📊 EV’s Charts of the Week #100
What's changed since I started this newsletter in 2015?
Hi, I’m Azeem Azhar. This is a very special week for Exponential View: we’re celebrating the 100th edition of the Charts of the Week and the 400th edition of the Sunday Newsletter!
We’ll take a step back, and reflect on what has changed since the first edition of the newsletter, back in March 2015.
CHARTS OF EXPONENTIAL CHANGE
Artificial intelligence
In early 2015, I was occupied thinking about the potential of artificial intelligence — I had just sold a marketing analytics startup that had used lots of machine learning. But I had been so narrowly focused on what customers needed, I had lost track of the advances coming from groups like DeepMind. The newsletter became a way for me to share as I learnt. That’s how Exponential View started. In one of the earliest editions of the missive, I share this article, showing the state of AI in Q1 2015. The two most demonstrative examples were (1) automatic image annotation using a combination of convolutional and recurrent neural networks, and (2) DeepMind’s deep reinforcement learning for playing Atari games.
This was the era in which DeepMind was focusing on games to advance its systems, before applying its designs to other projects such as uncovering the structures of proteins with AlphaFold. I would discuss this with the co-founder and CEO Demis Hassabis few years later on the podcast.
Today, we use AI not only to generate highly accurate images from text and accelerate research by orders of magnitude, but also to how to improve AI itself. One indicator of this transformation is the explosion in the number of parameters in ML systems. Increasing complexity in these statistical models means that they can perform better.
And it’s not stopping there — the number of AI papers has been growing exponentially.
Moore’s Law
In the third edition of the newsletter, on April 5, 2015, I shared an article along with the chart below: Transistor Production Has Reached Astronomical Scales.
Since then, we’ve seen the demand for compute expand, largely driven by those machine learning models. ML workloads have grown with the complexity of models and demands on them. At the same time, we want to pack machine learning into our phones and edge devices.
I have been surprised how Nvidia has bested many startup chip challenges to hold onto its crown as the provider of choice for the demanding machine learner. It is worth seeing the speed with which, a decade into the GPU-for-AI era, they continue to make strides. Nearly 7-fold performance improvements in 2.5 years.
Biotechnology
In 2015, I spoke about how CRISPR was booming, and how it was laying the groundwork as a “novel, low-cost and precise gene-editing technique.” Since then, biotech has only continued its revolution, with two major trends at the forefront: miniaturisation and AI-enabled research. Novel research techniques have enabled large-scale experimentation and fast learning, bringing down the cost of genome sequencing.
Energy and climate
Are we really on the cusp of an energy revolution? Can the combination of solar energy, decentralised production, better storage and a move to electric vehicles tackle our largest sin? We seem to be very close to some epochal shifts which can bring an end to the auto-petrochemical hegemony. (Azeem in #EV15, June 2015)
When I started writing EV, Lazard reckoned the unsubsidised cost of solar electricity was around $60 per MWh. By 2021, they were touting a price of ~$30. Combined cycle gas hovered between $60 and $65 in the same period — and that was before this year’s excitable natural gas market.
Since then, solar has increasingly led the way into our new greener world.
The move to electric vehicles, too, has taken place around the world. Countries like Norway at the forefront, China and others following suit. There’s a growing realisation that batteries will be a defining technology of our transport future, with green hydrogen being more important in industrial processes.
Battery prices have significantly declined in recent years. Current questions are mostly about the extraction of the materials needed for the batteries, and how the mining for rare earth minerals can be done in an environmentally safe(r) and ethical way.
However, we continue to pump CO2 into the atmosphere at an increasing rate, despite the Covid slowdown. In 2015, global annual CO2 emissions were 35.5 Gt, this year they will end up around 36.6 Gt.
Your turn
Now I want to hear from you: think back to 2015 — what (exponential) projections and expectations that you held seven years ago have played out?
The power of networks that included people were just taking off then, but their exponential growth was expected. Yes we had LinkedIn and Twitter and Facebook, but none of them had started influencing our society so profoundly. Their combinatorial creativity was visible and alluring, but we didn’t expect that algorithmic decisions would yield such profoundly complex outcomes.
2015 seems an age ago. I remember feeling very positive about the trajectory of global development: beyond digital and energy leapfrogging there was the steady downward trend of people in poverty. Wind forward one pandemic and some politics later, and many of the gains been wiped out, the climate damage accelerated, and political will for solidarity eroded. I hope we can re-engage with a new, sober awareness of our interconnectedness to tackle global inequality again, and in a way that is adaptive to what the future will bring.