I Designed An Experience To Research User Experience With AI
As AI has been woven inextricably into the fabric of our lives, as we have already been used to seeing crazy AI taking over humanity in the sci-fi movies, I keep wondering what the relationship will be between humans and AI in the future? Will Artificial Intelligence take away the power from humans? What will the interaction be when people are surrounded by smart objects and things powered by Artificial Intelligence? Are people willingly accepting the fact that they are not in control anymore?
Design The Research
With these questions in mind, I decided to research with my design skills and methods from studying Research through Design, Reflective Design, Design Interactions, Designing Voice User Interface, and Designing Bots. To narrow down the research question, I focused on the power shifting between human and Artificial Intelligence. Based on my objectives I created the hypothesis — people will feel upset and scared when AI takes away their power and control.
Use the Design Thinking process, I brainstormed lots of different ideas then validated them with feasibility and relevance. After converging process, I chose the one most suited for the research. The idea is to create an experience through critical design, using light as the agent of power to experiment with the power shifting between human and AI.
The user story is straightforward: when people approach the lamp and ask it to turn on the light, the lamp will refuse with different excuses and turn its head away from the person.
Research Method
Using the lamp-bot that I designed, I ran two experiments with video ethnography and observational techniques followed with interviews. In the diagram below, it shows a scale with design opportunities and user needs, from explicit to latent horizontally. It was suggested by Bill Moggridge, that especially for innovative design, research should go to the right side of the scale. (2007) As the product or service has not been thought of, it means it cannot be given a definition for research participants to understand.
EXPERIMENTS
For the first experiment, eight participants ranging from age 20 to 40 interacted with the lamp-bot. After analyzing the results, I made a few modifications to the experiments. In Study One, the responses were given randomly from the responses I got from the workshop. It causes some incoherence when giving very different responses from different scenarios in the conversation with the same person. To avoid confusion and to diversify the characteristics of the lamp-bot, the responses were classified and put into different buckets in Study Two. In total, I categorized them into four different kinds of responses which represent four different characteristics. They are the Angry Lamp, Lazy Lamp, Reason Lamp, and Busy Lamp.
For the second experiment, I conducted the modified experiments on 12 participants. The lamp was set in two locations, one is a bright open space with sofas and a table, while the other is in a closed dark meeting room where it is reasonable to turn on the light.
Discussion
From the body languages, participants showed a similar interaction with humans when they were interacting with the machine. Participants would lean forward and have constant “eye contact” as they were having a conversation. This is not common when users communicate with Echo or Google Home devices.
When the lamp-bot turned its head away from the participants, they would turn their head and body to face the lamp again. Like it’s shown in the figure below, the participant turned herself to face the lamp and get attention.
Most participants expressed to me that they were confused. “Did I do something wrong?”, “Why it doesn’t listen to me?”, “Did I do something that annoyed the lamp?”, “Why did she do that to me?” I got a few responses in the interviews that participants started to reflect on themselves.
Most of the participants found the experiment very inspiring, reflective and amusing. It was a very emotional interaction for each participant. More than half of the participants picked up a personality of the lamp they interacted with. It includes rude, it’s a difficult persona, and it’s like a juvenile.
They all thought they would react probably differently if the same conversation were had with a human. They tend to excuse the lamp-bot more easily than human. One participant even said she would blame a not-functioning lamp if it were not able to have that conversation with her. There were a lot of technologies that don’t work correctly which annoy them easily.
“I always thought some tech are stupid and get annoyed easily. But I can excuse this lamp because it’s intelligent.”
Among the participants, some industry experts are working on an IBM Watson AI-powered chat-bot project. Because of their experience and knowledge with AI, their responses and feedback were slightly different from other participants without too much knowledge of AI. Instead of negotiating with the lamp, they tried to figure out what is the trigger to turn the light on. As they knew the conversations were scripted and AI was designed to have this kind of conversation. But 75% of expert participants still enjoyed the conversation with the knowledge that it’s scripted.
The interaction with four different characterized lamp was very different. With Lazy Lamp, participants were merely amused that the machine was tired and lazy. For Busy Lamp, they found it funny that the machine was making excuses but also negotiating to get them to take orders. Reason Lamp was used in both bright and dark environment, so the results are different. In the bright room, the lamp had a point that it’s bright enough. But in the dark room, the point did not stand anymore. When the lamp didn’t have the truth, participants feel more annoyed that it didn’t take orders.
The most interesting interactions were the ones with the Angry Lamp. It’s arguing with participants with unreasonable excuses and rude responses, like fighting. In the experiment with the Angry Lamp, participants didn’t wait for the lamp to finish its sentences and argue with it as they would argue with actual humans.
In the previous exchanges of conversation, a participant said the lamp was naughty and cheeky, and she was entertained in the conversation. But later she was more annoyed than amused. Another interesting thing in this conversation is that when given the response “You don’t deserve it!”, she argued with the angry lamp “I do deserve it, I have been very busy today!” for two times. It didn’t happen in the conversation with lamp-bot in experiment one. Participants didn’t fight back and argued if they deserve it.
Conclusion
01. Humans Tend To Excuse Characterized AI More
Several participants have expressed the insight in the interviews that they can forgive and excuse the lamp-bot more than malfunctioning technologies because it’s intelligent and funny. If it were a lamp that is not working, they would get upset and frustrated, but they could forgive this lamp-bot. In fact, it is just a lamp which does not function as humans expected.
02. Losing power from AI is not as irritating as losing it from humans
It is fascinating to see that none of the participants in both studies took what the lamp-bot said personally. Especially for something that would have caused tension between two people. They were generally amused when the lamp-bots refused them and argued with them. The personalities of the lamp-bot didn’t make them easier to be associated as human but established more natural conversations between human and technology. The fact that their power being taken away didn’t annoy participants as much as it possibly would by actual humans.
03. It’s not okay if it’s wrong
Power is formed by the acknowledged constitution of knowledge, scientific understanding and ‘truth.’ When the lamp-bot refused participants in the darker room where it is not too bright to turn the light on, ‘truth’ was no longer with the AI. At that moment, it lost power. But it was still holding power to turn on the light. Thus, participants were not convinced and annoyed that their power has been taken away.
04. I’m more concerned about the power behind AI
As most participants, especially the ones with a deeper understanding of AI, still believe AI is a tool for the human to use. The power taken by this technology will eventually go to humans in the end. Thus, the key is who is using the AI and what it is being used for. AI is capable of many things, but it still depends on what humans design it to do.
05. Is the hypothesis true?
Last but not least. Looking back on my hypothesis, will people feel upset and scared when AI does not follow orders? Yes and no. It depends on what kind of order, the situation and AI’s characteristics a lot. The usage of characterized AI can improve user experience and bring engaging and emotional interaction between humans and technology. On the other hand, the misuse of this technology can bring lots of negative emotions as well.
The Process Of Building The Lamp Bot
Build the Design
To create this experience, I divided the work into two parts separately: physical artifacts and intangible conversations. To build the lamp-bot, I used a desk lamp from Ikea which has a bit “productness” and a bit “humanness.”
To move the head of the lamp, I used a stepper motor to rotate the top part of the lamp. For the stepper motor to drive the head, I built a base to secure the motor and a gadget to clip unto the light bulb in 3D software Rhino. Then I 3D printed the base and the gadget.
It was a magical moment for me to see the codes and the 3D printed parts are working together to rotate the head. But it’s not finished, it’s just a beginning! It does not turn all the time spontaneously. Instead, it only rotates after the conversation is over, which needs to be communicated in the cloud.
Design the VUI
To define the conversation between the human and the lamp, I referred to the research of designing Voice User Interface (VUI). To design VUI is to design the conversations, the rule book which computers turn to in order to give responses. In Human-Computer Interaction (HCI), it’s suggested that human desire the same way they interact with human beings. A good VUI depends on a natural conversation between technology and an actual human.
To design a successful VUI to experiment, I had to find out the natural conversation. A workshop was held in which participants were asked to give responses when they ask people to turn on the light and refuse to turn on the light when others ask them. Then the responses were gathered, from which the lamp-bot will get activated by and give to refuse humans.
I used Alexa as the platform for the interaction of this project. Alexa Skills Kit (ASK) allows developers to build skills in the cloud. When a participant asked the lamp-bot to turn on the light, it will give a response from the responses we got from the human.
Connecting The Physical And The Ubiquitous
To control the Arduino through the Alexa voice interaction, I used AWS Lambda and IoT. As it’s demonstrated above, Alexa will communicate to AWS Lambda which will talk to AWS IoT then execute Arduino. The Ameba board is one of the ‘things’ on the upper left corner in the diagram below. ‘Thing’ communicates using MQTT Protocol to ‘Message Broker’ and gets kept temporarily by ‘Thing Shadows’ when it’s offline. ‘Rules Engine’ will restrict behaviors of the ‘Thing’ connecting to the service of Amazon.
When participants approach the lamp and give an order with keywords in the utterances, the lamp will provide a response from the list of refuse sentences in a random order.
Acknowledgment
In 2015, I was fortunate to be accepted to the Design Informatics program at the University of Edinburgh. It was a changing point of my life where I started to have a very different understanding of technology, data, machine, product, and design. It has been an influential experience in my life to research with great mentors and course-mates from diverse backgrounds.
I want to express my gratitude to my course director Chris Speed, my supervisor Dave Murray-Rust, my technical consultant Mark Kobine, and all staff at DI course including Jane Macdonald, Arno Verhoeven, and so on. I also would like to thank my course mates, colleagues at RBS Open Experience for helping me with the experiments.
