Democratising AI needs a radically different approach

Urvashi writes on alternative pathways to democratise Al in The Hindu. She expounds on how trying to replicate or compete with the Big Tech model by investing in expensive computational infrastructure, bigger models and more and more data collection is unlikely to be successful and will only reproduce existing harms.

Urvashi further shares how we are locking ourselves into a Big Tech-dominated Al future as long as we are locked into a deep learning imagination of Al. She goes on to recommend that we need to start by building a theory of change and then building curated data sets and models that correspond to that theory, and theory, not big data, must guide Al development. This further needs prioritisation of domain expertise and interdisciplinary collaboration.

Read the full piece here.