Beyond the Valley by Ramesh Srinivasan

Author:Ramesh Srinivasan
Language: eng
Format: epub
Publisher: The MIT Press

16    Discrimination Technologies

In our conversations about automation and its potential effects on the future of work, we should all be asking questions about how work will be chosen in an economy increasingly driven by algorithmic and automated systems. But let’s probe deeper and ask an important ethical question: How do we prevent automated and algorithmic systems from targeting the most vulnerable members of our society, and how do we continue to ensure that these systems remain equitable?

Technologies are not neutral but socially constructed, created by and for particular people with particular worldviews, agendas, values, and priorities. Engineers, designers, and coders build into each system a projection of their own thoughts, preferences, and assumptions about the world. There is nothing inherently wrong with that, except that as these technologies become increasingly central to the lives of everyone on earth, the mental habits and cultural biases of a tiny group will be disproportionately influential and powerful. Sometimes it makes sense to allow a small group of specialists to have this level of influence or power—but usually we do so by electing them. Yet now more than ever, hidden, private, and corporate-employed designers have all the power over even the most intimate aspects of our lives.

It might seem obvious that content on the internet, as well as on the most popular websites, largely comes from the Western world (with the exception of the Chinese internet).1 But less obvious is the extent to which biases that shape the content we see come from the “white Western male” perspective. I have discussed gendered and racial bias earlier, especially in chapter 2. I’ll put it more bluntly here: artificial intelligence (AI) systems have a “white guy problem.”2

We have already seen AI learning systems make mistakes that follow the classic lines of discrimination: Google Photo’s image recognition algorithm misclassified black people as gorillas.3 A “beautifying algorithm” developed in Russia, called FaceApp, turned Barack Obama’s face whiter (and younger).4 A Microsoft-developed chatbot called TayTweets, which was designed to mimic and converse with teen users on Twitter, quickly developed racist, sexist, homophobic, and xenophobic tendencies.5 Within hours it began to echo positions of President Trump while tweeting #MAGA.

Algorithmic mistakes and biases are real. We all have implicit biases no matter who we are: after all, much of them are simply about what we leave out, forget, or assume. But algorithmic discrimination takes the problem to a new level. It can blow up hate, bigotry, sexism, and other forms of bias, which disproportionally target already-vulnerable peoples, to inflammatory extremes. Similarly, the very human temptation to treat computational recommendations as objectively “correct” is just another mechanism by which any error an AI system makes can be duplicated (and further spread) with the potential to create disaster.

Let’s consider perhaps the most troubling facial recognition software of them all: Amazon’s Rekognition. The system misidentified 28 members of the US Congress as criminals when the American Civil Liberties Union (ACLU) of Northern California matched images of all 535 members to a publicly available database of 25,000 mug shots.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.