Can We Build AI Without Losing Control Over It?

Great TED talk, by the brilliant Sam Harris, which serves as solid a introduction to understand the significance of AI and the enormous challenge we face regarding how to create safe conditions for AI as we will most likely only have one chance of doing so.

Sam Harris is a neuroscientist, philosopher, New York Times best-selling author, host of the Making Sense podcast, and creator of the Waking Up app. One of Sam Harris’ best books is called “Lying”, for the key points visit this post “Why lying is so pervasive in society and what to do about it“. See also Sam Harris on Lex Fridman’s podcast.

Key Points

Two possible scenarios with regards to AI

1. We stop making progress in building intelligent machines.
   
   This can only happen is something destroys civilization as we know it, such as:
   • A full scale nuclear war.
   • A global pandemic.
   • An asteroid impact.
   • (A gamma ray from a collapsing neutron star).
     
     
2. We continue to improve our intelligent machines, year after year.
   
   At a certain point we will build machines that are smarter than we are, upon which they will start improving themselves. The we risk an “intelligence explosion“.
   
   If you believe that we will not be able to build super-intelligent machines, then you must find something wrong with the following assumptions:
   
   • Intelligence is the product of information processing (we already have narrow AI performing at a superhuman level).
     
   • We will continue to improve our intelligent machines.
     
   • We are not near the summit of possible intelligence.

Electronic circuits function about 1 million times faster than biochemical ones. If our intelligent machines was at the level of intelligence as we are, then by the virtue of speed alone, after a week in operation it would produce the equivalent of 20 000 years of human intelligence work. This would continue every week. How can we even fathom what such intelligence would come up with, much less constrain a mind making this kind of progress?

Building a super-intelligent general purpose AI is a winner-takes-it-all scenario. Thus, wars could be potentially waged just based on merely the rumors that someone has built or is about to build such a machine. What we need is perhaps a Manhattan Project to understand how to avoid an arms race and to build the AI in a way that is aligned with our interests.

Sam Harris’ TED talk “Can we build AI without losing control over it?“