Warning: this post contains a major spoiler for Three Body Problem and minor plot point for Dark Forest, book 2 of the series

Last year (2021), I spent the summer working as a Tech-to-Market fellow at the Department of Energy’s ARPA-E helping develop and design a quantum computing program. I read various technical papers and reports1, conducted various stakeholder and industry insight interviews, and presented my market findings and technology roadmap hypotheses to the program director.

(This isn’t a post about quantum computing, but the rough picture is: quantum computing requires quantum entangling multiple objects, where each object is referred to as a qubit. This is really hard to do because it requires bringing objects really close together while preventing other stuff from interfering with them (noise, heat, light etc). There are a variety of methods that people use to accomplish this and the differences generally revolving around what object you try to entangle and how you keep the object “still” enough so they don’t “move” around or run into interference)

During that summer, one interview I conducted with a former national lab researcher who had since jumped to AWS in particular stuck with me. His observation was (and I’m paraphrasing/interpolating) that the US quantum computing industry tended to adopt a superconducting approach to building qubits as a result of strong basic research funding for superconductors during the past several decades in the US, driven by its use in MRI applications and the afterglow of the 1972 Nobel Prize in Superconductivity awarded to 3 US researchers. There are a few exceptions for other hardware approaches in North America e.g. PsiQuantum for photonics, D-Wave for adiabatic, Ion-Q for trapped ion, but the bulk of the investment is in superconducting.

But on the other side of the Atlantic, the balance of European startups focused on photonic approaches and his insight was that this preference in hardware technology occured because in the 80’s, European funding agencies favored basic research investments in photonics, while their US counterparts preferred superconductors.

This observation matched with something I had noticed anecdotally, from AI accelerator research as well, with a strong group of European startups developing photonic chips2.

More concretely, I modified this admittedly European-heavy map to circle all the European startups and research groups. Notice how most of the quantum computing heavweights in the US are using superconducting e.g. IBM, Google, while every listed photonics research group is in Europe. Another figure here also shows the geographical distribution of quantum computing hardware startups (notice EMEA has numerical superiority only in photonics).

Around this same time last year, I was also working my way through Liu Cixin’s Three Body Problem series, a sci-fi trilogy meditating on the nature and limits of scientific progress mixed in with novel takes on time travel and aliens. In the second book, as humanity faces an impending doomsday space battle with limited time to develop new technology, one deeply thoughtful military officer has the following conversation regarding space propulsion technology and the path-critical technology choice they face.

Doctor, you know about the two current research forks: media-
propelled spacecraft and non-media radiation-drive spacecraft. Two opposing
factions have formed around these two directions of research: the aerospace
faction advocates research into media-propelled spacecraft, while the space
force is pushing radiation-drive spacecraft. The projects will consume
enormous resources, and if the two directions can’t progress simultaneously
on equal footing, then one direction must take the mainstream.
…
It’s the old guard left over from the chemical rocket era that’s pushing for media
spacecraft, but forces from other disciplines have entered the sector. Take the
people on our fusion system. They’re mostly pushing for radiation spacecraft.
These two forces are evenly matched, and all that’s needed is three or four
people in key positions to break the equilibrium. Their opinions will decide the
ultimate course of action. But those three or four key people are, I’m afraid, all
part of the old guard.”
“This is the most critical decision in the entire master strategy. If it’s a
misstep, the space fleet will be built atop a mistaken foundation, and we might
waste a century or two. And by that time, I’m afraid there will be no way to
change direction.”
Dark Forest, Liu Cixin, pg 241-242

Shortly after, that character goes on to take drastic action to ensure the right technology path is chosen.

Over the past year, I’ve started to reframe a lot of observations about the technology we have today in terms of critical pathing. Some more examples of technology pathing in the real world (delayed, incorrect, or hypothesized):

Record federal investment in solar R&D during the 1973 oil embargo was sharply curtailed once Reagan was elected and only really recovered in the 2010s. 50 years later, solar costs in 2017 pretty much matched the cost curve prediction made in 1974. The solarpunk future we could have had

Japanese automakers obsession with hydrogen-powered cars for the past decade has made them laggards in electric vehicle technology and will reshape the balance of the global auto industry in the next decade (unless they somehow catch up, which would be a whole story by itself)
I’m mildly bullish on nuclear fusion within 20 years, but maybe we could have it sooner if we actually funded fusion R&D like we said we would in 1976?

File:U.S. historical fusion budget vs. 1976 ERDA plan.png

The AI models we have today are not globally optimal in some universal sense, but are locally optimal and contingent on the computing hardware we have available to us today. Though they did not realize it at the time, the original CUDA and GPU developers and designers have invisibly crafted the models available to use today as state-of-the-art. The development of new AI accelerators and the cambrian explosion of boutique semiconductor chips will lead to greater diversity of AI models. In particular, we’ll see greater discrepancy between US-China AI model design as China loses access to the same common hardware substrates like GPU’s and is forced to design their own chips. Already, Tsinhua researchers are training large language models on chinese-designed chips.
The history of the mRNA covid vaccine is filled with human contingencies and decisions, including a DARPA program manager who funded Moderna in 2013 and helped them, and the technology, stay afloat
Henry Ford and Thomas Edison’s relationship and how they almost commercialized a battery-powered Model T in 1914
There’s probably a long blog post or book on why France gets 70% of their power from nuclear power plants and the US doesn’t (something something three-mile island)
90 years after the Hindenburg and the false start of passenger aviation, Sergey Brin is relaunching blimps for sustainable aviation. (c.f Kim Stanley Robinson’s Ministry for the Future, which also prominently features airships as a climate solution)
Microsoft bet big on an exotic type of quantum computing hardware, which offered alluring theoretical properties around stability that conventional hardware types (e.g. superconducting, trapped-ion) did not. 3 years later, their seminal 2018 Nature paper was retracted. They gambled on a novel technology pathway and lost

We shouldn’t be too harsh on our predecssors - hindsight is 20/20 and even today, there are a variety of pathings choices and its not clear how things will turn out (at least to me).

future of warfare: the role of drones and autonomous vehicles, hypersonics, capital naval ships; the pace of decision making driven by AI-enabled systems
where we will fall on the spectrum between centralized power grid and distributed micro-grid as we transition to clean energy (this is a good example of a question that is primarily determined by political and regulatory choices, not technology)
what the portfolio of carbon removal technologies will look like (how much DAC vs BiCRS, nature-based restoration vs permanent sequestration) and how much CDR we will end up with
industrial heating decarbonization: hydrogen (local or networked), electrification, biomass, or carbon capture
will we ever achieve quantum advantage and how significant will it be? domain-specific computing architecture + never-ending improvements in classical algorithms have made finding practical applications with quantum advantage elusive. Which quantum computing hardware system will get there first?
How will US-China export controls, sanctions, and geopolitical tension splinter and fragment the development of a variety of technologies?

The technology we have today is certainly not the result of inexorbale, singular technological progress nor is it (solely) the product of individual, serendipitous discoveries. Rather, technology development is contingent on thousands of compounded human decisions and judgements made in a complex world, interwoven with regulatory environments, media landscapes, and black swan events.

My time at ARPA-E was the first concrete moment where I began to recognize the agency we have in building the future we want and how technology and its development is not something imposed on us but rather the result of many human interventions in our world. In each decision we make, on what to work on and who to work with, we are taking part in jostling and guiding our society towards the infinite possible futures.

I finally got around to writing more, so of course, I’m starting with the reflections I’ve had queued up in my head from a year ago 🙄 Here’s to less procrasting and more active reflection and writing!

The only theme/topic for this newsletter is strictly whatever I am thinking about at the moment. If (for some reason) you find that appealing, definitely subscribe!

Subscribe now

And if you’re interested in science and tech policy, federal funding apparatuses, or specifically working in Department of Energy, please reach out! I always to love to meet new people

1

For the interested, the National Academies Press has an accessible yet technical handbook on Quantum Computing, if slightly outdated.

2

These observations are also biased by the fact that the European VC environment is weaker than the US, though that is slowly changing as well. This makes the EU’s relative strength in startups in the photonics approach for quantum computing all the more significant.

Rough Drafts

Real-life Examples of Path Criticality in…