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

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

In today’s post: Following my lookback at the evolving media and AI industries, this week we’ll recap 1) how the labor shortage is fueling automation, 2) algorithmic and neural trickery, and 3) some of my favorite celebrity interviews – after we cover a couple new stories from last week.

AI Research Assistants

One of the bigger use cases of AI could be finding overlooked data from published academic research. One such example is Elicit by Ought, a non-profit search engine that leverages GPT-3 to find relevant research from open papers and a handful of academic journals (the majority of academic publishers are not participating). Given the volume and density of academic research, having an AI engine that can understand, synthesize, and connect dots (much like a good review article) could dramatically increase the speed of fundamental research in a variety of fields. Google has long had the Scholar academic search engine, and they have scanned every book in the library, so it seems plausible they could apply one of their AI engines in a similar fashion. If the major journals could get on board and see that, rather than being a threat, such AI projects could widely accelerate overall research and the importance of peer-reviewed journals, AI meta research could really take off.

Gov’t Registers Labor Deficit

The Biden administration, along with the Federal Reserve, is honing in on our problematic structural labor shortage. The WSJ reports that the White House is looking at ways to assist with childcare to help bolster workforce numbers. While working from home can (in some cases) make it easier for parents to stay in the labor force, for the majority of families lacking such options, the cost of childcare is outpacing wage growth, causing a structural issue. In the past, the government has used large-scale social engineering programs to promote births, home ownership, etc. (especially following WWII), and we could see a coordinated effort to revive or embolden these policies. Another area that could help strapped labor markets is the rapid rebound in US immigration. It’s too early to tell if this increase is a backlog from pandemic lockdowns frustrating international travel, or if it’s a true return to pre-Trump levels of immigration. Meanwhile, putting even more pressure on the pool of available labor, COVID and opioids have again decreased life expectancy in the US, which is now back to 1996 levels.

1) Structural Labor Shortfall Drives Automation

As I’ve covered several times over the last couple of years, the structural limitations of labor are likely to speed up the adoption of automation in both information- and labor-intensive jobs. Back in #331, I wrote:

Another trend that stands out in my population model is the steady decline of working-age adults in the US. Driven by the lack of immigrants, the increased death rate for middle-aged workers from the opioid crisis, and lower births some twenty years ago, folks aged 20 to 64 look to be slightly down over the next decade. On the flipside, the diminished working age population puts the spotlight on folks over 65, who will grow at 1.68% per year through 2030. Traditionally, retirees spend and consume far less than the working-age population, so an aging population tends to be a headwind for consumption growth. However, Boomers have accumulated significant wealth in markets and home equity, thanks to decades of accommodative stimulus and rate policies, so perhaps they will drive more consumption in retirement than expected. None of these projections is revolutionary, it’s simply the continued outcome of what’s been happening for several decades: declining birth rates in developed countries. The incremental change is the slowdown in immigration and the risk of reaching a breaking point on labor availability, which could be inflationary for many years until technology and automation advance to offset it.

Of perhaps greater note, if you roll my model forward with the same assumptions, the US population begins to decline in 2035, and, by 2050, there are several million fewer people in the US than there are today. This is a far cry from various models out there which optimistically show 50-70M more people in the US than we have today by 2050. There seems to be a massive disconnect between the general expectation and the reality of where we are headed. If the US aimed to keep the population flat through 2050, we would need to attract an average of 100,000 more immigrants per year than 2021’s level, and simultaneously keep the birth rate from falling while stabilizing life expectancies...

The real problem with automation of course is the difficulty of replicating the incredible efficiency of humans – from #351:

Amazon reportedly was worried about running out of workers by 2024 based on the growth path they identified during the pandemic. Of course, we now know that projection was largely based on an error in their forecasting systems, an example of the pervasive recency bias we saw in the pandemic. At some point though, as the analog-to-digital transition of the economy creates natural power laws, Amazon might again find itself facing a worker shortage. This threat puts the burden of filling the labor gap on automation. On this front, Amazon has always struck me as being well behind the curve. While they excel at software automation, they have largely built their logistics around people. Even with warehouse robotics, they’ve only just announced that they have an autonomous pallet-moving robot that can work alongside people (previously, Amazon fulfillment center robots only worked in caged-off areas). It’s been over ten years since Amazon acquired the warehouse robotics startup Kiva with the goal of modernizing logistics with such free-range, autonomous, “human equivalent” bots. The pace of progress seems painfully slow. Part of the issue is the complexity and fragmentation of the robotics industry and use cases, as well as the lack of a platform. And, of course, humans are still just more efficient at most tasks...

The pace of progress in automation doesn’t appear high enough to meet the potential demand from reshoring or a sizable green infrastructure push. From Reshoring Rising (#354):

This FT op-ed addressing the changing nature of global trade contains several useful charts. Notably, they illustrate the peak of trade occurring over fifteen years ago (something I’ve covered before) as well as the decline in China’s wage competitiveness, which began leveling off around ten years ago. I believe fears that deglobalization will be inflationary have little merit based on the seemingly incidental impact of years of ebbing international trade (although, it’s admittedly hard to parse causation from correlation given everything happening, especially with aging populations and technological progress). The inflationary impact could certainly change if there is a significant uptick in reshoring, but I suspect there is a natural cadence to just how fast supply chains can move capacity and repopulate their labor force. If it took half a century to globalize, it will probably take that long to reverse. A breakthrough in general-purpose humanoid robotics like EVE or other automation technology could accelerate reshoring, but such technological leveraging would provide compensatory deflation. The WSJ reports on a perfume company with annual sales of ~$1B that has rapidly shifted capacity back to the US and is now sourcing 70% of inputs from US suppliers. It would take a significant further reduction in Chinese costs to shift back overseas. Meanwhile, Bloomberg reports a rapid acceleration in reshoring and nearshoring of manufacturing, 10x above pre-pandemic levels. The construction of new manufacturing in the US is up 116% y/y (note: expensive new chip fabs in Arizona may be a big factor in that number). The CEO of GE Appliances (owned by Chinese parent Haier) began reshoring to the US in 2008 and sees it as the way to go for producing large items with higher quality for less cost. Generator maker Generac has shifted from China and now sources more than half of their supply from the US and Mexico. I’ve theorized that, after decades of shifting overseas, deglobalization is a challenge given the lack of labor and infrastructure – not to mention the lost know-how – but there is clear evidence building that, at least in some cases, reshoring is economically and logistically feasible, in part thanks to technology. As I’ve said before, remaining largely a global trade society is far better for peacekeeping and progress, so finding an equilibrium between domestic and international trade/manufacturing would be ideal for ongoing prosperity.

During last summer’s extreme heat wave, I covered the labor headwind to green infrastructure and other upgrades necessitated by climate change – from #355:

Extreme heat is taxing a world built for lower temperatures. Melting roads and runways, warped train tracks, and data-center cooling failures are just a few examples. Some of these issues – like improving data-center cooling – can be solved with technology and AI, but most of the problems require expensive, labor-intensive efforts. The demographically shrinking labor force in developed countries means that we might need to simply adjust to a world with Internet outages, unreliable transportation infrastructure, power rationing, etc. From #310:

As I was thinking about the cost to upgrade infrastructure to handle more extreme weather swings, it seems like there are a lot of ~$20B projects under consideration. A couple weeks ago, I mentioned (#306) it would cost about that much for PG&E to bury a portion of its power lines in high risk areas, and for Detroit to upgrade their stormwater drainage system. Apparently, it was also determined that damming the Golden Gate Bridge to keep rising tides at bay would cost...$19B. It’s certainly easy to see how expenses could add up to well into the trillions. I can’t help but wonder where the labor will come from (and with what incentives) to even consider some of these projects. 21,000 people were involved in the construction of the Hoover Dam. Is it less people-intensive to build a dam today than during the New Deal era? It’s possible that governments won’t even be able to contemplate breaking ground until deflationary automation/robotics renders construction more affordable.

In the long term, we will solve these problems through technological innovation, but, in the meantime, try to stay cool and pray that TikTok has good cooling for its servers.

And, I’ve gone a whole quarter without mentioning Sippy, Flippy, and Chippy, so here they are in an encore appearance from #364:

Robots may increasingly be sold as a service. I’ve highlighted Miso’s Flippy and Chippy fry-cook-bots before, which have a monthly cost of $3500. Essentially, robots can be value priced as an ongoing subscription against the cost of human labor. The robots can even cost the same as an employee because they come with fewer pesky issues, like needing healthcare benefits or time off for vacation/illness/family, and they can work 24 hours a day. Further, the complexity of humanoid replacements necessitates ongoing upgrades and maintenance. WaPo has a detailed report, including several videos of Miso’s kitchen-bots, aimed at replacing the dull, dirty, and dangerous jobs humans would rather not do. Miso also has a new robot, Sippy, which can make fountain drinks with new spill-proof lids faster than humans.

2) Algorithms and the Gullible Human Brain

A recurring topic in last year’s newsletter was the creeping influence of algorithms over all aspects of life and the economy. Such AI engines take advantage of our own neural shortcuts and biases to influence our behavior in surprisingly subversive ways. They also introduce new, programmed biases of their own, which appear to have markedly exacerbated economic volatility. I kicked off 2022 with the Algorithmic Threat to the Illusion of Free Will (#328):

To minimize energy input and optimize survival, the human brain evolved as a prediction machine, attempting to anticipate what might happen based on prior experiences, and then adjusting predictions to match input from our seven senses (the traditional sight, sound, touch, smell, and taste along with thoughts and emotions, which are best perceived as sensory inputs). One of the predictions our brains have to make is how other people will behave. And, of course those other people have their own neural algorithms making the same types of predictions about others as well. Historically, we’ve lived in small tribes with many shared experiences, and that’s important because the main factor the brain uses to make future predictions is prior knowledge. When people have a shared culture and common history, then they are likely to make more similar predictions, which makes life, well, more predictable. In the global, always connected world, we increasingly lack a common culture (see Digital Tribalism), which makes it more complex for the brain to predict the behavior of others.

Adding to this prediction complexity, we are now operating alongside a growing number of algorithms that are also making predictions about us and others based on prior behavior. These algorithms might determine whether your rental application/work resume is considered, who you date, what news you read, what medical care you receive, etc. Last week, the WSJ reported that more than 30,000 US churches are using data amassed by Colorado-based Gloo to recruit new members by targeting vulnerable individuals whose stats suggest they are experiencing personal struggles. For example, churches can home in on people who Gloo identifies as going through a divorce (based on connecting up credit card activity, travel bookings, and health attributes). Your browsing data might cause you to become a Baptist, a Catholic, or enter a rehab center depending on who pays the most for your data and is able to influence your brain’s future decisions through social network ads. Algorithms now even cause us to smile less, according to Allure Magazine, as people emulate the growing trend of influencers who emulate the models who stopped smiling in the 1990s for a variety of reasons.

This clash of complex prediction engines puts a spotlight on our already tenuous relationship with the concept of free will. While we feel like we have agency over our actions, neuroscience has informed us that the brain typically makes decisions well before we are consciously aware of them. In a small, isolated community, it’s perhaps easier to maintain that all-important illusion of free will because everything seems more predictable and rational. But, there are two ways this new complex set of interacting prediction engines highlights the illusory nature of free will. First, because we increasingly lack common culture, we can now see many other people making decisions that we simply don’t understand (and they, in turn, may view our decisions with the same confusion!). Second, we are becoming more aware that black box algorithms beyond our control are making decisions and predictions that impact our lives, sometimes in profound ways. These external algorithms, it turns out, are not that dissimilar from our own internal neural daemons – they just have different inputs and programmers. Our brain is doing its best to guess its way through life in a way that preserves its all-important vessel (that would be us), based on information at hand from prior experiences, all the while giving our conscious self the sense that we are in the driver’s seat. The state of our free agency, however, is not completely as hopeless as it might sound. A couple weeks back, I mentioned Lisa Feldman Barrett’s suggestion that the best way to gain some control over your brain’s decision making process is to actively change your behavior today so that, in the future, your brain has new sets of patterns on which to base predictions. Put simply, good habits can pay off for your own behavior. But, what about the increasing control that these enigmatic external algorithms wield over our futures? There are no good habits we can adopt to alter their impact on our lives without completely forgoing use of technology.

With more and more black-box algorithms interacting and influencing us, I see five ways to respond to this increasing lack of predictability and control in modern life: 1) try to imagine – and then follow – good intentions and habits you want your future self to use as prediction-engine inputs; 2) try to create a landscape for good luck to come knocking, or, at the very least, learn to see good luck when it comes your way (e.g., by cultivating mindfulness as discussed in our essay: Time Travel to Make Better Decisions); 3) build adaptability into as many aspects of your life as possible so that you can respond flexibly no matter what unpredictable things happen; 4) realize that bad luck is just as likely as good, and if you see someone who is missing out on good luck, try to help them; and, lastly, 5) cultivating an awareness of these prediction machines, whether it’s your brain, someone else’s brain, or an external algorithm, gives great perspective on daily life.

I covered the topics of misleading algorithms hawked to consumers and AI-induced economic volatility in Magic AI-Ball (#357):

Companies are being increasingly conned into buying decision-making software and tools claiming to use “AI” and algorithms to predict their future path through a complex adaptive system like the economy. In one example, McKinsey makes the following sales pitch for their AI forecaster QuantumBlack: “Transform faster. Innovate smarter. Anticipate the future. At QuantumBlack, we unlock the power of artificial intelligence (AI) to help organizations reinvent themselves from the ground up—and accelerate sustainable and inclusive growth. We do this by harnessing the foresight and precision of data and technology with the creativity and understanding of people. The result? Hybrid intelligence, a source of competitive advantage that transforms how companies think, operate, and disrupt.” (Seriously!? Hopefully, I am not the first person to break the news that consultants are full of bologna.) Like “dotcom” twenty five years ago, “AI” is fast becoming a standard marketing gimmick that won’t materially change the underlying businesses for at least a decade or two (see AI is the New Dotcom for more). Complex adaptive systems science teaches us that we can only prepare and adapt to the future, not forecast it with any accuracy. However, most peddlers of prediction engines either don’t realize this paradox or choose to ignore it. A great recent example of the failure of highly sophisticated tools/algorithms to predict the future is Amazon’s SCOT system, which, along with human influence, incorrectly predicted future ecommerce demand during the pandemic, leading to substantial overbuilding of capacity. Despite AI being largely a catch phrase (for now), the increased use of AI tools/software add-ons will have one tangible impact: a significant increase in the amplitude of feedback loops in the economy. Amazon's SCOT error is one such example as the company over hired and overbuilt, and is now reversing what would have otherwise been a much smaller increase in capacity. In the stock markets, we saw volatility rise with increasing implementation of quantitative strategies and autonomous algorithmic trading, in some cases creating feedback loops that impacted the underlying securities’ fundamentals. If a lot of corporations are using similar algorithms from a handful of software companies to forecast demand, and those algorithms are using similar data sets, the collective reactions will cause positive and negative feedback loops, depending on the situation. In many cases, elements of chaos will be introduced, meaning small changes to the initial conditions of the predictions will be amplified throughout the system. Economies, unlike software, move slowly; but, as industries become more and more digital, the pace of change will speed up dramatically, allowing the feedback loops to express more speedily. The silver lining to Amazon's SCOT debacle is that they were the first major retailer to adjust to the slowdown in consumer spending, leaving us with some hope that eventually digital tools will dampen outcomes, rather than amplify.

Regarding the current flood of AI snake oil: First, we should be highly skeptical of all tools (and humans!) that claim to help predict the future. Second, we should expect increasing volatility and reaction speeds across the economy, with an accompanying level of chaos and unpredictability. The antidote is to build systems with resilience and adaptability at their core. This strategy applies to companies, portfolios, and any system that has a network of interacting agents, and it should provide some level of inoculation against rising volatility. Lastly, I’ll propose one area where I think AI tools could be quite useful but which is currently lacking good, practical examples: explaining the present. Rather than predicting the future, using AI tools to explain why things are the way they are, why systems function the way they do, little of which we seem to understand, could have far more positive implications for successfully plotting a path through future uncertainty.

And, I talked about how one piece of pricing optimization software might have caused an artificial increase in rents, which in part falsely fueled the aggressive Fed rate hike policies in Algorithmic Distortion of Apartment Rents Fuels Interest Rate Hikes (#368).

As algorithms and AI increasingly take over everyday life, it’s more important than ever to understand the vulnerabilities of our own brains, a topic I covered in: Living in the Past Mitigates Chaos (#336) and the Brain’s Pessimistic Default (#363). I also recently addressed a favorite topic of mine, the intersections of magic, comedy, and cognitive bias, in the Art – and Science – of Magic Tricks and the subsequent Edit Everything (#374).

3) Insights from Interesting Characters

To close out this lookback edition of SITALWeek, below are a few links to my favorite interviews and tributes from last year. Lastly, do not miss this new, killer Bob Dylan interview in the WSJ.

The creator of Choose-Your-Own Adventure books looks back at life from 90.

SNL’s caretaker Lorne Michaels appears in this two-part podcast with Dana Carvey and David Spade.

A pair of singers: Willie Nelson’s Long Encore in the NYT and The Mountain Goats’ John Darnielle in The New Yorker.

A pair of actors: Winona Ryder in Harpers and Brad Pitt in GQ.

A pair of directors: Ben Stiller in Esquire and Francis Ford Coppola in GQ.

And, a pair of tributes: Super Dave (Bob Einstein) on HBO and Gilbert Gottfried on Peacock.

✌️-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.