Moral Labyrinth is an art installation, workshop, and website inspired by the Value Alignment Problem in AI, and considers the complexity of conflicting values, and reflects on the difficulty of embedding such values into technological systems.
The art installation, created by artist and researcher Sarah Newman in 2018, has been adapted into five different versions, and has exhibited in Berlin, Linz, London, Boston, and Miami in 2018-19. The Moral Labyrinth workshop, led by Newman, Mindy Seu, and Jie Qi, premiered at RightsCon 2019 in Tunis. A new version of the workshop was conducted virtually in June 2020, and hosted by Northeastern University’s Center for Law, Innovation, and Creativity. This website invites anyone in the world to contribute a question (in any language!), for possible integration into a future physical or digital labyrinth. The site also shares installation shots from previous exhibitions and makes the 2020 version of the labyrinth available for digital exploration.
As machines get smarter, more complex, and able to operate autonomously in the world, we’ll need to program them with certain “values.” Yet we do not agree on what we value: across cultures, across individuals, even within ourselves. We often do not act in accordance with what we say we value, so should these systems learn from what we say or what we do? What are the implications of how our current belief systems manifest in the swiftly approaching technological future? As we anticipate such change, can we use this technological moment to become more honest, humble, and compassionate? Moral Labyrinth is an art installation that has taken shape as a physical walking labyrinth and a virtual 3D labyrinth. The work is a meditation on perennial—and now particularly pressing—aspects of being human. Engaging with the difficult task of aligning values, it gently reveals the gravity of the problem, and creates an open space to reflect on questions. It also hopefully allows us to see our own values more honestly and critically, as a first step toward any solution.