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: a wave of productivity gains for white collar jobs is arriving, and it’s unlike anything we’ve seen in prior technology cycles; chatbots are already matching the proficiency of doctors for answering medical questions; the potential for a new class of weight loss drugs to have far reaching impacts on the healthcare system – and perhaps even the food supply chain; demographics are driving a voter tipping point; an essay on the shift from overwhelmingly positive feedback loops to frustratingly stubborn negative feedback across the economy; the magic of Juan Tamariz; and, much more below…

Stuff about Innovation and Technology

DoctorGPT

Chatbots could soon become powerful healthcare tools. A recent arXiv paper from DeepMind/Google demonstrated that large language models (LLMs, or chatbots as I prefer to call them), such as Med-PaLM (a derivation of Google’s PaLM), can give correct answers to medical questions over 90% of the time, which is comparable to clinicians (PDF, p. 3). A large part of the current patient-doctor relationship is Q&A regarding symptoms, and AI chatbots are all about answering questions and making associations. As I’ve noted for quite some time, I think chatbots are the future of all of our digital interactions and devices, and they will increasingly be our medical consultants too. And, their health analysis won’t be limited to just patient questions. For example, ChatGPT from OpenAI is able to screen for signs of Alzheimer’s disease by noticing unusual speech patterns. Eventually, AI should also be able to incorporate data from health wearables into their recommendation algorithms. The ultimate goal in healthcare is to shift from treatment to prevention, and it appears AI could become instrumental in identifying a variety potential/early-stage diseases before they become entrenched. The combined effect of all of these new AI tools could be heavily deflationary on the healthcare system as diseases are prevented or treated far earlier.

Productivity Tsunami

Anecdotally, doctors are using ChatGPT to engage with insurance companies, and I have no doubt we’ll soon have doctor chatbots conversing with insurance chatbots to resolve claims or treatment recommendations. Beyond the potential to handle routine tasks to shift human focus to higher value objectives, I’ve been thinking a lot about the productivity gains from chatbots and generative AI. For example, some programmers claim AI tools have doubled their coding output while producing superior code. Think about what other types of jobs stand to see such gains. Most office jobs are largely concerned with answering questions and/or moving through a series of repetitive workflows. All of this information handling – including real-time problem solving – is now easily in the domain of AI. Could we see a doubling of productivity across nearly every information-based job? It’s such early days, yet the results are so promising, that I am willing to venture into the extremely dangerous territory of making predictions – and declare that we just might see massive productivity increases from chatbots and generative AI unlike anything we have yet seen over the course of the Information Age – outweighing even PCs, smartphones, and the Internet. I hate the expression “buckle up”, but it might be called for here.

Chatbots in Schools: Tool not Threat

The recent decision by the NYC education department to block ChatGPT use at schools is akin to them blocking access to the library or Internet. Tools are tools. They can be used inappropriately, but the potential for chatbots to dramatically improve education far outweighs potential negatives from cheating. Further, ChatGPT is an extremely powerful tool for teachers to leverage, as Professor Ethan Mollick pointed out here. Mollick has more examples of how these tools should be embraced rather than feared in the education setting. Given that Microsoft has plans to integrate ChatGPT into Word and other programs (something we predicted was the next logical step in our post titled Artificial Homework, suggesting it would give LLMs a major boost in training), does the NYC education department just plan on eliminating computer use from the curriculum? Chatbots may not replace teachers, but they will augment both the teaching and learning process.

Improving Chips for LLM Training

AMD CEO Lisa Su recently touted the 8x increase in performance and 5x increase in efficiency of the company’s latest AI chip at her CES keynote address: “MI300 delivers 8x more performance and 5x better efficiency than our MI250X which was already powering the world's fastest supercomputer. And let me tell you what this means. MI300 can train much larger AI models faster, at lower cost, and with less power. And just to put this in perspective, over the holidays, there's been a lot of talk about ChatGPT and what you can do with these large language models. What you probably didn't know is that it takes months to train on thousands of GPUs that consume millions of dollars of electricity. MI300 can reduce the time to train these models from months to weeks with dramatically lower energy costs. And, more importantly, it can also support much, much larger models that can be used for even more advanced and more powerful AI services in the future.” Chip makers are known for grand claims about next-gen chips, but if we do assume a ~5x efficiency gain to train LLMs every ~18 months, we might get to a point 2-3 generations from now where it’s economical to train models daily, with wildly unpredictable ramifications for that speed of learning. (GPT-3, for example, was trained in early 2020 at a significant cost over the course of several months; within a few years, that could drop to 1-2 days at a cost of maybe single-digit millions of dollars, implying a few billion dollars a year, which is probably less than what Google spends to maintain its search engine; note: I am speculating on most of these numbers to provide a rough framework.)

Miscellaneous Stuff

Magical Maestro

If you liked my bit on how the art of magic helps us make better decisions, referencing the incredible performance of Spanish magician Dani DaOrtiz, then you’ll enjoy this NYT profile on Juan Tamariz, the 80-year-old Spanish maestro who teaches close-up magic to many of the world’s greatest performers, including DaOrtiz.

“Going out to dinner with Juan Tamariz in Madrid is a little like accompanying a cartoon character on a journey to the real world. As I walked with the 80-year-old magician on side streets off the city center’s main drag, the Calle Gran Vía, heads turned left and right. Tamariz has been a professional magician for 52 years, and in that time, he has managed the singular feat of becoming both a household name in his home country and a living legend in magic everywhere…

In the 1970s, Tamariz decided that magic needed an established school of thought, like the French surrealist movement, and composed a manifesto. It became the founding document of the Escuela Mágica de Madrid, a collective dedicated to the advancement of their craft. If the group modeled itself on an artistic movement, it operated much like a research laboratory: The magicians conducted clinical trials, gathering spectators to witness their performances and soliciting feedback, and produced a peer-reviewed journal, the Circular.”

The Impact of Eating Less on Food Supply Chains and Healthcare

Our brains and hormones have been hijacked by unhealthy foods along with the advertising and supply chains that create the delicious threats to our health. The apparent ongoing success of the new class of drugs addressing type 2 diabetes and obesity by targeting the appetite and satiety pathways creates an interesting thought exercise: can you positively impact a broken system in reverse order? The drugs in question are the glucagon-like peptide-1 (GLP-1) receptor agonists from Novo Nordisk and Eli Lilly that work by making people feel more full, mimicking the body’s natural signals to stop eating and eat less overall. Across clinical trials, they have reduced body mass by roughly 15-20%, and several drug variants have received FDA approval for type 2 diabetes and weight loss treatment for adults and teenagers. With respect to people looking simply to lose some weight, there are even online Rx apps for the mostly injection-based medicines.

The broken system I referred to is of course the government-subsidized mono-crop industrial farming complex (and the restaurants and food brands that it feeds), whose products are transformed into cheap calories of low quality (high sugar, high saturated fat, low complex carbohydrates, low micronutrients, etc.), combined with our woefully misaligned healthcare system. The entire system fuels fast food, unhealthy options, and, in particular, takes advantage of lower income individuals who cannot afford to buy healthier ingredients and/or don’t have time to prepare meals at home. Ideally, the government could remove subsidies that favor low-quality foods, incentivize healthier choices (e.g., targeted food vouchers, broadly offer high-quality school lunches), and improve the healthcare system to focus on prevention rather than just treatment. But, frankly, impacting the supply-side of the food chain seems like an impossible task. And, the US health insurance system – which is incentivized to keep patients sicker for longer so that they can make ever larger profits with diagnostic tests and symptomatic treatment – appears equally impervious to disruption. So far, insurance companies are not covering the costs of these drugs when prescribed for weight loss, which can run over $1,000 per month (it’s as yet unclear if their usage is required to be ongoing or can be stopped after weight loss goals have been met).

As I noted nearly two years ago in #296, when first discussing this new class of drugs, approximately 8% of healthcare costs were attributed to obesity, and 26% overall were due to lifestyle-related diseases. Does inverting the solution by solving it from the demand side rather than the supply side have potential to overhaul the entire healthcare and industrial-food systems? There are an estimated 37M diabetics in the US, but there are closer to 100M obese people. If an appetite curtailment drug was used widely enough, could it create a measurable decrease in food demand and drive a positive feedback loop of overall health improvements? Would the fast food industry have to shrink their advertising and change their menus? Would even a modest decline in demand for unhealthy foods and snacks ultimately make its way back to the industrial farming system? Would the healthcare system, faced with a sudden onslaught of millions of heathier clients, be forced to rethink how it makes money? I try to defend against cynicism, but I am skeptical that any of what I just described could ever happen. The cynic would say it’s more likely that we would see even bigger farming subsidies to offset the apparent plight of healthier eating; and, I’m not holding my breath for insurance companies to start doing right by their patients. Regardless, if there is even a small health revolution from a class of drugs that might reverse the hijacking of our hormones/metabolic system by the agribusiness industry, that alone would be a huge win.

Stuff about Geopolitics, Economics, and the Finance Industry

Political Demographic Tipping Point

In #372, I noted that there was evidence for a demographically driven voter tipping point – moving from a conservative to progressive majority – as discussed in this Big Think article. The FT has some very interesting data on the trajectory of Millennial voters. Typically, as people age, they become more conservative. However, Millennials, perhaps because of the repeated economic crises of the 21st century, high student debt levels, unaffordable housing, etc., are becoming more progressive. The data include UK and US populations, and, if you look at the charts, you can see another interesting trend: Gen Xers in their 40s in the US are unexpectedly becoming more progressive as well. If these trends hold, we should expect a decades-long shift to progressive politics and agendas, which is likely to include ongoing social engineering efforts (such as government stipends for childcare, which I referenced last week) and broader infrastructure spending to support the green economic transition.

When Positive and Negative Feedback Loops Collide

One of the ways that complex adaptive systems teach us to envision the world is through the ongoing opposition of positive and negative feedback loops. Positive feedback cycles are the self-reinforcing attributes whereby growth begets more growth (e.g., network effects), while negative feedback is the stubborn, real-world challenges that offer resistance to unbounded growth. This push-pull scenario is very common when industries go from analog to digital. Take EVs for example: there is a steady growth in demand for electric vehicles; however, people only buy new cars when they need them, and charging infrastructure, battery range, cost, lithium mining/refining capacity, and form factor requirements (sedan vs. SUV vs. truck vs. minivan etc.) are all real-world needs that push back on the obvious reasons for EV adoption. In our 2014 paper Complexity Investing we wrote:

In nature, we see positive and negative feedback loops with regularity. For example, the pine beetle ravaging the forests of the Rocky Mountains represents a classic positive feedback loop. Due to the loss of extended cold winters (which normally act as the negative feedback loop), pine beetles find their growth unchecked. They will continue to prey on susceptible pine trees until there is literally no more food left. Then their population growth will come to a crashing halt. We see something similar happening with the invasion of non-native Burmese pythons in the Everglades. Their inclusion at the top of the food chain has significant nonlinear implications for the ecosystem. As python numbers have grown, wildlife sightings have fallen some 90%.

In the world and in companies, we observe the same thing. Positive feedback sets things in motion through self-reinforcement, while negative feedback ensures stability against disruptions and excesses. We’d argue that when a company comes into a large, existing market with a disruptive product or business model, it’s very similar to someone releasing a non-native Burmese python into the Everglades: a new variable in a complex system changes the nature of the overall system in a nonlinear fashion. Sometimes there’s no negative feedback loop to check the new variable’s growth, which leads to hyper growth and flame out. Sometimes hyper growth can go on for a VERY long time because the opportunity is so vast.

We can apply this idea of tension between positive and negative feedback to various industries that have been impacted by technological disruption over the last two to three decades. Amazon is a classic example of a company that has benefitted from positive feedback loops as they grew their market share of retail sales, their Prime customer base, their marketplace of third-party sellers, their advertising business, Prime Video, and warehousing/delivery logistics including fulfillment by Amazon. The advent of smartphones, 5G, and consumer behavior changes all came together, driving ecommerce to a high-teens percent of overall retail sales in the US (and even higher in other countries). Some retail categories have become digital at well over 50%, while others, like groceries, have remained quite low. However, in hindsight, this growth has actually been rather slow thanks to negative feedback, garnering less market share than one might expect. And, overall, ecommerce is only gaining about one percentage point a year (on trendline after a pandemic acceleration and mean reversion). Real-world issues of shipping capacity, customer behavior, supply chains, etc. are all factors that keep ecommerce from experiencing unbounded growth. Inside of ecommerce, the positive feedback engines I referenced previously, like Prime, have created less resistance for Amazon than other retailers.

Positive feedback cycles can be supercharged by a variety of factors. The disinflationary era of globalization and tech-driven productivity over the last few decades provided an easy-money and low-interest-rate backdrop – classic lighter fluid for unchecked growth. The pandemic-fostered excessive fiscal stimulus was like directing a flamethrower at the lighter fluid. Then the music stopped. Those decades of positive feedback loops – the engines of digital transformation – are still there, but the negative feedback loops have been awakened from their slumber and are feeding on high rates, economic slowdowns, and sticky consumer preferences/behavior. Much of the low-hanging fruit across several industries was plucked at an accelerated pace over the last few years, and the next wave of analog-to-digital transformation in the economy is likely to be significantly more challenging.

There will be new positive feedback engines – like AI – that run unencumbered and have the potential to greatly amplify volatility across the economy. But, the positive feedback sources that we’ve become accustomed to over the last couple of decades (such as the digital transformation of advertising, media, and retail) might have gone too far too fast, and may even unwind (see #376 and the de-powerlawing discussion in #377 for more). To have some idea of where capital will be increasingly allocated across the economy in the future, we need to identify the next set of positive feedback loops – where the overwhelming benefits of disruption will overcome the resistance of the deep grooves of the real world’s behavior and habits. As discussed throughout the past year (e.g., see #377), the LLMs and transformer models enabling chatbots appear to be probable engines of largely unbounded growth. AI has the potential to double the productivity of white collar jobs, and automation and robotics are increasingly coming to disrupt the labor market. The latter is likely to experience much stronger headwinds given the greater challenge of replacing human physical labor compared to brain power. In general, the more digital the activity, the more likely positive feedback will overwhelm, while the more analog, the more likely we are to see negative feedback impinge growth. Alternatively, we can look for the trends enabled, but not necessarily transformed, by disruption. For example, in retail, while the ecommerce transition might be slowing down, brick-and-mortar stores adopting digital technology might be just beginning. In some sense, we are moving from the primary function to the first and second derivatives of the digital transition of the economy, i.e., from the locomotives of change to the companies that leverage the benefits of having railroads. This slower, more distributed transition will benefit from bounded positive feedback loops that can create benefits for a long period of time.

However, this next set of positive feedback engines will be messier and slower to establish than the unbounded, runaway growth of the Internet and the cloud. They will require hard decisions and hard work, i.e., they are heart transplants rather than Botox injections. The recent meltdown of Southwest Airlines, resulting in cancellation of ~70% of holiday flights, is one such example. The challenging task of moving a decades-old software tool off legacy mainframe products and into the cloud/mobile was kicked down the road one too many times, and their colossal failure has now forced existential change. Take another example: IEEE estimates the US needs one million new miles of electrical transmission cables to achieve a carbon neutral electrical grid by 2050; however, we currently only install around 1,000 miles a year. How could we add an order of magnitude more cable without people, permitting, materials, etc.?

While it might appear that we have reached a point of stability between positive and negative feedback forces after decades of supercharged growth, we know from complex adaptive systems that the only equilibrium is disequilibrium. Therefore, we should hold any predictions very lightly. Since the low-hanging fruit of the digital disruption have been consumed in a binge of low interest rates, global trade, and tech-driven productivity, we may have prevailing headwinds for the next several years. There will no doubt be surprising new areas of positive feedback loops, but, as complex industries – like healthcare, financial services, and energy – experience technological disruption, the effort to engage them will be far greater than what was necessary for media, advertising, and ecommerce.

I’m fond of the expression: pessimism sounds smart, but optimism wins in the long run. Between pessimism, which assumes the worst, and optimism, which assumes the best, lies skepticism. In positive feedback cycles supercharged by things like low interest rates or deflationary trends, skepticism is less valued because those megatrends tend to always support the optimistic outcomes. However, when negative feedback loops become stronger, skepticism can be a more valued framework. That said, in eras of constricted growth, pessimists may sound even smarter – but be no less wrong about the future – and there is still ample room for boundless optimism.

✌️-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.