21 February, 2019
I’ll leave you with our scariest story as you head off into your workday. Researchers at the King Abdullah University of Science and Technology, in Saudi Arabia, are teaching AI how to think abstractly, i.e. recreate the mind’s ability to deduce complex patterns. Humans have an innate ability to understand complex patterns of cause and effect, for example that an apple appears purple because it’s being shown under a blue light. It’s very difficult to teach a machine to disentangle multiple causal mechanisms like this because they use an observer-centric point of view instead of an object-centric point of view. They can’t yet put themselves in the place of the apple like humans can, i.e. they can’t consider what’s happening to the apple that’s causing it to appear purple. The researchers intend to use information probability to help machines generate inferences like this. They’ll use “algorithmic complexity to isolate several interacting programs, and then search for the set of programs that could generate the observations.” This could be a huge step in complementing machine learning with abstraction and identifying cause and effect, which current methods like deep learning cannot currently handle. If the thought of machines with human intuition doesn’t scare the pants off of you, maybe it’s time to recalibrate your own intuition.
"KAUST shall be a beacon for peace, hope and reconciliation, and shall serve the people of the Kingdom and the world."
King Abdullah bin Abdulaziz Al Saud, 1924 – 2015
Thuwal 23955-6900, Kingdom of Saudi Arabia