Welcome to Stuff I Thought About Last Week, a personal collection of topics on tech, innovation, science, the digital economic transition, the finance industry, Nicolas Cage, and whatever else made me think last week.

Click HERE to SIGN UP for SITALWeek’s Sunday Email.

In today’s post: using AI to control rapidly changing systems from fusion to driving; the use of simulated data lowers barriers to AI creation; robots that pick strawberries and turn them into smoothies; big tech's misguided VR plans; health wearables continue to gather groundbreaking data; Gen X finally matters...a little bit; does technology always lead to centralization?; and much more below.

Stuff about Innovation and Technology
VR Hazards
British insurance company Aviva notes a 31% rise in damage claims from VR-related accidents in homes, with an average cost of £650 (~$880), largely due to damaged TVs. I’m a broken record when it comes to touting the benefits of augmented reality over full virtual reality for almost every type of app (there are perhaps some exclusions around truly immersive games, but, even then, AR can accomplish some amazing effects). Having a layer of real-world presence greatly improves the experience (and decreases the injuries and insurance claims!) and creates a special magic in the brain by improving reality instead of replacing it completely. I think passthrough VR, which uses video of the setting you are in to simulate AR, has all the drawbacks of VR as well. For now, most of the big tech platforms are going in the wrong direction by focusing on immersive VR.

AI Speedsters and Synthetic Data
GT Sophy is an AI trained to win Sony’s Gran Turismo game. Driving games require faster reflexes and continuous judgements, in contrast to simpler, more discrete games previously mastered by AI. GT Sophy might be part of Sony’s broader auto efforts, if it turns out to be useful for self-driving AI. Training in simulations is part of a larger AI trend, as Nvidia’s VP of simulation technology, Rev Lebaredian, explains in this IEEE interview. Nvidia is developing Omniverse software to simulate wildly complex systems like the entire planet and everything on it. Despite the allure of omniscience these models present, we know from complex systems that we cannot predict the future (since tiny, chaotic perturbations spiral into massive unknowns as time passes), but simulations will give us glimpses at many possible futures. Of course, useful, detailed, real-world simulations are extremely computationally intense, which helps sell more Nvidia chips. Another interesting consequence of simulations is their extensive use of simulated data. We tend to think of Big Internet companies as having an advantage thanks to their massive, exponentially increasing troves of data. However, simulated data is comparatively easy to come by. Those of us trained to expect the unexpected, however, will assume that not all data sets, rather real or simulated, are terribly comprehensive or useful in predicting the future.

AI-Controlled Magnetic Fusion Cage
Google’s DeepMind has trained a neural network to control the plasma in a fusion reactor. I first noted Google was working on this problem back in #277, and the team recently published a Nature paper highlighted by Wired. One of the trickiest parts of harnessing fusion energy is holding the plasma – the superheated soup of colliding hydrogen nuclei and electrons – stable long enough to reliably extract power. As Wired notes, 19 magnetic coils have to hold the plasma in place without touching the plasma (you know, like Nic Cage in The Sorcerer’s Apprentice). The trick is to continuously adjust the magnetic fields to optimize the shape of the plasma, which DeepMind accomplished with a deep reinforcement learning system that can handle an ever-changing set of circumstances (another example of AI moving beyond games like Chess and Go that have only a limited number of options and discrete steps). In a main-sequence star like our Sun, nuclear fusion happens as hydrogen turns into helium. The thermal fusion core of the Sun is both promoted and held in place by the gravitational pull of the outer layers in what’s known as hydrostatic equilibrium. Without the massive gravity of the outer layers of a star, fusion reactors on Earth need to run at a much higher temperature (150M °C) vs. the center of the Sun (14M °C). Once fusion runs its course in a star, the gravitational forces take over and collapse continues. If the star's mass is large enough, a rapid collapse fuses helium into heavier elements eventually resulting in a supernova that throws off the ingredients necessary for life in the universe and leaves behind a black hole. Luckily, fusion reactors on Earth can avoid this evolutionary path!

AI-Optimized Video Streaming
In other DeepMind news, YouTube is now using the AI division’s MuZero reinforcement learning algorithm to stream videos. AI improvements to VP9 open-source compression resulted in a 4% decline in bandwidth with no change to video quality. I wonder how much improvement we would see if AI were used to design an entirely new compression algorithm from scratch and a custom chip to more efficiently run such an algorithm? MuZero was originally designed to master the strategy game Go, and, in so doing, to create novel ways to win the game against humans. Google has previously applied other algorithms from DeepMind in its data centers to decrease power consumption.

BlenderBots
As labor shortages persist, Restaurant Dive reports that 50% of restaurant operators plan to deploy automation in the next 2-3 years. Jamba Juice is rolling out more robotic smoothie stations powered by Blendid. The stations are essentially a robotic arm and a series of dispensers, as you can see in this video. The speed at which the robot arm moves seems far slower than humans. At least with digital orders, I suppose being able to batch and schedule overcomes some of this slowness…or maybe they are just afraid to crank up the speed (also, the bots leave it to the human customers to attach the lid and insert the straw). In related news, White Castle is deploying Miso’s Flippy 2 robots in 100 additional locations. Flippy 2 is able to operate an entire fry station. Notably, both Blendid and Miso have recently opted for crowd equity funding despite the unlimited amount of free money that VCs are handing out. With the progress being made in robotic strawberry-picking machines, we may soon have farm-to-blender smoothies from our robot overlords.

Miscellaneous Stuff

Good to Feel Cruddy After Vaccination
The latest TemPredict Study investigated whether temperature data from health wearables correlated with COVID immunization success, as measured by antibody production. I previously covered (#272) the first TemPredict study’s successful early detection of fevers in patients before they noticed COVID symptoms. This second study shows that elevated temperature following vaccination was positively associated with higher subsequent antibody production, effectively showing that individuals with a more significant physiological response to the shots ended up with more immunity.

Stuff about Geopolitics, Economics, and the Finance Industry
Economic Torch Passes to Gen X
Live sports continued their rebound from pandemic lows as the Super Bowl saw a 14% increase over 2021 to 112M viewers, including 8% more linear viewers and a doubling of streamers to 11.2M. I caught the 1990’s/early-2000’s nostalgic halftime show, which, as I noted to my family, seemed appropriately geared toward us Gen X’ers as Boomers hand us the consumption torch. Gen X currently ranges from 42 to 57 years old, with peak household consumption in the US taking place between ages 45 and 54. Gen X is a smaller generation by birth, which is somewhat mitigated by higher immigrant numbers. The biggest trough is people aged 43 to 50 born in the mid-1970s. The millennial generation, just now starting to turn 40, is coming up behind in greater numbers (see The 30-Something Sneaker Wave), but with more debt and less savings than Gen X and Boomers had at the same age.

Precision Targeting Inflation
I’d like to see governments use more tactical and targeted policy to fight inflation rather than shocking the system with interest rate hikes. Raising overall rates fights a very long trajectory of rate declines, as economies become naturally more leveraged over time (see The Improbability of Rising Rates). This leverage is not something to fear as long as it’s understood by policymakers (which, it clearly is not), but can be a problem with misguided rate hikes. For example, if a shortage of labor is a primary driver of inflation, rather than raising rates, governments can incentivize workers to re-enter the workforce with targeted benefits. A good example is Texas’ Service Industry Recovery Child Care Program, which pays the full cost of child care for parents working more than 25 hours a week for households at or below 75% of the state’s median income. Of course, these work incentives could be sponsored by the private sector as well. Employers struggling to find enough employees could directly offer such services or high enough wages for employees to cover their own costs. There is obviously a double-edged knife here. I am never one to ask for more regulation (e.g., I would much rather companies see the value in providing childcare than leave it to the government), but if the alternative is the blunt rock of monetary policy, I favor experimentation with targeted incentives.

Is Decentralized Technology an Oxymoron?
I read with interest these lecture notes on cryptocurrencies by David Rosenthal, an early employee of Nvidia and author of an award-winning 2003 paper on proof-of-work networks. Rosenthal discusses, among other drawbacks to crypto, a paradox that arises from the desire to decentralize systems. Essentially, the idea comes down to increasing returns (and Rosenthal notes Brian Arthur’s seminal paper that we often reference at NZS). According to Rosenthal: “Information technologies have strong economies of scale, so the larger the miner the lower their costs, and thus the greater their profit, and thus the greater their market share.” Further along he notes: “If a system is to be decentralized, it has to have a low barrier to entry. If it has a low barrier to entry, competition will ensure it has low margins,” and thus we can assume that any for-profit (i.e., VC-backed) decentralized business is attempting to create high, centralized barriers. Rosenthal dives into several other problems with crypto, from energy consumption to crime, and concludes: “It would be wonderful if we could figure out how to build a Web that would resist centralization. But all the technical and financial cleverness that's been poured into cryptocurrencies hasn't succeeded in doing that.” At the risk of removing too much nuance from Rosenthal's more detailed arguments, my takeaway was the following: when technology is the enabler of a system, centralization ensues. And, perhaps it is unavoidable when technology and profits combine to enable innovation. Power laws are natural tendencies in information-based systems, which is why we currently have a small number of very large Internet platforms (see How I Learned to Stop Worrying and Love the Monopoly). As I noted in #314, I think it’s logical to assume we will see continued efforts to ban cryptocurrency speculation (at least for the next several years, if not longer). Governments fear competitors to fiat and the damaging, crypto-enabled ransomware plague, among other factors. Hitting the pause button might be what is needed for engineers and decentralized visionaries to regroup and try to solve this paradox of decentralization (if it is indeed a problem). I remain optimistic as always that bits and pieces of all new technologies will separate, combine, break apart, reform, etc., and continue to create a more efficient digital future that moves everyone forward in progress. Blockchain seems like it will be one of these additive elements over time. Debate, failures, and reinvention have always been a part of human progress, and these cycles of innovation are speeding up.

✌️-Brad

Disclaimers:

The content of this newsletter is my personal opinion as of the date published and is subject to change without notice and may not reflect the opinion of NZS Capital, LLC.  This newsletter is an informal gathering of topics I’ve recently read and thought about. I will sometimes state things in the newsletter that contradict my own views in order to provoke debate. Often I try to make jokes, and they aren’t very funny – sorry. 

I may include links to third-party websites as a convenience, and the inclusion of such links does not imply any endorsement, approval, investigation, verification or monitoring by NZS Capital, LLC. If you choose to visit the linked sites, you do so at your own risk, and you will be subject to such sites' terms of use and privacy policies, over which NZS Capital, LLC has no control. In no event will NZS Capital, LLC be responsible for any information or content within the linked sites or your use of the linked sites.

Nothing in this newsletter should be construed as investment advice. The information contained herein is only as current as of the date indicated and may be superseded by subsequent market events or for other reasons. There is no guarantee that the information supplied is accurate, complete, or timely. Past performance is not a guarantee of future results. 

Investing involves risk, including the possible loss of principal and fluctuation of value. Nothing contained in this newsletter is an offer to sell or solicit any investment services or securities. Initial Public Offerings (IPOs) are highly speculative investments and may be subject to lower liquidity and greater volatility. Special risks associated with IPOs include limited operating history, unseasoned trading, high turnover and non-repeatable performance.