Generative AI in the era of 'alternative facts'
|
MIT Open Publishing Services
Research
The idea of equal opportunity enjoys wide acceptance because of the freedom opportunities provide us to shape our lives. Many disagree deeply, however, about the meaning of equal opportunity, especially in algorithmic decision-making. A new theory of equal opportunity adopts a structural approach, describing how decisions can operate as bottlenecks or narrow places in the structure of opportunities. This viewpoint on discrimination highlights fundamental problems with equal opportunity and its achievement through formal fairness interventions, and instead advocates for a more pluralistic approach that prioritises opening up more opportunities for more people. We extend this theory of bottlenecks to data-driven decision-making, adapting it to cetre concerns about the extent to which algorithms can create severe bottlenecks in the opportunity structure. We recommend algorithmic pluralism: the prioritisation of alleviating severity in systems of algorithmic decision-making. Drawing on examples from education, healthcare, and criminal justice, we show how this structural approach helps reframe debates about equal opportunity in system design and regulation, and how algorithmic pluralism could help expand opportunities in a more positive-sum way.
|
MIT Open Publishing Services
|
Harvard Business Review Press
|
Arxiv
|
Arxiv
|
bioRxiv
|
Nature
|
Arxiv
|
Pancreas
|
Science
|
Cell Systems
|
Arxiv
|
Radiological Society of North America
|
Nature
|
Arxiv
|
Science Direct
|
PNAS
|
Nature
|
Arxiv
|
Journal of Clinical Oncology
|
Proceedings of Machine Learning Research
|
Dynamic Ideas
|
Science
|
Little, Brown and Company
|
Arxiv
|
Dynamic Ideas
|
Advances in Neural Information Processing Systems
|
International Journal of Computer Vision