One of the themes explored in my discussion with Yuval Harari was the move towards algorithmic management of the world. How, increasingly, algorithms make automated decisions in our everyday lives and in the deep bowels of business and the economy.
📊 Pew Research recently interviewed 1,302 experts on the impacts of algorithms on society in this rather excellent survey
. The conclusions: algorithms will go everywhere and that’s mostly a good thing. But we need to be wary about losing our own decision-making prowess, guard against bias, manage filter bubbles and ensure we have oversight of those algorithms.
My take is that correctly designed, algorithmic decision-making will continue to be an incredible boon to human society. However, such algorithms needs to be holistically designed, which means understanding the wider context of a decision including any externalities arising from it. If ‘algorithmic decisions’ are narrowly focused on efficiency for the operator of the algorithm, unintended consequences are likely to abound.
In November 2012, it took four employees 25 hours to compile and post just a fraction of the election results manually. In November 2016, Heliograf created more than 500 articles, with little human intervention, that drew more than 500,000 clicks.
This is a far cry from 1995 when my then boss at The Guardian clamped together a bunch of scripts to autogenerate weather reports.
DeepMind’s Pathnet has got AI researchers buzzing because it seems to be a precursor to the kind of architecture that could support artificial general intelligence (or AGI). Pathnet combines several hot areas of AI research in a single architecture: meta-learning, reinforcement learning, adversarial & cooperative learning and transfer learning. Carlos Perez has a reasonably accessible overview of it.
Adversarial examples are a datum that can force a machine learning system to make a mistake. It’s an intriguing area of cyber-security, especially as more decision making switches to algorithms. OpenAI has a nice paper on the subject