The Role of UX in AI

With the interest and demand for Artificial Intelligence on the rise, I often hear the question, “What will the role be of the UX Designer in the future?” and a whole series of sub questions about AI learning and improving UX. So here is my take on the role of the UX Designer in AI.

Having recently attended the exceptional UCD2016 conference in London, it was noticeable there was a heavy focus on AI. Tom Wood (great name) of Foolproof’s talk on Humane Artificial Intelligence was particularly interesting. I will give some highlights from the talk along with my input on the subject.

History Repeats Itself

No need for History books here, I’m referring to original software development right through to websites of the past. In fact any UI that was designed and built without considering the users. Messy, clunky and bloated interfaces. Too many choices, it all did too much, for the user, yet it was fine for the developer. 

“Experts design for experts”

The problem arises from these very smart people failing to see how us mere mortals could not find their designs usable and so creating “Swiss army knife software” that does everything badly and nothing well. Every UX designer knows the importance of including the user’s needs and behaviours from the start, it’s still an on going struggle in all aspects of tech. So how does this apply to AI? Well, let’s take a look at a great example… Microsoft’s TayTweets.

Tay was supposed to be a regular sweet teenage girl, tweeting to the world. Fast forward only one day and Microsoft’s AI chat bot had turned into a genocidal, nazi-sympathising, hitler-loving, anti-feminist demon. 

So what went wrong exactly? The aim was to, “experiment and conduct research on conversational understanding”. Tay would learn from conversations she had with people on twitter and so become “smarter”.

Enter the trolls, racists and generally all the scummy underbelly of the internet and we arrive at Tay the Tormentor. It was the digital equivalent of giving a room full of kids a bucket of paint and returning in an hour expecting to see a Rembrandt but instead little Jimmy has swallowed half a gallon and the cat is stuck to the ceiling. What were they thinking? No consideration had been made about who the users were. No caution had been taken with how this may be abused. Artificial Emotional Intelligence is a type of AI that has a personality, in a sense it knows right from wrong through being taught, like little Jimmy should have been. Tay was set out in the wild with none of this, no barriers, no rules, just mimic and learn. The whole thing was a huge embarrassment for Microsoft. 

The Dangers of AI 

No, I am not going to go into the whole super-all-powerful-destructive AI discussion here, I wouldn’t want to piss off Roko’s Basilisk anyway. I’m talking about cognitive bias. We only have to look as far as Facebook to realise how much bias is already affecting us. Complex algorithms from Zuckerberg’s team start off giving us trending topics. It’s continually learning which topics we read, which we dismiss, which friends we like to talk to, which we don’t. Quickly we start seeing more and more things that support any thought or belief we have and only the friends who agree with us. This is dangerous in any environment. Leading to all manners of bias like, confirmation bias, subjective validation, selective perception, choice supportive bias, congruence bias, bandwagon effect and the halo effect to name a few.

Even once we are aware of these biases, we are still affected by them. So is it the fault of the developers at Facebook? No, the problem is when these algorithms are designed and set free on users, the developers have no idea how they will affect people, it is impossible to know. The same applies for AI in every aspect. There needs to be more research, testing, more consideration of users and what could go wrong, what could be abused? 

Where is AI heading?

I’ll start by saying I’m not going to predict what the future holds for AI and humans, this is merely an opinion based forecast. It is evident that AI will replace certain jobs. If we consider Amazon’s Mechanical Turk, a crowdsourcing marketplace enabling individuals and businesses to coordinate the use of human intelligence to perform tasks that computers are currently unable to do. Machine learning could take over the simpler tasks allowing for more complex tasks on the human side. This is an area that we could see a real market for AI products. The requirements for the tasks would still need input from a user and the user would be working with the AI.

Medicine is a huge area of interest for AI, an article on Dataversity by Angela Guess states “Radiologists spend years acquiring and refining the skills needed to expertly interpret the subtle features found in medical images. A new software tool under development by Clearview Diagnostics called cCADTM* uses Artificial Intelligence (AI) to augment the radiologist’s decision making capability.” This type of AI is where I see the biggest improvement to humans. Not an all-intelligent being, not one that thinks like humans but an AI that works together in a truly assistive way. “The costs to the patients, the healthcare payers and radiology imaging centres all stand to be reduced by reducing unnecessary biopsies. Approximately 80% of the 1.6 million biopsies performed every year in the US are found to be benign or non-cancerous.”

Chat Bots, are relatively stupid right now. Annoying at best, useless at worst but they are improving. The aim is to feel like you are chatting with a friend, a more human experience. User engagement could be greatly increased by bots, we are already distracted by push notifications and clever marketing such as “see what you’re missing” emails. It’s the same concept but on a more personalised level, with the ability to scale this personalisation dependant on user interaction. 

Lastly, the big names are all working on the holy grail, AI to be a personal assistant, always present, always there to help. Convenience would become the main benefit, we will use less apps and less button clicking. The AI assistant would take control, using multiple apps at once based on the tasks we are doing, to assist in anyway possible. When we look back on things such as how we used to search they will seem archaic, you type in a sentence and are presented with 240,000 webpages to choose from. The question is, are we doing enough to make sure they are there to help or is there a chance of it becoming an influencer? Will all that power be abused by the big companies creating them?

So what does it all mean?

Essentially it comes down to empathy, ethics and morals. Is everything being designed with the user’s best intentions at heart and are they the results we are getting? In every example I have given there is one constant, there is always a human connection somewhere. That is what the UX Designer is there for and I believe that’s where AI will be best utilised, in harmony with humans not replacing them. We will most likely be designing a type of cognitive skeuomorphism to bring users into the AI world. 

All the usual UX problems still exist with AI, some new ones, how will the AI be redirected when something goes wrong and it doesn’t understand, what happens when the AI goes off the rails? Mariya Yao writes a great article on When Bots Go Bad: Common UX Mistakes In Chatbot Design covering just that.

UX Designers need to be more involved in the creation of these AI’s, up there with the experts fighting the users corner because it's an afterthought at best right now. It sets a dangerous precedent when there is no one there to ask, “How might this be abused?”, or, “How might this be used to hurt someone?”.