AI Correctness Is Not The Same As AI Ethics

As the capabilities of deep learning algorithms have improved exponentially over the last few years, there has been increasing awareness of the ethical considerations in deploying technology that can autonomously capture the underlying patterns of data and make decisions based upon it with precision and nuance unthinkable even a few years ago. This rapid developmental pace is enabling deep learning applications that push the boundaries of computational decision-making, from today's facial recognition algorithms to tomorrow's driverless cars to future autonomous "killer robots." At the same time, the impact of algorithmic bias is becoming more visible as AI systems are being deployed into ever more influential roles. As society reacts to AI adoption in sensitive areas like military, judicial and surveillance use, governments and companies have responded by arguing that so long as their algorithms perform as intended, they are ethical, suggesting there is considerable confusion about the difference between AI correctness and AI ethics. Like all computer code, deep learning algorithms and the data-driven models that power them are designed to perform specific tasks within specific operating constraints with a guaranteed accuracy rate.