The Ethical Dangers of AI

The AI community has begun to take the downside risk of AI very seriously. I attended a Future of AI workshop in January of 2015 in Puerto Rico sponsored by the Future of Life Institute. The ethical consequences of AI were front and center. There are four key thrusts the AI community is focusing research on to get better outcomes with future AIs:

Verification – Research into methods of guaranteeing that the systems we build actually meet the specifications we set.

Validation – Research into ensuring that the specifications, even if met, do not result in unwanted behaviors and consequences.

Security – Research on building systems that are increasingly difficult to tamper with – internally or externally.

Control – Research to ensure that we can interrupt AI systems (even with other AIs) if and when  something goes wrong, and get them back on track.

These aren’t just philosophical or ethical considerations, they are system design issues. I think we’ll see a greater focus on these kinds of issues not just in AI, but in software generally as we develop systems with more power and complexity.

Will AIs ever be completely risk free? I don’t think so. Humans are not risk free! There is a predator/prey aspect to this in terms of malicious groups who choose to develop these technologies in harmful ways. However, the vast majority of people, including researchers and developers in AI, are not malicious. Most of the world’s intellect and energy will be spent on building society up, not tearing it down. In spite of this, we need to do a better job anticipating the potential consequences of our technologies, and being proactive about creating the outcomes that improve human health and the environment. That is a particular challenge with AI technology that can improve itself. Meeting this challenge will make it much more likely that we can succeed in reaching for the stars.

Ref: Interview: Neil Jacobstein Discusses Future of Jobs, Universal Basic Income and the Ethical Dangers of AI – SingularityHub