🔮 Discovery; foundational AI; learning & cost of energy production; nuclear fusion++ #376
Hi, I’m Azeem Azhar. I convene Exponential View every week to help us understand how our societies and political economy will change under the force of rapidly accelerating technologies. Today’s edition is quite rich, so find a comfy spot and settle in.
If you are reasonably new to this wondermissive, then a great way to get the baseline of my thesis is to buy my book Exponential (called The Exponential Age in America.) If you like audiobooks, give it a shot. I’m the narrator! 😄
The near future
🧶 The burden of knowledge
Harder, smaller, slower, strugglier: this is the state of scientific research, suggests Matt Clancy. It is a deeply thoughtful piece. Perhaps one anchor weighing discovery down is the burden of knowledge:
If you need more and more knowledge to make a discovery of a given size, then you can probably expect discoveries of a given size to require more time or manpower to bring about.
My view is that there may be structural reasons (in current academia) that are reducing the incentives for creative risks for scientists. This could also be tied to the commercialisation of much research, especially in computer science and biological domains. Another measure, for example, is that the time to win a Nobel prize may reflect sociological rather than scientific phenomena. It may also be that the growing complexity of research, often involving distributed teams, has had a drag on creative productivity. It’s only in the past decade that it seems the distributed researchers are able to be as, if not more, innovative than collocated teams. And that phenomenon seems to be spreading to firms, according to McKinsey.
🧱 The building blocks of AI
The Economist has a solid summary of foundational AI models. This is the notion that high-performance, complex neural network models (of the type made by Google or OpenAI) will be the baseline building blocks of many future AI applications. There are risks, of course: of the concentration of power, challenges of inclusion and problems of access. My take is that some models seem so fundamental that we need to think about how they operate in the broadest benefit (and that “how” is a political, rather than a commercial question.) Others, such as industry-focused models proposed by my buddy Fernando Lucini, are less challenging. The deeper briefing is worth reading. (Sam Arbesman explores alternative histories by throwing hypotheticals at GPT-3. What does GPT-3 predict would happen if JFK hadn’t been assassinated?)
💬 Build it and they won’t come
Usage on some of the top metaverse platforms has collapsed to as little as 1,000 users a week. Still smacks of a problem looking for a solution (at least today), and the question is whether the groups running these entities have a way (short of handouts) of getting product:market fit. The technology is also proving to be troublesome. Meta has delayed the launch of augmented reality goggles by several years. Apple did not announce anything remotely VR-goggle-ish at this week’s World Wide Developer Conference.
I’m willing to be wrong about the metaverse, but I’m still struggling to see a general use case for it with the current state of the hardware and software stack.
🍎 An everything company
So much to say about Apple’s World Wide Developer Conference. The absolutely remarkable silicon on display in the M2, proving that you don’t need to rely entirely on Moore’s Law to improve processor performance exponentially. Architecture and specialisation helps, too. I’m excited about the arrival of Passkey (Apple’s implementation of FIDO) and the end of the password. Apple has gotten into “Buy now, pay later”, relying on its own internal credit chops. This proves it can be an everything company, based on its very strong relationship and understanding of its customer. Elsewhere, USB-C will become mandatory for all phones sold in the EU from 2024 - no more Apple Lightning cables. The USB-C standard seems, in any case, to be technically stronger than Apple’s flavour.
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: “Learning by doing is a theme that runs across all our stories of pragmatic optimism in climate this week. Most importantly, it’s not about learning individually – but collectively. Let’s do, and learn, and do again better, together.” (Azeem: I wrote an op-ed for the Financial Times on learning by doing this week. Check the end note.)
#Landback. In what could be the largest single land buy-back purchase in American Indian history, the Bois Forte Band of Chippewa has regained title to more than 28,000 acres of what’s now called Minnesota. The land was purchased from a private landowner by The Conservation Fund and the Indian Land Tenure Foundation, leaders of the National Indian Carbon Coalition. Many thousands of acres of land are being given and sold back to Indigenous tribes by federal, state, and local governments around the US but most are in increments of a few hundred or thousand acres at a time. Indigenous land rights are cited 58 times in the latest IPCC report as a key lever for mitigating climate change. I recommend this podcast interview with Dr. Kyle Whyte for a powerful explanation of Indigenous climate science and analysis by the IPCC. What does returning land look like? It can take many forms. Meeting-tolerant readers may delight as I did in the understated beauty of this ten minute discussion last summer between the government of Lincoln County, Oregon (pop. 50K) and the local confederated Siletz tribes about the donation of 90 acres. Later in that same meeting, county leaders thanked the Siletz for what the Federal Emergency Management Agency called the first case they’d ever seen of a tribe providing shelter to non-Native populations displaced by climate-worsened wildfires. May such mutual aid proliferate world-wide.
ICE ban advances in Europe: New internal combustion engine (ICE) vehicles will no longer be legal to sell after 2035 in Europe if regulation advancing through the EU is agreed to by member states. Substantial amendments that would have softened the ban were defeated this week. Bloomberg writes, “If an agreement is reached, it effectively spells the end of the combustion-engine car in Europe, marking a radical overhaul of a form of transport that has been dominant for more than a century. It would also be a crucial win for the bloc’s climate agenda, with the transportation sector proving one of the hardest to decarbonise.” The publication Elektrek, though, says that the 2035 timeline, the continued sale of used ICE cars, and the remaining option to export ICE vehicles manufactured in Europe make the scheme too little, too late. British architect and fuel-watcher Katy Duke updated her tracking of fuel costs per mile this week and found that it now costs 12.9X more per mile to gas up an ICE vehicle at the pump than it does to charge an electric vehicle overnight.
Playing defence: The Biden administration has announced that it will use the Defense Production Act (DPA) to accelerate the development of clean energy technology, including solar technology, heat pumps, insulation, green hydrogen, and grid components like transformers. The DPA was famous for transforming the capacity of the US auto industry into deployment for military production in times of need. The US defence industry, with its 750 military bases around the world, is both a leading contributor to the climate crisis and is as clear-eyed as the insurance industry about the threat of climate change.
Learning faster: Researchers at the Lawrence Berkeley National Laboratory in the US have quantified the impact of learning on the huge declines of levelised cost of renewable energy. That’s the cost of energy production understood in terms of the whole lifetime of its plant production. Learning-by-doing drives down the cost of energy production and financing, at a rate for wind of 15% per epoch (if I’m understanding this correctly, with epochs delineated by big jumps in energy output and R&D investment) and 24% for solar. There have been 3 distinct epochs in large scale wind production (since 1982) and 2 in solar (since 2007). The impact of learning on price has been growing as the industries have matured: over the most recent epoch it was up to 40% for wind and 45% for solar. This research cited Wright’s Law, or the law of the “learning curve” in technology development, which is said to be even more predictive than Moore’s Law. As guest author Ramez Naam wrote in Exponential View two years ago regarding solar power production: “Wright’s Law states that, for most technologies, every doubling of cumulative scale of production will lead to a fixed percentage decline in cost of the technology. This happens through learning-by-doing, a mixture of innovation that improves the technology itself and innovation that reduces the amount of labor, time, energy, and raw materials needed to produce the technology.”
Perhaps analogous processes may be found in the work each of us is doing.
Short morsels to appear smart while experimenting with Dall-e
🖼️ A new approach that builds AI for better image classification: an integrated end-to-end photonic deep neural network.
📈 Why building a nuclear power station is getting increasingly expensive.
☢️ Getting closer to nuclear fusion: how far we’ve come using the Lawson Criterion.
🚴 E-bikers tend to exercise more than regular cyclists, and tend to be older as well.
☀️ A bottom-up solar revolution in Puerto Rico.
💻 Get a (mini) taste of Dall-E.
🥧 Google Cloud managed to calculate pi to 100 trillion digits, 37.2 trillion (almost 40%!) more than the previous record. That’s about twice as many black holes there are in the universe. Or 10 million times the number of sheep in Wales.
🌾 Food shortages are forcing restaurants to rethink food.
👍 Japanese scientists found a way to cover a robot finger with living skin.
I made some recommendations in the Financial Times about how governments can employ new technologies to create markets and deliver their benefits more quickly. Do have a read.
A reminder that this week you’ll be able to catch me at a few events:
Talking at the Kite Festival in Oxfordshire on Sunday.
Speaking at CogX on Monday. I’ll be around on the other two days too. CogX looks great as usual. As a wondermissive aficionado you can attend CogX with a 100% discount. Visit this link and use the code AzeemSFP22 . (I think SFP stands for starfighter pilot.)
Thursday, I’m speaking at Founder’s Forum with Stuart Russell, Nigel Toon and Lila Ibrahim on brain-scale AI. That is invite only but if you are going to be there, say hi!
Have a great week!
P.S. Are you engaged on crypto regulation issues? Let us know by filling out this form, we’d like to loop you into our member insights network.
💭 Member comment: Luca Taroni is a member of our Exponential Do community. He shared the below take on Numerai with us in the Exponential Do Slack, and I thought it too good not to be shared more widely. Enjoy.
Hi, Luca Taroni here. Given my background in algorithmic trading and inspired by the announcement of the Numerai podcast, I’ve decided to dive deeper into Numerai. Here is a recap (the model, the dataset, market neutrality, users and incentives) and why I think this is a great project from a trader’s perspective, but not so great for users.
An ensemble-learning based meta-model
The core of the project is the proprietary meta-model, the notion building upon ensemble learning, which is basically machine learning with multiple models. In its simplest form, we can use multiple instances of the same algorithm with different random seeds or data slices, and average them to achieve more robust and less overfitted results. In this case we only require each model to have a predictive ability higher than 50% (randomness). As meta-models get more complex, we can use different kinds of algorithms and we no longer strictly need that each contributing model does better than randomness. We do need each model to positively contribute to the overall meta-model though.
That’s exactly what Numerai is trying to do: They are more and more focused on rewarding the marginal contribution of each model to the meta-model. Let's say we have a crappy model which is outstanding only in rare, extreme volatility market conditions. On average this model runs at a heavy loss, but if the meta-model is smart enough to activate it only in the appropriate market conditions, we have a positive marginal contribution. Ensemble learning thrives on marginally successful models - they have to be diverse. Diversity is core. You want your contributing models to be orthogonal or as orthogonal as they can be.
Numerai uses this approach to manage a fund of financial time series across countries, markets and sectors. Their track record shows success and is improving over time.
The rest of Luca’s comment is available here, and you’ll read:
Why Numerai use a market-neutral approach
The implications of using blockchain tokens as a basis for building a meta-model
Why Numerai is closer to a Web2 project than it is to Web3
What you’re up to – notes from EV readers
Vass Bednar, executive director of the Master of Public Policy in Digital Society at McMaster University, received a President’s Award for Outstanding Service.
Rafael Kaufmann’s startup Digital Gaia is live on Gitcoin’s June funding round. Digital Gaia is on the mission to revolutionise impact integrity by channelling the money and energy going to climate into projects that have a measurable, lasting, positive impact on the world.
Allison Agre and her team at Amogy demonstrated the first Ammonia-powered, zero-emissions tractor.
Adam Oskwarek at Zopeful is launching a free 3-day email course about Carbon Removal - sign up now to receive it next week.
Ken Pucker wrote two complementary articles about HBSCs head of Responsible Investment Stuart Kirk’s controversial speech: “Kirk’s inconvenient truth about ESG investing” and “ESG: Trouble and Truthtelling”.
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.)
How happy are you with today's wondermissive?
Unhappy | Meh | It's good | Great | Extremely happy, will forward