3️⃣ AI Ethics with Lauren
It's about time the AI industry started disrupting itself!
Data centric AI is the push we need to focus on the quality of data. Its technical progress is evidence that technical and ethical benefits often overlap. Too long have we milked success from accumulating data for the sake of “more=better” for developing models, molding them to be whatever is best for what is fed in. The quality of data is the source of various ethical concerns, from privacy rights (ex. what personal information should be collected and how it is used/treated) to encoded bias (what story the data tells and how it’s fortified through the model). Data centric AI repositions our focus, allowing us to more easily get in front of problems before they become entrenched in the model. Now, we have an added performance incentive to be more conscious of the impact that data has on models, and the impact those models have on everything they touch.
Accessibility is a valuable outcome of data centric AI. It bolsters the abilities of knowledge experts to implement their mastery directly by creating labeling functions themselves. This shift decreases the likelihood of error due to gaps in understanding the data and better discernment in how data should be treated. It also results in a more equitable AI landscape, as the scope of who holds power is broadened and subsequently diversified.
This renewed focus on the quality of data could be considered an expansion in ethical considerations. Implementing more steps doesn’t have to result in a slowdown or a performance decrease – as stated, it bolsters both the technical and the ethical by creating more points at which we can identify problems and generate solutions. We won’t know what is broken if we can’t see it! Data centric AI allows us to more easily get in front of problems before they are algorithmically entrenched in the model, and encourages a wider variety of experiences to shape those outcomes for the better.
- AI Ethics segment by Lauren Keegan