?? Where can deep learning take us?

Hi, I’m Azeem Azhar. I convene Exponential View to help us understand how our societies and political economy will change under the force of rapidly accelerating technologies. This is a brief overview of some of the most important trends for the near future I observed last week. Full overview is in my Sunday newsletter.

?? Join me on Friday for a live 30-minute discussion with Bloomberg’s Editor-in-Chief, John Micklethwait, whose latest book looks into how the Covid-19 pandemic has exposed weaknesses in the West. We’ll stream live on Linkedin—RSVP to get a notification once we're live.

Earlier last week, I briefly wrote about China’s chip opportunity, and hosted a members’ discussion about the future of citizenship.

The near future

?? “I do believe deep learning is going to be able to do everything, but I think there are going to have to be quite a few conceptual breakthroughs. The human brain has about 100 trillion parameters, or synapses, whereas what we now call a really big model, GPT-3 for example, has 175 billion. It’s a thousand times smaller than the brain,” says Geoff Hinton, one of the pioneers of deep learning. In an interview with the brilliant Karen Hao, Hinton made the claim. Critics of this view include the CEO and founder of Robust.AI, Gary Marcus, with whom I spoke a year ago. My view: I do think that we’ll be surprised by what we’ll do with deep learning. The test will come when GPT-n has more parameters than the brain which may be only five-to-ten years away. Will a 100-trillion-parameter neural network outperform a 100-trillion-connection brain? I doubt it. Will it be an amazing piece of technology? Certainly. What about a 10-quadrillion-connection neural network? What will that be like? And what will we call the things it can do? 

?? Why do we need to end the data economy, and how do we do it? Last week Carissa Véliz joined me to answer these questions. I’ll leave you with one thought from Carissa and hope that you’ll listen to the whole episode: “People might think it’s very radical to call for the end of the data economy. But what is extreme is a business model that depends on the systematic violation of rights.”

?? Misinformation was rife during the US presidential election, particularly when it came to the already fraught issue of voting and ballots. A group of researchers took five hours and 120 people to fact check and trace the origin of a conspiracy theory called SharpieGate. See also, an exposé of how YouTube channels make money off fake and misleading livestreams of election results. YouTube has confirmed some channels will still be allowed on the site, although the owners will be demonetised. 

?? An interesting profile of the CEO of Volkswagen, Herbert Diess, on the electrification opportunity to which VW has committed more than €30bn. “The danger is that electric cars, which contain far fewer parts than combustion engine models, will be commoditised. It is a scenario Mr Diess hopes to combat by owning the valuable customer data generated by vehicles that are ever more automated and connected to the web.”

To receive this quick preview of the near future, follow me on Linkedin.

Today's bonus...

Joe Biden won the US election but I was very curious about the forces that underly some of our politics, especially in the last decades of accelerating change. Here are some interesting views.

Philosopher Michael Sandel talks to The Guardian about where he thinks the left has gone wrong – its pursuit and belief in meritocracy – ahead of his new book, The Tyranny of Merit. The interview is fascinating. Sandel says:

"On globalisation these parties said the choice was no longer between left and right, but between ‘open’ and ‘closed’. Open meant free flow of capital, goods and people across borders. To object in any way to that was to be closed-minded, prejudiced and hostile to cosmopolitan identities."