away. It focused on its core competency, to use modern language. Similarly, with these other mediums, they'll be freed to an extent to focus on what they are best at. That said, there's just so much interesting stuff going on on all the others that designers will have to be fluent in all the others. There's edges where they start operating together that will be really fertile.

Christine: At the end of the day, it comes back to human experience—what we want to do, what we need to do. There are different levels of problems that need to be solved. All of the disciplines have different ways of addressing those problems, and we come together creating these products. We have many different bags of tools to choose from. They get all thrown out on the table in this exploratory period, in this new phase of technology development, the IoT. With all these new kinds of products that are emerging, everything's up for grabs. Whoever can solve the problem will get to solve the problem.

Timoni West on nailing the virtual reality user experience Mar 24, 2016

Show notes

The O'Reilly Radar Podcast: VR UX hurdles, bringing VR mainstream, and preparing for user behavior.

This week, I chat with Timoni West, the principal designer at Unity Labs, where she specializes in virtual reality (VR) user experience. We talk about VR, the UX hurdles designers are tackling, what will drive mainstream adoption, and what we can expect from VR in the future.

West will be talking more about VR at Strata + Hadoop World San Jose 2016 in her session "Virtual reality in 2016 and in the future."

Here are some highlights from our chat:

UX hurdles in VR

The biggest UX challenges I've seen people tackling in various ways are, first, locomotion: how do you move around a space if the space is larger than the physical space you have available to you? If you're going to use some sort of movement mechanic to move the user's camera forward, how do you do that without getting them sick? There's a couple of really brilliant solutions to this already, so I think long term maybe that won't be so much of a big issue as it will be just deciding which one you want to go with. Do you want to use blink locomotion? Do you want to have a slow moving track? Do you want to have a portal-like mechanism, and so on.

Another big one is how do you interact with objects in the world? Obviously, you can use the triggers to grab in a lot of different VR experiences right now. But having things like secondary hot keys if you want to interact with something in a slightly different way—what is the equivalent of the alt key in VR? That's something that comes up a lot in my particular line of work because we're taking a very complicated piece of software and trying to translate it into VR. There's also a lot around button mappings because no one has used controllers like these before. The Oculus Touch controllers are a bit more like a conventional game controller, but the Vive controllers definitely have a fairly new interaction, having grip buttons on the side and having the thumb pads that you can use as sort of a secondary radial menu.

So, teaching people how to use that or figuring out when it's best to use those types of interactions—it's all new now. There's no standardization, nor do I think there should be at this point, but just trying to set things up so that people can smoothly move into interacting in your particular world. I think it's one of the biggest hurdles if you're making a game, or you're making an experience, or if you're making an app, or whatever for VR. Right now, there's a lot of special controllers that have huge text instructions next to them: 'Point here. Click here. Swipe your thumb right here. Right here, look at the arrow, right here.' Even then, people don't necessarily get it.

Bringing VR mainstream

I've shown off a lot of demos to ordinary people or friends—I had my parents come in and my brothers come in, none of whom do anything related to technology. The gear stuff is fairly compelling, especially little kids love it, but when people try out things like Fantastic Contraption or Tilt Brush, where they're actively creating and they are actively manipulating and making their own space in the world, that seems to be where people get the most excited—when they think they have some part of it or some ownership, they're not just looking at a beautiful scene. You can play a lot of video games and never have it occur to you that you can make a video game, right? When you have these creation tools, then I think people really feel like they can do something with it, they can own it. I think that is the thing that tumbles you over that cliff into actually considering maybe dropping $2,000 on a VR headset and computer. It is a lot of money right now. On the other hand, I'm carrying around a $700 tiny computer in my pocket all the time. Clearly, people get used to it.

Hand gestures vs controllers

When I first started in VR design, I was very bullish on natural hand gestures. I was like, 'Yeah, of course. Of course that's what we're going to do—we're going to use nail polish that doubles as a sensor and have our fingers be tracked in space. It seems intuitive.' But the longer that I work in VR, the more I'm bullish on controllers because they have very definite and definable actions attached to them. If I make a gesture with my hand, it could be a yes, it could be a no, it could be a thumbs up, it could be snapping my fingers. Those are things you want to do anyway, and having them remap to a specific verb in a specific app isn't always what you want. I'd like to be able to wave my hand without it opening up a menu item.

Heads-up VR UX designers: Users will try to break everything

There's an Apollo 11 VR experience. It was a pretty well-known Kickstarter. We have one of the demos on our computer. You're actually in the Apollo 11 shuttle and there's two astronauts sitting next to you, you're the furthest one to the left. I make everyone stand up and actually walk out of the capsule because then you can see the moon going toward you and you can see the Earth getting farther away from you. It's a really cool view.

I did the demo once just trying to peer out the tiny little capsule window before I was like, 'Wait, I'm in VR, I can just go through the wall and go look. I don't need to be looking through this tiny little capsule window.' It feels so uncomfortable and people are like, 'What? You want me to walk through the...what? I can do that?' They sort of shuffle along really cautiously and then they get outside and they're happy. Stuff like that is pretty great. VR experience designers should definitely keep in mind that everyone will try to break everything and stick their heads in weird places.

The sharing economy: A big step toward making Marshall McLuhan's Global Village a reality Mar 10, 2016

Show notes

The O'Reilly Radar Podcast: Alyssa Ravasio on founding a company, mining government data, and the future of the sharing economy.

In this week's episode, I sit down with Alyssa Ravasio, founder and CEO of Hipcamp. We chat about navigating the challenges of founding a company, mining government data, and the role the sharing economy will play in the future.

Here are some highlights:

Mining government data

Some states don't have any database we can query, and in that case, we actually have a team of researchers who are reading the websites and creating our own data sets. That's actually how we do almost all of our work. There are some cases where—Oregon has a really cool API for their state parks with photos, and they were really helpful. The static data is readily available and publicly accessible. One of the biggest challenges, though, has been accessing the real-time data, specifically about availability—is the campground booked next weekend or are there two spots left? This is public data. It happens to sit in the servers of a private company that has contracts with the government. Our biggest challenge this year has been trying to gain access to that data.

Hipcamp's Land Share program

Land sharing is this idea that there's a lot of land out there that people own. Sometimes they're struggling to find ways to pay for the mortgage or their property tax, and there's a really good revenue stream out there just from campers who want to get outside. It's kind of a new way to get outside because you're not going to a developed campground where there's a ranger and 40 sites, and everyone is in a circle. It's really a more open and free experience, where you're often getting a gate code to a 600-acre ranch and getting to go camp by yourself on the side of a river.

Making Marshall McLuhan's Global Village a reality

Before the Internet, it would have been really hard to know that there's someone down the road with a car, with a house, with a surfboard, with a beautiful piece of land that they'd be happy to let me rent or borrow for the night. With the Internet, we're so empowered to find these people and to connect in this really new way. I think that's going to result in a society that's much better organized, much more efficient, hopefully a lot less buying, and waste, and consumerism, and just a lot more of a community feel. I like to think of Marshall McLuhan's Global Village and imagine that this sharing economy is really a big step in making that real.

Building your company's team

Team building is like creating a family, so you're choosing people you want to spend a lot of time with and that cultural fit is essential. I think one big thing I've learned this year is you really want to hire for super powers. When you're small, it's tempting to try to hire someone who seems like they can do a lot of things, and it's good to have versatile people, but as you get a little bigger, like five, 10, 15 people, you can start to hire someone who definitely has weaknesses but they have this super power that none of us have, and now other people can help cover their weaknesses. As long as we're aware of what those are, it's a lot better to have those strengths, those super powers.

Finding your co-founder

In terms of co-founders, for me, the hardest part was before having a co-founder. As soon as I had a co-founder, now we're crazy together, but also you don't want to choose the wrong person just because you want to have someone in the trenches with you. I think the best thing you can do—in my case, what I did was build Hipcamp. I built the website all on my own, and I was totally embarrassed of how it looked but it was OK, so I emailed it to 50 people and I just took that first step of saying 'here's what I'm kind of thinking we should build.' By presenting it to my small group of friends, they were able to share it, and then my co-founder came through a friend of a friend who had the same idea. I think if you want to find a co-founder, you should just start building your company and put that out there, and trust they'll find you.

Matt Harris on fintech sectors ripe for innovation and those where elephants have entered the dance hall Feb 25, 2016

Show notes

The O'Reilly Radar Podcast: The maturing payments battleground, bitcoin and blockchain, and insurance innovation.

In this week's episode, Hannah Grenade, a tech entrepreneur and former partner at McKinsey, chats with Matt Harris, managing director at Bain Capital Ventures. They talk about the most interesting areas in fintech innovation, taking a look at some hits and misses, and potential untapped areas of opportunity. Harris also talks about why the merchants payment battleground is no longer a great space for startups and why insurance is poised to be the final frontier for fintech innovation.

Here are some highlights from their chat:

Elephants in the dance hall

Many of those [payments] battles are kind of reaching a conclusion, and that the entry of players like Facebook, and perhaps most notably Apple, have signaled that perhaps this merchants payments battleground is not the best place for startups to be choosing as the market opportunity, that there's a maturity happening, and there's also really a sort of expectation of ubiquity that companies like Apple and the other major technology players have a chance to offer, though, notably, Google certainly failed every time they've tried in payments. So, even if you are already ubiquitous and global and dominant doesn't mean you can introduce a new payment type. I'm seeing less and less in the way of new startups in the merchant payment space. I think there's an acknowledgement that the elephants have entered the dance hall.

Behavior change opportunity

The only company that's really demonstrated large-scale behavior change is Starbucks, and I'm not sure that that's an example that can be followed by too many other players, retailers or other. Starbucks has this incredibly advantageous position, where the customers go once a day or multiple times a day, so it really lends itself to habit formation. They have 95% gross margin, so they can offer, in effect, a discount of 6% or 7% for their loyalty program, and they've got an early smartphone-using, wealthy demographic who are sophisticated and adaptive; I don't know of anyone else who has that same set of characteristics, so I think Starbucks is a little bit of a false positive. ... We've now seen a number of other retailers: Walmart, first, followed by Target, followed, oddly, by Kohls, launch their own eponymous payment method: Walmart Pay, Target Pay, etc. In general, I think there's a couple things that work that could give you reason for optimism. In the case of Walmart, they actually serve millions of underbanked people, so this is the opposite approach, the opposite opportunity of Starbucks.

Solving problems of infrastructure

There's a lot [of companies interested in tackling infrastructure problems]. This really has become the thrust of the bitcoin movement these days, with entrepreneurs like Blythe Masters, who's a fabulously talented executive, with a 20-year career at JP Morgan and now runs Digital Asset Holdings, and she and a couple dozen other companies tackling this opportunity of financial markets, and banking and payments infrastructure, leveraging the distributed ledger idea, the distributed ledger architecture that underpins bitcoin and/or the bitcoin blockchain itself, as ways to think about real-time, fault-tolerant, secure architectures for moving money around. I think it's a very much an "of the moment" kind of idea. It's one that's really hard. I mean, it is one that frequently requires more than one financial institution, many times a dozen or more financial institutions, to kind of sign off because you're talking about counter-parties, and you're talking about, fundamentally, transactions that involve, inherently involve, multiple parties. Those are really difficult social problems layered on top of really difficult technology problems. While it's clearly a popular problem set right now, it's one that I don't think you're going to see any quick wins in, although in the long term you may see some really big companies built.

Reaching maturity

I think that this sort of maturity phase of fintech has pretty firmly kicked in, and that more and more of these one-time [simple bank] renegades are not knuckling under to the realities of our actual financial services world, but rather, I think, maturing to the fact that if they want to truly have scalable impact they've got to have deeper relationships with incumbent financial institutions.

Innovative insurance

If you really want to innovate [in insurance], I think you have to be a carrier. I think the sort of gussied-up brokers...that opportunity existed in corporate insurance, but I don't think there's a breakout opportunity in auto, or a breakout opportunity in life, just for kind of a tech-enabled broker, per se. We have a company called Justworks that is growing very quickly, that is also competing, effectively, with Zenefits, but we think solving a more fundamental problem, which is that for employers who have 50—actually, as of the end of the year, 99 employees or fewer—they're technically "small group," which means they end up getting very bad prices for health insurance. Large group, on a per employee basis, can be 30 or 40 times less expensive than small group.

Justworks is what's called a PEO, meaning they effectively bundle the lives of a bunch of small employers. Now they have thousands of lives, and they've grown 5X this year over last year. They can get large group pricing for small groups, tying everyone together through a really elegant technology and risk management process to make sure that they're taking on risk prudently. That's the kind of thing where, and again, it's a young company, but we feel like if they can execute, there's a value proposition for smaller companies there; on the insurance side, that is a fundamental disruption, that nobody offering...however you polish a small group policy, it's going to be 40% more expensive than what Justworks can get you, and we think, ultimately, that's the kind of innovation that can attract a large part of the market.

B2B opportunities

It sounds like a funny thing to be passionate about, but I am quite passionate about B2B payments. The statistic that shocks most people and still shocks me is that 62% of business to business payments in the U.S. are made by paper check. That to me is like, "How can that be?" It's totally a solved problem in most other countries, and in this country, in terms of retail payments, checks have effectively gone away. I think that this doesn't get enough attention. It gets a lot of my attention, and I think that if you look at the cost and risk, and just lack of modernity that is implicit in that statistic, I think it tells you all you need to know. That is going to change. It may take five years, it may take 10 years, but it's going to change, and in doing so save a lot of people a lot of time and money, and that's the kind of dynamic I want to be on the right side of.

Risto Miikkulainen on evolutionary computation and making robots think for themselves Feb 11, 2016

Show notes

The O'Reilly Radar Podcast: Evolutionary computation, its applications in deep learning, and how it's inspired by biology.

In this week’s episode, David Beyer, principal at Amplify Partners, co-founder of Chart.io, and part of the founding team at Patients Know Best, chats with Risto Miikkulainen, professor of computer science and neuroscience at the University of Texas at Austin. They chat about evolutionary computation, its applications in deep learning, and how it’s inspired by biology. Also note, David Beyer's new free report "The Future of Machine Intelligence" is now available for download.

Here are some highlights from their conversation:

Finding optimal solutions

We talk about evolutionary computation as a way of solving problems, discovering solutions that are optimal or as good as possible. In these complex domains like, maybe, simulated multi-legged robots that are walking in challenging conditions—a slippery slope or a field with obstacles—there are probably many different solutions that will work. If you run the evolution multiple times, you probably will discover some different solutions. There are many paths of constructing that same solution. You have a population and you have some solution components discovered here and there, so there are many different ways for evolution to run and discover roughly the same kind of a walk, where you may be using three legs to move forward and one to push you up the slope if it's a slippery slope.

You do (relatively) reliably discover the same solutions, but also, if you run it multiple times, you will discover others. This is also a new direction or recent direction in evolutionary computation—that the standard formulation is that you are running a single run of evolution and you try to, in the end, get the optimum. Everything in the population supports finding that optimum.

Biological inspiration

Some machine learning is simply statistics. It's not simple, obviously, but it is really based on statistics and it's mathematics-based, but some of the inspiration in evolutionary computation and neural networks and reinforcement learning really comes from biology. It doesn't mean that we are trying to systematically replicate what we see in biology.

We take the components we understand, or maybe even misunderstand, but we take the components that make sense and put them together into a computational structure. That's what's happening in evolution, too. Some of the core ideas at the very high level of instruction are the same. In particular, there's selection acting on variation. That's the main principle of evolution in biology, and it's also in computation. If you take a little bit more detailed view, we have a population, and everyone is evaluated, and then we select the best ones, and those are the ones that reproduce the most, and we get a new population that's more likely to be better than the previous population.

Modeling biology? Not quite yet.

There's also developmental processes that most biological systems adapt and learn during their lifetime as well. In humans, the genes specify, really, a very weak starting point. When a baby is born, there's very little behavior that they can perform, but over time, they interact with the environment and that neural network gets set into a system that actually deals with the world. Yes, there's actually some work in trying to incorporate some of these ideas, but that is very difficult. We are very far from actually saying that we really model biology.

OSCAR-6 innovates

What got us really hooked in this area was that there are these demonstrations where evolution not only optimizes something that you know pretty well, but also comes up with something that's truly novel, something that you don't anticipate. For us, it was this one application where we were evolving a controller for a robot arm, OSCAR-6. It was six degrees of freedom, but you only needed three to really control it. One of the dimensions is that the robot can turn around its vertical axis, the main axis.

The goal is to get the fingers of the robot to a particular location in 3D space that's reachable. It's pretty easy to do. We were working on putting obstacles in the way and accidentally disabled the main motor, the one that turns the robot around its main axis. We didn't know it. We ran evolution anyway, and evolution learned and evolved, found a solution that would get the fingers in the goal, but it took five times longer. We only understood what was going on when we put it on screen and looked at the visualization.

What the robot was able to do was that when the target was, say, all the way to the left and it needed to turn around the main axis to get the arm close to it, it couldn't do it because it couldn't turn. Instead, it turned the arm from the elbow or shoulder, the other direction, away from the goal, then swung it back real hard; because of inertia, the whole robot would turn around its main axis, even when there was no motor.

This was a big surprise. We caused big problems to the robot. We disabled a big, important component of it, but it still found a solution of dealing with it: utilizing inertia, utilizing the physical simulation to get where it needed to go. This is exactly what you would like in a machine learning system. It innovates. It finds things that you did not think about. If you have a robot stuck in a rock in Mars or it loses a wheel, you'd still like it to complete its mission. Using these techniques, we can figure out ways for it to do so.

Our TOPPODCAST Picks

Follow Us

Stay Connected

O’Reilly Radar Podcast – O’Reilly Media Podcast

Problem finding

Technical depth and UX

Make security usable and delightful

Getting the right values out in the world

Human-in-the-loop recommendations

Deep learning inspiration

Learning to build a cake

Structuring the unstructured

The rise of the mini data center

Experiment and test

Bottom-up approach

Reality distortion fields

The last problem we ever solve

AI in the short term

AI's fundamental shift: Supplemental thinking

Winning at AI

Design, code, and computation

Creative coding approach to programming

P5.js

Design and visualization

The STEAM evolution

UX hurdles in VR

Bringing VR mainstream

Hand gestures vs controllers

Heads-up VR UX designers: Users will try to break everything

Mining government data

Hipcamp's Land Share program

Making Marshall McLuhan's Global Village a reality

Building your company's team

Finding your co-founder

Elephants in the dance hall

Behavior change opportunity

Solving problems of infrastructure

Reaching maturity

Innovative insurance

B2B opportunities

Finding optimal solutions

Biological inspiration

Modeling biology? Not quite yet.

OSCAR-6 innovates

Related Podcasts

1

2

3

4

5

6