Here are our most interesting finds from our week at SXSW Interactive 2016 in Austin. This is not a summary or trend report of all that was going on, just a personal wrap-up from the perspective of us as user experience designers with curious minds.
We had the luck of listening to a great panel talk (they are usually quite terrible!) called “Get the Message! The Rise of Conversational UI”. There has been some buzz around conversational UIs with for example Facebook’s experiment with the virtual assistant M, the shopping assistant Operator and virtual health coach Lark. On the panel was amongst others Julia Hu, CEO & Co-founder of Lark, sharing some good insights about their work. At the core, they see this as a new way of interacting with a service where AI has the power of bringing some emotional connection where a relation is built over time.
The things that stuck with us that we talked a lot about after the panel was:
- Concept of emergence — The service gradually evolves over time when more data is available and based on your interactions giving the user a sense of allusion
- Non judgemental reinforcement — Using best practices from the field on how to give the user positive reinforcement without being judged when not doing the “right” thing.
- Too much freedom means stupid — If you give too much freedom in the interactions the service automatically becomes stupid since it can’t really help you. Constraints are key.
- Real / Not real — Users are aware that they are interacting with a machine still they create an emotional connection. For example, most users think that Lark is female.
- Enough is enough — One thing we really appreciated when trying out Lark is that there is a clear ending to each interaction instead of a non-stop feed of options and information.
She also mentioned that the amount of messages they have handled in the service would equal 21 000 full time nurses (!) operating the service manually. In the panel was also Jeff Xiong, Seven Seas Venture Partners, with an long background in digital services that talked a bit about China’s No 1 messaging app WeChat that has taken messaging way beyond what we are up to in the West. It was surprising to see that most of the audience had never heard of WeChat. On the topic of AI and bots he mentioned that when interacting with one of the bots available in the service, 25% had said I love you to the bot. A bit scary, huh!
What is fascinating for us as designers is that a lot of the problems we solve today concerning interaction with different types of screens will become less and less important. The focus will instead be about designing for service and conversation regardless of device and platform, which will require new tools and competences where we might need to go even deeper into the field of psychology. Being interaction designers this is of course not news it just becomes clearer where things are heading. It’s interesting times and we are really looking forward to integrate this even more in our daily work.
Designing for no UI experiences
Jared Ficklin from Argodesign held an inspiring and thought provoking talk about designing for no UI experiences and what the future in this space might hold. You might have seen some videos from their work in the last years. He started off from today with where we spend a lot of time immersed in our screens, often choosing this over the physical world. Then taking us on a journey not into Virtual Reality or Augmented Reality but Integrated reality.
In this design space digital integrates in a subtle, more seamless and human way with user experience moving out in the open introducing new ways of interacting with gestures, through marked up physical objects, touch and (sometimes) voice. Also using projections (on walls, tables etc) and interfaces on horisontal surfaces. What really struck a chord in us was his very human approach to design and tech with a focus on designing for individuals, not market segments, and cooperation starting in the physical space. Not seven screens in one room as he nicely put it.
Also he talked about how machine learning can give computers social intelligence and also get to know people which will give us designers great possibilities but even greater responsibilities. Basically we’ll need to more seriously consider the moral and ethical implications and values of the systems we design. Not an easy task but definitely something worth pursuing. Even though all this might seem a bit overwhelming to get into he ended with a nice final call to action:
“For a long time we have been creating big systems. Now is a great time to find gaps in those systems. Great value will come from closing small gaps.”
Here are some of the examples Jared showed:
- Room E — A room that interacts with you
- Smart Dumb Things — Ordinary things brought to life through technology
- Trash Barrel Robot — Stanford Center for Design Research project looking at how people interact with an everyday robot in the wild
- And the full presentation from SXSW 2016
Finally on a similar topic, here is a demo from Capitol One’s session on conversational interfaces in banking where they demo access to financial services using voice via Amazon Echo.
Building a better self driving car than Tesla
In the end we were very happy that our workshop ”Synthetic Biology” got cancelled because instead we got to see 26 year old George Hotz talk about his project where he built a self-driving car in his own garage. For those of you who don’t know Hotz he’s the first person to hack the iPhone and later on the PS3.
Hotz is a very smart guy. One of the smartest we’ve ever listened to. We understood most parts of what he talked about but far from it all (Kolmogorov Complexity, Neural Networks and Convex Space etc). But even so: we enjoyed every second of it.
He told the story about when Elon Musk tried to hire him to help Tesla build their version of a self-driving car:
”Elon calls me, says, We’re almost ready to sign, but we need to make one change. We need to have the right not to buy it. No big deal, right? No, no! That’s not a contract, that’s an option for you to buy at an incredible price! I was so frustrated, I said, I’ll build it myself.”
So he bought an Acura and started building his own self-driving car. The most interesting part is the way Hotz approach the solution. Instead of programming the car to do (and don’t do) certain stuff his system is teaching the car how to drive — which makes the car drive more like a real human. He calls this supervised learning:
“I’m teaching the car to drive. I would take the car out, I would turn all the sensors on. What we are coming back to is, it is a supervised learning problem”
The plan is to constantly gather driving patterns from human drivers and create an AI-system that is more intuitive and can work in all potential situations and on every possible road across the globe.
“If you take a bunch of human drivers, say 100 drivers, you will get something like the committee of the superhuman driver”
Grand plans and big words, sure — but the fact is Hotz’ Acura is already working and he wants to ship the product before this year’s end. For less than $1,000!
Stop trying to trick the users
In “Checkbox that ruined my life: Manipulative design”, Dennis Ellis and Kate Swindler discussed the science of behavioral change and where to draw the line between manipulation and persuasion. Really interesting stuff, and something we at Common Ground debate quite often. Not only the methods we use, but how to spot the pitfalls when they occur. In the talk it was also stressed that there exist a lot of evidence that avoiding the dark patterns of design will, in the long run, be beneficial to the business since users don’t like being tricked.
And yes, Swindler did point out the irony of her last name.
A good way to find the risks of manipulation is to look at who benefits from a product or a function. Is it the user, a general win for society, or is it the business? If the two first has the most to gain, you’re normally home free, but if the business alone has the most to gain there is a greater risk for manipulation to occur. Just ask any marketer about their use of emotional drivers in “good” marketing, and you’ll understand why things can go wrong.
You can also ask two simple questions:
- Would i use it?
- Will it improve the users life?
If both questions are answered “no”, you are just a dealer, selling dangerous stuff you wouldn’t touch yourself. Make it your business to avoid those products and projects.
As a bonus, one thing we loved to find out was that when searching for flights on hipmunk.com, the default sorting of the flights is by “agony”. Absolutely brilliant, for long flights this is the best parameter to use when selecting which flight to take. Unfortunately, we can report that their algorithm needs some work. When searching for flights from Stockholm to Austin, the flight we have learned have absolutely the least amount of agony (shortest flying time and no stops in the US) only came in ninth on their list. So: great idea, questionable execution.
A new look at A/B testing
Dan Chuparkoff called his talk “Everything you think about A/B testing is wrong”. Well, what he really meant was that the usual “click-races” are flawed because there is never just one answer. The best answer may only please your biggest group, and it may actually hurt others. In trying to reinvent A/B testing using extensive data science, Chuparkoff discussed how we can incorporate the premise that all people are not the same, and that segmenting is needed to take A/B testing forward.
In short, there are three main problems:
1. There is no single right answer. Different people like different things, and we must accept this.
2. Least Common denominators. This can be described using this puppy-monkey-baby commercial. While Mountain Dew apparently believes that three great things will always be great together, we all know this is not the case.
3. It’s about influence, not clicks. You have to know what the non-clicks mean: are we alienating too many people for example?
The method is using segments. So, you must understand where your users came from and where they can go, and create user segments based on product usage. Talk to different segments over different channels (maybe using segment-optimized landing pages) and match the right contact to the right segment.
Anyway, while all this is possible, it requires a lot of work, data and dedication. So, don’t try this at home, folks.
Last but not least
All the good fun and weirdness that makes you want to come back to Austin again….
Austin. Can’t wait to see you again in 2017!
Common Ground is a UX design studio based in Stockholm.
The Common Ground toolbox holds research, strategy, UX design, user testing, prototyping and visual design.