Defense Fordism - Lux Capital

Defense Fordism and AI

The war in Ukraine, if it comes (or came overnight — these things move fast), may well be one of the last wars we can process with human perception.

Much as the Boer War in South Africa at the turn of the twentieth century offered a savage preview of the weapons and tactics to come in the Great War, today’s conflicts foretell how war will be experienced in the coming decade. Unmanned and in some cases autonomous drones have already been used offensively in North Africa and Syria. Moore’s Law, which so far seems to apply equally to drone technology as it did to semiconductors, implies that the capabilities of autonomous hardware will increase exponentially while costs decrease. Full autonomous war is just around the corner.

The Pentagon is certainly aware of the future, but is it positioned to succeed? Drones and autonomous weapons have been studied for decades, and concerns about doctrine have been just as forthcoming. Paul Scharre at the Center for New American Security wrote a great book titled Army of None exploring the intricacies of the future of this warfare back in 2018. Yet, it’s one thing to ponder and strategize — another to actually experience and fight such a war firsthand.

Sue Halpern of The New Yorker had a great story this week on how this revolution is going to be experienced. She explores DARPA’s autonomous dogfighting challenge, with a specific focus on how human pilots are adapting to this new future of artificial intelligence:

In one scrimmage, his plane and the adversary’s chased each other around and around—on the screen, it looked like water circling a drain. The pilot told [Katharine] Woodruff [a researcher working with SoarTech] that, though he let the A.I. keep fighting, “I know it is not going to gun this guy anytime soon. In real life, if you keep going around like that you’re either going to run out of gas or another bad guy will come up from behind and kill you.” In an actual battle, he would have accepted more risk in order to get to a better offensive angle. “I mean, A.I. should be so much smarter than me,” he said. “So if I’m looking out there, thinking I could have gained some advantage here and A.I. isn’t, I have to start asking why.”

Latent throughout the piece is the same laconic dread that has animated American manufacturing for the past few decades — the sense that machines are taking over, production is being outsourced to “others,” that a way of life is vanishing. The Fordism that brought the postwar economic golden era in America is today’s wasteland of Fentanyl and social bile. Dreams of being a fighter pilot will soon go the way of securing a high-paying union manufacturing job in a prosperous industrial Midwest town.

Yet despite awareness of the future, one senses that nostalgia and inertia are still driving U.S. security policy. Far from punching ahead, the Pentagon’s goal with the Air Combat Evolution program is to support a human-computer hybrid model where humans are “battle managers” overseeing the work of their AI fighting counterpart. Yet, it’s already obvious that the artificial intelligence has little need for the human intelligence sitting in the cockpit:

Trust will also be crucial because, with planes flying at speeds of up to five hundred miles an hour, algorithms won’t always be able to keep pilots in the loop. [Peter Hancock, a psychology professor at the University of Central Florida], calls the discrepancy in reaction time “temporal dissonance.” As an analogy, he pointed to air bags, which deploy within milliseconds, below the threshold of human perception. “As soon as you put me in that loop,” he said, “you’ve defeated the whole purpose of the air bag, which is to inflate almost instantaneously.”

The dogfighting trials are early and ongoing, but AI pilots are arriving quickly. In fact, we will almost certainly have deployed AI fighter planes earlier than autonomous vehicles on our roads. Our cars have to halt at fallen stop signs, weave around construction sites, and navigate the intricacies of pedestrians, delivery vans, bikes, and more all without making a single, tragic mistake. An AI fighter plane? It may be hesitant to target and neutralize an enemy plane, but the constraints of land blow away in the air.

There is a discontinuity coming in war, where the costs won’t be born by the humans fighting, but exclusively by the humans who are being fought over. If, as some political scientists believe, the elimination of an active draft in the United States encouraged military adventurism by lowering the costs of going to war, the same will hold even more true when there aren’t much of any costs at all. Even worse, such a calculus will equally apply to America’s adversaries, from China and Russia to an insurgent.

Swarms of autonomous, cheap, violent autonomous weapons that can’t be stopped with conventional hardware. It’s the new Fordism of defense, but unfortunately, the United States is still obsessed with the old Fordism.

The Pentagon’s testing report for the new Gerald R. Ford class of aircraft carrier, which was obtained by Bloomberg this week and is expected to be submitted later this year to Congress, notes widespread shortcomings in the ship’s ability to defend itself as well as with its operational systems. The ship, which cost about $13 billion and first started construction in 2005, could see operational deployment later this year.

(That’s only slightly better than the James Webb Space Telescope which finally reached its destination this week and took nearly 24 years and about $8.8 billion to plan and construct).

Compare that prodigious spend on a single vessel to the entire autonomous warfare budget. According to Halpern at the New Yorker, the Pentagon will spend $1 billion on AI this year.

That’s the fundamental tension of defense acquisition today. The multi-decade commitments that advanced warfighting platforms like the Gerald R. Ford require are slamming straight into the Moore’s Law of autonomous machines. The Ford’s construction was underway prior to the introduction of the iPhone, and it’s still not in service because experimental Chinese weapons like supersonic missiles will likely prove effective in neutering it (an exercise that Beijing has already made a clear priority).

There’s the canonical line attributed to William Gibson about the future not being equally distributed, but what happens when the future is staring you right in the face and nostalgic blindness prevents a leap to the next generation? It’s a Brave New World out there, and the Fordism of the Navy’s past and all the rest of defense is running up against the Fordism of the future. Cheap and abundant will beat expensive and rare.

There are some positive motions to adapt to these new threats. Lux portfolio company Anduril announced a nearly $1 billion contract with the U.S. Special Operations Command (SOCOM) on an indefinite delivery indefinite quantity basis (which is government speak for an open-ended, fixed-time contract). Anduril will work as a systems integration partner with SOCOM on countering unmanned threats.

Down payments on the future are good. But we need more, faster — just like the AI weapons systems that will buzz around past the barrier of human perception.

Speaking of ARMing yourself for the future

Nvidia won’t be ARMing itself with ARM perhaps after all. The $40 billion September 2020 marriage proposal was one of the most important acquisition announcements in recent tech memory, but a year and a half later, the chipmaker looks set to abandon the deal according to a report in Bloomberg this week.

This was a fraught deal right from the beginning. Given ARM’s centrality to chip design, particularly in mobile and increasingly in embedded and automative applications, customers cried foul, worried that Nvidia would leverage its influence over ARM’s design roadmap for its own market gains. Then there were the antitrust reviews triggered in four(!) jurisdictions (U.S., U.K., EU and China) that are all quite frosty on the issue of national security and semiconductors.

It was something of a slow-motion train wreck, but Nvidia will skip ahead just fine. The same cannot be said of SoftBank, which currently owns ARM and paid $32 billion for the company back in 2016. SoftBank desperately needed liquidity from the ARM sale to pay back its colossal debt, and a breakup would be damaging. SoftBank has lost 55% of its value since a peak in February 2021, and Masayoshi Son’s right-hand man, Marcelo Claure, left the firm over what was reported to be a pay dispute.

With future ambiguities eliminated and tens of billions of acquisition money back on the table though, Nvidia has the bank to go to market for other properties, boding well for semiconductor exits, particularly given wider market turbulence.

A different direction on chips

Chiplets, a perennial favorite of our very own Shahin Farshchi, are in the limelight. As he puts it: “Chiplets are here, and they will become ubiquitous.”

Today, a single chip is often a combination of multiple processing systems manufactured together (what’s dubbed “System on a Chip” or SoC). If you’ve seen the Apple M1, then this is familiar: you have separate cores for processing, ML and GPUs all packed together onto one piece of silicon.

That model has great economics, but there is a clear downside to this all-in-one approach. Fabs are overwhelmed with demand right now, putting an extremely high premium on nodes with the smallest sizes such as TSMC’s 5nm and coming 3nm process. Machine learning cores need to run at the highest performance, but the same is not true for, say, memory or wireless modem systems. Right now, whichever system requires the smallest node constrains the selection of a fab for manufacturing.

Chiplets unbundle the systems on a chip. Each system can be manufactured on its own silicon (perhaps at a larger node with cheaper economics) and packaged together later. What gets more interesting is that engineers are stacking these chips in the Z-axis for much greater performance while using significantly less power. By working in three dimensions rather than two, the distance — and thus latency — between systems can be significantly reduced.

Once experimental, these technologies are hitting early scale, but challenges linger. AMD and Intel (via its Foveros technology) have incompatible standards and approaches, which is slowing progress, plus the fabs aren’t entirely ready for this new paradigm.

Nonetheless, there’s an opportunity here for startups to come in and build something just as this evolution in semiconductors crescendoes. One key area that Shahin has been interested in is technology that makes the multi-chip packages that connect chiplets together into a single unit more efficient and less expensive. Another opportunity is to build a “chiplet-native” startup that uses the enhanced performance of this chip design to outcompete on certain workloads.

Chiplets and 3D stacking are getting underway, but they will be definitive in the coming years and represent one of the largest macro trends coming up in semiconductors.

Lux Recommends

Peter Hébert recommends a WSJ piece on a breakthrough in regenerative medicine. Scientists have been able to trigger the regrowth of limbs in frogs via a drug cocktail applied to the injury.
Deena Shakir recommends this viewpoint published in JAMA on adding health equity as a fourth objective to the traditional trio of ”improving population health, enhancing the care experience, and reducing costs” of healthcare improvement.
Adam Goulburn highlighted new research out of Caltech that begins to elucidate where our feeling of thirst comes from. Increasing our understanding of the gut-brain axis will hopefully offer us new therapeutics, an area that Lux portfolio company Kallyope is focused on.
Our scientist-in-residence Sam Arbesman offers this essay by Dennis Overbye on the seminal meaning of the James Webb Space Telescope’s deployment.
From me, a fascinating conversation on philosophy and science fiction (along with five book recommendations) between Eric Schwitzgebel and Nigel Warburton.
And for something fun, Sam offers this analysis of what happens when you mix breakfast cereal press releases with GPT-3. It’s not tasty.

That’s it, folks. Have questions, comments, or ideas? This newsletter is sent from my email, so you can just click reply.

Forcing China’s AI researchers to strive for chip efficiency will ultimately shave America’s lead

In incididunt ad qui nostrud sint ullamco. Irure sint deserunt Lorem id officia dolore non. Anim dolor minim sit dolor et sint aliquip qui est. Ex in tempor laborum laboris dolor laboris ullamco quis. Enim est cupidatat consequat est culpa consequat. Fugiat officia in ea ea laborum sunt Lorem. Anim laborum labore duis ipsum mollit nisi do exercitation. Magna in pariatur anim aute.

In incididunt ad qui nostrud sint ullamco. Irure sint deserunt Lorem id officia dolore non. Anim dolor minim sit dolor et sint aliquip qui est. Ex in tempor laborum laboris dolor laboris ullamco quis. Enim est cupidatat consequat est culpa consequat. Fugiat officia in ea ea laborum sunt Lorem. Anim laborum labore duis ipsum mollit nisi do exercitation. Magna in pariatur anim aute.

‍

Right now, pathbreaking AI foundation models follow an inverse Moore’s law (sometimes quipped “Eroom’s Law”). Each new generation is becoming more and more expensive to train as researchers exponentially increase the number of parameters used and overall model complexity. Sam Altman of OpenAI said that the cost of training GPT-4 was over $100 million, and some AI computational specialists believe that the first $1 billion model is currently or will shortly be developed.

As semiconductor chips rise in complexity, costs come down because transistors are packed more densely on silicon, cutting the cost per transistor during fabrication as well as lowering operational costs for energy and heat dissipation. That miracle of performance is the inverse with AI today. To increase the complexity (and therefore hopefully quality) of an AI model, researchers have attempted to pack in more and more parameters, each one of which demands more computation both for training and for usage. A 1 million parameter model can be trained for a few bucks and run on a $15 Raspberry Pi Zero 2 W, but Google’s PaLM with 540 billion parameters requires full-scale data centers to operate and is estimated to have cost millions of dollars to train.

Admittedly, simply having more parameters isn’t a magic recipe for better AI end performance. One recalls Steve Jobs’s marketing of the so-called “Megahertz Myth” to attempt to persuade the public that headline megahertz numbers weren't the right way to judge the performance of a personal computer. Performance in most fields is a complicated problem to judge, and just adding more inputs doesn't necessarily translate into a better output.

And indeed, there is an efficiency curve underway in AI outside of the leading-edge foundation models from OpenAI and Google. Researchers over the past two years have discovered better training techniques (as well as recipes to bundle these techniques together), developed best practices for spending on reinforcement learning from human feedback (RLHF), and curated better training data to improve model quality even while shaving parameter counts. Far from surpassing $1 billion, training new models that are equally performant might well cost only tens or hundreds of thousands of dollars.

This AI performance envelope between dollars invested and quality of model trained is a huge area of debate for the trajectory of the field (and was the most important theme to emanate from our AI Summit). And it’s absolutely vital to understand, since where the efficiency story ends up will determine the sustained market structure of the AI industry.

If foundation models cost billions of dollars to train, all the value and leverage of AI will accrue and centralize to the big tech companies like Microsoft (through OpenAI), Google and others who have the means and teams to lavish. But if the performance envelope reaches a significantly better dollar-to-quality ratio in the future, that means the whole field opens up to startups and novel experiments, while the leverage of the big tech companies would be much reduced.

The U.S. right now is parallelizing both approaches toward AI. Big tech is hurling billions of dollars on the field, while startups are exploring and developing more efficient models given their relatively meagre resources and limited access to Nvidia’s flagship chip, the H100. Talent — on balance — is heading as it typically does to big tech. Why work on efficiency when a big tech behemoth has money to burn on theoretical ideas emanating from university AI labs?

Turn to China now, and the story is in a very different place. The Biden administration has placed a series of export controls on advanced computation and semiconductor fabrication to China, including blocking the export of Nvidia’s H100 and A100 chips. Nvidia responded in March by offering tweaked versions that comply with U.S. export controls (with what it dubs the H800 chip), but the general policy remains in force: prevent China from acquiring the best chips that Earth produces.

Without access to the highest-performance chips, China is limited in the work it can do on the cutting-edge frontiers of AI development. Without more chips (and in the future, the next generations of GPUs), it won’t have the competitive compute power to push the AI field to its limits like American companies. That leaves China with the only other path available, which is to follow the parallel course for improving AI through efficiency.

For those looking to prevent the decline of American economic power, this is an alarming development. Model efficiency is what will ultimately allow foundation models to be preloaded onto our devices and open up the consumer market to cheap and rapid AI interactions. Whoever builds an advantage in model efficiency will open up a range of applications that remain impractical or too expensive for the most complex AI models.

Given U.S. export controls, China is now (by assumption, and yes, it’s a big assumption) putting its entire weight behind building the AI models it can, which are focused on efficiency. Which means that its resources are arrayed for building the platforms to capture end-user applications — the exact opposite goal of American policymakers. It’s a classic result: restricting access to technology forces engineers to be more creative in building their products, the exact intensified creativity that typically leads to the next great startup or scientific breakthrough.

If America was serious about slowing the growth of China’s still-nascent semiconductor market, it really should have taken a page from the Chinese industrial policy handbook and just dumped chips on the market, just as China has done for years from solar panel manufacturing to electronics. Cheaper chips, faster chips, chips so competitive that no domestic manufacturer — even under Beijing direction — could have effectively competed. Instead we are attempting to decouple from the second largest chips market in the world, turning a competitive field where America is the clear leader into a bountiful green field of opportunity for domestic national champions to usurp market share and profits.

There were of course other goals outside of economic growth for restricting China’s access to chips. America is deeply concerned about the country’s AI integration into its military, and it wants to slow the evolution of its autonomous weaponry and intelligence gathering. Export controls do that, but they are likely to come at an extremely exorbitant long-term cost: the loss of leadership in the most important technological development so far this decade. It’s not a trade off I would have built trade policy on.

The life and death of air conditioning

Across six years of working at TechCrunch, no article triggered an avalanche of readership or inbox vitriol quite like Air conditioning is one of the greatest inventions of the 20th Century. It’s also killing the 21st. It was an interview with Eric Dean Wilson, the author of After Cooling, about the complex feedback loops between global climate disruption and the increasing need for air conditioning to sustain life on Earth. The article was read by millions and millions of people, and hundreds of people wrote in with hot air about the importance of their cold air.

A few weeks ago, Bloomberg had a great piece about the rising toll of air conditioning under the similarly scary headline, “A Billion New Air Conditioners Will Save Lives But Cook the Planet.” The numbers, as always, are staggering:

Demand for air conditioners is surging in markets where both incomes and temperatures are rising, populous places like India, China, Indonesia and the Philippines. By one estimate, the world will add 1 billion ACs before the end of the decade. The market is projected to
before 2040. That’s good for measures of public health and economic productivity; it’s unquestionably bad for the climate, and a global agreement to phase out the most harmful coolants could keep the appliances out of reach of many of the people who need them most.

This is a classic feedback loop, where the increasing temperatures of the planet, particularly in South Asia, lead to increased demand for climate resilience tools like air conditioning and climate-adapted housing, leading to further climate change ad infinitum.

And there doesn’t appear to be a way out. Scientists have determined that there is a zone of comfort for humans where we can function at peak performance. Much of the industrialized world happens to sit in these temperate zones, which is one argument for their long-term economic success compared to the tropics. But as climate change intensifies, the world’s temperate zones are expected to narrow dramatically, and some countries like India may have next to no temperate zones left to protect its people. Air conditioning is a necessity (with some even calling it a human right), but greener coolant alternatives are still in their infancy. In other words, cooling innovation needs to heat up in a big way.

Lux on the Farm

Josh Wolfe speaking at Nvidia Auditorium at Stanford. Photo by Chris Gates.

Josh Wolfe gave a talk at Stanford this week as part of the school’s long-running Entrepreneurial Thought Leaders series, talking all things Lux, defense tech and scientific innovation. The
.

Lux Recommends

As Henry Kissinger turns 100, Grace Isford recommends “Henry Kissinger explains how to avoid world war three.” “In his view, the fate of humanity depends on whether America and China can get along. He believes the rapid progress of AI, in particular, leaves them only five-to-ten years to find a way.”
Our scientist-in-residence Sam Arbesman recommends Blindsight by Peter Watts, a first contact, hard science fiction novel that made quite a splash when it was published back in 2006.
Mohammed bin Rashid Al Maktoum, and just how far he has been willing to go to keep his daughter tranquilized and imprisoned. “When the yacht was located, off the Goa coast, Sheikh Mohammed spoke with the Indian Prime Minister, Narendra Modi, and agreed to extradite a Dubai-based arms dealer in exchange for his daughter’s capture. The Indian government deployed boats, helicopters, and a team of armed commandos to storm Nostromo and carry Latifa away.”
In the happier news department of science, Ina Deljkic recommends “’Quite science fiction’: Brain-spine interface system helps man with paralysis walk again.”
Sam recommends Ada Palmer’s article for Microsoft’s AI Anthology, “We are an information revolution species.” “If we pour a precious new elixir into a leaky cup and it leaks, we need to fix the cup, not fear the elixir.”
I love complex international security stories, and few areas are as complex or wild as the international trade in exotic animals. Tad Friend, who generally covers Silicon Valley for The New Yorker, has a great story about an NGO focused on infiltrating and exposing the networks that allow the trade to continue in “Earth League International Hunts the Hunters.” "At times, rhino horn has been worth more than gold—so South African rhinos are often killed with Czech-made rifles sold by Portuguese arms dealers to poachers from Mozambique, who send the horns by courier to Qatar or Vietnam, or have them bundled with elephant ivory in Maputo or Mombasa or Lagos or Luanda and delivered to China via Malaysia or Hong Kong.”
Finally, Sam recommends Brick Technology’s video of “Making A Billion-Year Lego Clock.”

That’s it, folks. Have questions, comments, or ideas? This newsletter is sent from my email, so you can just click reply.

‍

January 22, 2022

Lab-grown Babies

<- Previous

continue

reading

February 1, 2022

The West’s self-defeating technological sovereignty

Next ->