🔮 Moore's Law; ML & work; solar panels in space; ants ++ #391
As machine-learning infused tools make their way through the workplace, we’re learning more about when they help.
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.
Earlier this week, I discussed the White House’s new crypto framework with one of the leading experts in the field, EV member Kevin Werbach. Listen to our conversation👇
The near future
♟️ Farewell, Moore’s Law (or not)
Jensen Huang, the boss of Nvidia, has declared that “Moore’s law is dead” and “[a] 12-inch wafer is a lot more expensive today. The idea that the chip is going to go down in price is a story of the past.”
But what does this actually mean?
Moore’s Law is a slippery idea. In practice, it speaks to the ability to cram more transistors onto a single wafer. More transistors per square inch means more computing power. It was this miniaturisation that drove computing improvements. Moore’s Law became a “social fact” which propelled the industry forward, drove costs down, and made computers ubiquitous.
The idea that Moore’s Law might die has been around for a couple of decades. Miniaturisation only goes so far: heat problems and quanta effects abound. But this isn’t what Huang seemed to refer to.
The issue is that Moore’s Law is a slippery concept that is colloquially used for the idea that the price/performance of computing will decline at very rapid rates. (So even if “chips” double in price, if their performance quadruples, price/performance has improved.)
I think Huang’s assertion is not quite correct. Individual chips may get more expensive. The Cerebras WSE1 (a dedicated AI chip) cost a couple of million bucks — but it packs a huge punch. Its cost per compute cycle is much lower than say that of Intel’s 1978 chip, the 8086 ($400 in 2022 terms). Modern chips (even if they are more expensive on a per unit basis) deliver computing more and more cheaply.
Price/performance improvements may not be the miniaturisation that was the keystone of Moore’s Law. They have come from architectural and packaging changes (like multi-core processors) or moving from 2-d to 3-d layered chips (as Graphcore has done with its Bow chip.) A large improvement has come from moving from generalised chips to specialist chips optimised for machine learning workloads (think of Google TPUs). There are even more narrowly focused systems. For example, hedge fund, DE Shaw has a computer called “Anton” optimised for accelerating molecular dynamics 180-fold over current high-performing computing systems.
So while the cost of Nvidia’s graphics units might increase in price year-on-year, the overall trend of computing getting cheaper and cheaper will be unaffected. Bring it on.
In addition, for a good scientific review of moving past Moore’s Law, I recommend John Shalf’s 2020 paper “The Future of Computing Beyond Moore’s Law.” See also, China’s growing semiconductor market slumped 25% due to sluggish demand and lockdowns.
🦾💪 When humans beat machines
As machine-learning infused tools make their way through the workplace, we’re learning more about when they help (and when they don’t). A new paper identifies this gap.
ML tools are beneficial in augmenting distinctions and certain classes of decision-making but adding ML to the mix worsens productivity with novel problems. It is the least capable workers who benefit the most from an ML-based teammate, top quintile workers see little improvement. But on novel or rare problems, humans do better on their own. There are a number of ramifications of this. The tools can level up newer or weaker workers (perhaps allowing outsourced workers a chance to compete on a level playing field). The relative importance of the top performing workers increases, as they have the experience set and judgement to handle novel tasks.
See also: Shipments of industrial robots in China grew 45% in 2021, representing about half of the global market. While the number remains small — 243,000 compared to the Chinese manufacturing workforce of 83 million — this growth may help bridge the gap in the local labour market which is suffering from China’s demographic reverse. Single robots may be more productive than teams of people. In one case, making headphone cases, a team of four people could manage 650 cases an hour while a single robot arm weighed in at 800. Great long-ish read in the WSJ.
See also, this study on hybrid work showed that hybrid workers have greater satisfaction and a lower sense of isolation than either stay-at-home or always-in-the-office workers.
Weekly commentary: The negative cost of Net Zero
There is a widely-held view that the cost of getting to Net Zero is going to be high. We are told that the sums of money required for the new kit (solar panels, batteries, wind turbines, small modular reactors, high-voltage direct current interconnects, microgrids and the like) will be larger than a typical oligarch’s Loro Piano budget. Certainly, running into the trillions of dollars. But is this really a large number in a sense of being either unaffordable or poor value for money?
I look into this in my Weekly Commentary which will go out to paying members on Monday.
If you’re not a member yet, upgrade your account to support our work and receive tomorrow’s commentary in full.
Dept of our climate future
In every Sunday edition, we track key metrics that tell us a little about our shared climate future. Our member, Marshall Kirkpatrick, takes the time to curate a view of our current climate status in this segment every week, and you can read Marshall’s curation below. Here’s Marshall:
In addition to billions of dollars of new loans and investments aimed to bolster renewables, and legal rulings against multi-billion dollar fossil fuel projects, here are some of the most interesting new developments in climate this week.
Renewables milestone: Wind and solar power provided more than 10% of the entire world’s power generation for the first time last year, Bloomberg NEF reported this week. 50% of the new power generation added last year was from solar, 14% from wind. The bad news is that coal grew faster last year, too. Just 89.5% of the mix left to go for renewables! Last month we shared a meta study by 15 academic institutions demonstrating that it is now effectively a global scientific consensus that 100% of the world’s energy needs could be filled by renewable sources by 2050.
Indigenous co-management for climate: The US Federal Government announced new guidelines last week for co-management with indigenous tribes of federal lands across the country, covering more than 70% (!) of the land under federal management. The guidelines will apply to three giant agencies: the Bureau of Land Management, National Park Service, and Fish and Wildlife Service (links to each agency’s order in PDF). Repeating for emphasis, because this really is amazing, more than 70% of the federal land in the US (by my math) now falls under orders with language like this from the BLM, which is directed to “engage Tribes in meaningful consultation at the earliest phases of planning and decision-making in order to provide an opportunity for Tribes to shape the direction of the BLM’s land management activities. Upon request by a Tribe, the BLM will meet and consult regarding co-stewardship opportunities and evaluate proposals submitted by Tribes, including proposals to build both Tribal and Federal capacity to carry out the SO.” Wow! Globally, Indigenous Peoples represent 6.2% of the world’s population, hold tenure over 25% of the world’s land surface, steward 80% of the world’s biodiversity, and manage at least 24% of the total carbon stored above ground in the world’s tropical forests. The IPCC’s February report cites indigenous land tenure 58 times as a key lever for limiting climate change.
Solar panels in space: China’s LONGi Green Energy, called the world’s largest solar company, has announced that it is going to launch solar panels into space. The experimental system will capture sunlight high above the ground and convert it into microwave beams. Then it will shoot those beams through the air to a receiver station on the ground, where they can be converted back to electricity. While such a system is intended to offer continuous access to the sun’s rays, many people have called it more trouble than it’s worth. Dutch renewables expert Auke Hoekstra posted a thoughtful Twitter thread in defense of solar panels in space almost four years ago. It’s certainly a very centralised solution.
Breaking the loss & damage taboo: Denmark has announced that it will break ranks with international leaders who refuse to pay money to poor countries and former colonies for the impact of climate change. This week the country announced a symbolic US $13M pledge to aid countries already suffering far more than they’ve contributed to climate change. Historically, global powers have been unwilling to give a little, lest they be expected to pay up the full sums that might be expected. Denmark built commercial colonial power with sugar and slaves, across Africa, Asia, Europe, and North America, for more than 400 years, until the 1950’s. Announcing the new climate damage funds, Denmark’s current development minister Flemming Møller Mortensen said: “I saw for myself in Bangladesh this spring that the consequences of climate change need increased focus… It is grossly unfair that the world’s poorest should suffer the most from the consequences of climate change, to which they have contributed the least.” Imagine what it would mean for fairness to be a priority in all such matters.
Let’s build on these examples of bravery!
Short morsels to appear smart while Putin tracks you
🕸️ Fascinating details on the surveillance and monitoring efforts by Russia’s internet regulator. See also, the Chinese government accuses the NSA of infiltrating core internet infrastructure.
🚨 Xpeng’s passenger saloon now has the regulators’ OK to offer fully-autonomous driving within Chinese cities.
🎳 The Helium crypto community voted to move from its native blockchain to Solana.
🎭 Getty bans AI-generated artwork, an attempt to avoid future legal challenges.
🌶️ The Danes want to develop the first quantum computer for life sciences research.
🐜 The Earth’s ants (probably) weigh more than birds and mammals combined.
🙃 Youtube’s “dislike” and “not interested” buttons don’t really do anything, a study found.
Hello, hello. It’s my birthday next week — and it’s a big one! 😎 So as my birthday present, I’d love you to do one of two things. If you are not a premium subscriber, consider subscribing for a year.
And then (especially if you are a paying subscriber), please share the newsletter with a few friends encouraging them to sign up!
What you’re up to – notes from EV readers
George Darrah published a piece exploring the connective tissue between nature and biotechnology.
Kevin Bankston is hosting a virtual panel on September 29th about the intersection between science fiction and speculative design, with an appearance from cyberpunk sci-fi legend Bruce Sterling.
Steven Ritchie joined The Knowledge Society as a director for their virtual program, training young people looking to solve the world’s biggest problems with emerging tech.
Aimun Jamjoom and his co-authors found a liability dilemma with autonomous surgical robotic systems in an international survey of public opinion published in Frontiers in Surgery.
To share your projects and updates, fill out your details here. Because of space constraints, we prioritise updates from paying members and startups I have invested in. (You can become the former by subscribing, if you have not already, and the latter by getting an intro to me via a trusted contact.)