Positive Intelligent Technologies
Intelligent technologies are changing every aspect of our lives. If designed with care and forethought, they have the potential to solve many of today’s problems. But if designed sloppily, they could be dangerous and harmful to humanity.
I’m Steve Omohundro and my students, colleagues, and I have been analyzing and creating intelligent systems for decades. We invented a wide variety of machine learning and machine vision algorithms and built systems that learned to visually control robots, read lips, learn grammars, search for images, recognize and parse visual shapes, and perform many other fundamental tasks. About a decade ago I began to realize that advanced versions of this kind of system could be harmful if they didn’t properly incorporate human values. I developed an economic methodology for analyzing and developing these systems safely. The analysis fits into a broader program aimed at creating a unified understanding of biological and economic intelligence.
I’m thrilled that this work has been getting more media attention recently. I was invited to give a TEDx talk in Estonia and a keynote address in Oxford at the first conference on Impacts and Risks of Artificial General Intelligence. James Barrat’s excellent book “Our Final Invention” was partly based on my work. There have been news stories in the Boston Globe, USA Today, and the New Yorker and a much longer piece in the Daily Beast. I discussed the movie “Her” on NPR’s Science Friday, did an interview on the Palo Alto cable show Future Talk TV, and did an interview with Singularity 1 on 1. This website links to additional talks and papers about these topics. For a quick introduction, watch the 18 minute TEDx video from May 2012:
For more detail, watch this December 2012 talk at Oxford University:
this July 2011 talk at Monash University:
or this October 2007 talk at Stanford University:
For more depth, read the papers:
Or read the key insights at these links:
- Autonomous systems are imminent.
- Autonomous systems will be approximately rational.
- Rational agents have universal drives.
- Today’s infrastructure is vulnerable.
- We can build safe systems using mathematical proof.
- Some simple systems would be very harmful.
- The Safe-AI Scaffolding Strategy is a positive way forward.
Send email to Steve Omohundro: