Possible Worlds: our company
We believe that a company should not only provide solutions; it should also ask questions. Here are some fundamental questions we have about our trade (Artificial Intelligence), about the kind of organisation we want to be, and ultimately about our planet. We have been thinking about them for a long time, so we link to past and current projects of ours (✍).
Rethinking the world
Possible Worlds does what it says on the tin: our organization is concerned about envisaging alternatives to the status quo. When it comes to AI, we find that current systems are built around a notion of prediction which encourages the reproduction of existing patterns – the good and the bad. The widespread adoption of such systems means more entrenchment in the patterns of the past and less freedom to think outside of the box. We want to campaign for radical change in the way AI algorithms are built and thought about, and make them support human creativity rather than restrict it. But how do we start shifting long-lived assumptions about what makes ‘good machine learning’?
✍ On radical change and promoting emancipatory social goals: The ends of utopian thinking (monograph, in press). More specifically on AI: The ethics of generalization, Infinity is not everything (blog posts).
A scientific understanding of language
Current AI has grown around a mistaken notion of language, which has far-reaching repercussions on the generation and sharing of information, as well as the way marginalised communities are represented by machine learning models. Can we correct false assumptions and revise the general public’s understanding of language?
✍ See Denotation IO, a site dedicated to bringing computational linguistics, and in particular scientific notions of meaning, to general audiences.
Building healthy data-sharing processes
Data can be used or abused. In the face of the current debates around AI and copyright violations, how do we build community-based infrastructures for healthy data sharing? Can we integrate our company’s data processes with the Commons? How do we organize the work of dataset documentation?
✍ PeARS is an open-source project aimed at developing a distributed, multilingual search engine. It allows users to build, share and search community-based indices of Web content. The general PeARS structure can be leveraged for data creation, documentation and licensing. The project is currently funded by the EU’s ‘Next Generation Internet’ Programme.
Building green technology
- Our planet is dying, due to unregulated energy usage within various industries. How do we promote technological degrowth? How do we give visibility and value to efficient algorithms that do not pollute?
✍ We have a track-record of working on small and efficient NLP systems using both conventional and unconventional architectures. For a sample of peer-reviewed publications, see here, here, or here.
The future of work
- How do we organise work to make it meaningful, at all levels of research and development?
✍ First, the obvious: we believe in diversity. The core PeARS team is gender-balanced and has members from Asia, Europe and South America. Beyond this natural prerequisite, we are thinking of structures and processes that will stress the importance of each and everyone of our colleagues. The issues listed on this page highlight that there is no small task, and we want everybody to feel that their work is fully essential, not just to the company but to the world at large.
But most importantly…
We know that we cannot solve these problems alone, and we want to consider all the possible worlds where they might find solutions. So bring your world to us and let’s chat!