The Alignment Problem

Resumen del libro

Brian Christian is the author of the best-selling books The Most Human Human and Algorithms to Live By. He holds degrees in computer science, philosophy, and poetry, and has won several awards for his insightful books on the intersection of technology and humanity.

In 2015, Jacky Alciné discovered that Google Photos had mislabeled a selfie of him and his Black friend as “gorillas,” exposing the racial bias embedded in AI systems. Despite Google’s quick response and eventual removal of the “gorillas” category, the issue remains unresolved. This problem is rooted in a long history of biased technology, dating back to photography’s early days when cameras and film were calibrated to favor white skin tones, excluding accurate representation of darker skin. Even as Kodak improved film in the 1970s, the damage of decades of exclusion persisted. Fast-forward to the 2010s, Joy Buolamwini encountered similar biases in facial recognition technology, where datasets overwhelmingly favored white males, rendering AI ineffective at identifying Black women. Her findings prompted IBM to refine its algorithms, but the incident highlighted a deeper issue: AI systems reflect the biases of their training data, shaped by historical and societal inequities. The next section will explore why these biases persist in modern AI and the ongoing efforts to address them.

Para leer el resto del libro puedes descargar Bitely