Design Project / Design Research
In collaboration with: Bureau Moeilijke Dingen
2018Download master thesis →
Artificial Intelligence (AI) is becoming increasingly influential in shaping our everyday experiences. From answering our search queries to shaping our music taste, AI technology is being utilized in more and more products and services. But how aware are we of this?
Breakthroughs in AI-fields such as machine learning have made AI more potent than ever, seducing us with the possibility to make life easier than ever. But this focus on convenience distorts our view on AI, creating a culture of magical thinking, where we are happy to take the output of these algorithms for granted. And with the field of AI being dominated by a handful of tech giants users are becoming increasingly vulnerable to the influence AI will have in everyday life.
Together with Bureau Moeilijke Dingen a future was explored in which machine learning models are not optimized by industry, but rather owned and trained by the user. By using speculative design methodologies an interactive exhibition was developed, allowing people to experience this future and engage in discussion. By reflecting on insights derived from this expo, as well as the design process behind it, a vision on design and AI has been developed for Bureau Moeilijke Dingen, serving to both inform future projects and as a tool spread literacy among designers and society. As AI is here to stay design will become crucial in shaping the relationship between users and the intelligent things that surround them.
This project was my Final Masters Project earning me my MSc. in Design Engineering from Eindhoven Univeristy of Technology.
One of the main outcomes of this project was the development of the Artificial Intelligence Training Center, a fictional laboratory filled with everyday products equipped with AI technology. Visitors could interact with these objects and train them to become better at their job. But the goal of this interactivity was to allow people to better understand what the AI was doing, and have them reflect on the implications of the technology.
The Artificial Intelligence Training Center was on display at 2018 Dutch Design Week, and I was invited to talk about the project during the DDW Talk & Walk: Future Living event. This talk can be watched below (Dutch only)
The telivision present in the Training Center uses an on-board camera and computer vision to detect when a new viewer is present. It uses Amazon Rekognition, an AI-driven image analysis tool from Amazon to estimate the demograpphics of the viewer (age, gender, facial expression) and uses these to recommend a tv program.
Viewers can see the estimations and assumptions the TV made about them. They can also provide feedback about the program choice of the TV by banging it on top when they don't like what they see - much like people would bang on old tv's when they would not function properly. This process would train the machine learning model to perform better in the future.
Having people experience how a machine judges them and choose content based on that judgement engendered a lot of discussion about how much value we can attach to these judgements.
The security camera is able to recognize individual people, and gives them a rating on how suspicious they look. It is powered by very potent, but easily implementable Python library called "face_recognition". With this library a single photo is enough recognize a face with a high degree of certainty.
Attached to the security camera are a monitor and interface. Here people can rate other people (who have already been captured by the security camera and agreed their image to be used) on how suspicious they look. By doing so they provide the security camera with data to train on.
The goal of the experience was to have people reflect on the subjectivity of data sets. Althought this is an extreme example, there is simply is no such thing as purely objective data.
The letterbox looks like a regular letterbox, but is outfitted with AI technology to detect spam or unwanted email.
Whenever a new letter is put into the mailbox it tries to determine whether or not it is spam. If it thinks it is, it shoots the letter back out again. If not, it keeps it nice and tidy. It can be trained to keep certain letters by keeping the lid shut and preventing it from launching the letter.
This artefact was meant to help people grasp the sheer amount of data needed to effectively train a machine learning model: having it unlearn to reject a certain letter is something that requires a whole of training.
The nightlight is outfitted with a microphone so that it can detect the sound of snorring: a cue that it can turn off for the night.
Once a training session is started visitors are invited to snore or make another sound into the microphone. The nightlight will try to do its best to determine what it just heard, and turn off when the sound is classified as snorring. However, if it is wrong it can be correceted through its pull switch. This action provides feedback to the system and trains it to become better at detecting snorring.
Visitors were invited to actively interact with the artefacts at the Artificial Intelligence Training Center, and engage in discussions about there view on AI.
The design of the artefacts is very archetypal. The AI functionality is highlighted with orange design elements. This was done so that viewers immediately understood what object they were looking at, making it easier for them to focus on the new functionality AI brings to these products.
While the Artificial Intelligence Training Center played an important role in the project, it served a borader objective: to explore how we might live with AI in the future everyday, and the opportunities and responsibilities of design in bringing this technology to everyday experiences. This resulted in the AITC Manual, a propsed framework for making design decisions in regard to AI.
The AITC Manual is a proposed framework for design decisions regarding AI. It is more extensively described in a booklet that can be downloaded below.