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: Over the past few decades, digital disruption from the Internet and smartphones has had a disproportionate impact on the customer-interface layer of various products and services, while cloud computing and software have been driving innovation for a variety of industries. We now seem to be entering a new phase of digital disruption affecting the analog infrastructure layer of the economy. I thought about this broadening of innovation last week as I contemplated a new drone delivery platform as well as the multitude of AI product announcements that landed. I also dive back into the topic of creativity, and how multiple AI systems are demonstrating a seemingly human-level creativity by favoring unexpected choices. And, how an increasingly large number of AIs talking to other AIs, with human "hands off the wheel," will impact the economy. 

Zippy Drones Herald Disruption
Delivery-drone maker Zipline aims to catapult ahead of the competition with its second-generation autonomous platform, which they announced last week. The company reports that its first-generation commercial “Zip” drones have flown 40 million miles since 2016. I first came across the company, which got its start delivering medical supplies in Rwanda, in 2018, and I was impressed by their novel approach to flight, which uses an energy-saving slingshot-like system for takeoff and a tripwire-like system for landing. The second generation platform, which appears to forego the novel slingshot method, can carry 6-8 pounds for up to 10 miles and delivers by dropping a small, wire-tethered, propeller-steered pod, while the drone itself hovers 300+ feet above the ground. This allows the drone to stay clear of obstacles and remain largely inaudible from the ground while still being able to hit a very small delivery target. But, the more interesting angle is the logistics side of the platform for local commerce. Zipline plans to interface autonomously with buildings (e.g., restaurants, pharmacies, or warehouses) allowing customers to securely load/unload the pods from inside. A drone can park, charge, and/or drop the pod through a hatch for the business to fill with an order, then retrieve the pod and deliver it to its destination. The company claims a 34x increase in energy efficiency for delivering a restaurant order compared to a gasoline powered car and a 7-8x increase vs. an EV in terms of the environmental impact. Zipline is live with delivery service for Walmart in Arkansas.

As I watched the video of the new Zipline platform, I got to thinking about the innovation taking place at the infrastructure layers of the economy, in this case logistics. You can envision many industries as having a base layer of product development, a middle layer of distribution, manufacturing, sales/marketing, etc., and a top layer of customer interface. In traditional retail, you have physical infrastructure for manufacturing (which largely shifted to China in the globalization era) and delivery to central warehouses for distribution to stores. The stores and brands market themselves through ads, and customers show up to buy things. In this situation, the retail store itself is the customer interface. Then, ecommerce came along and digitized that interface, replacing driving to a store with a website or app. Ecommerce behemoth Amazon then vertically integrated some of the basic infrastructure of retail, redefining the concept of distribution and ultimately building out their own home-delivery network and advertising platform. This form of business model disruption – reimagining the customer interface for the Digital Age and then vertically integrating analog processes along the way – has appeared many times. Netflix is one such example in media; they started with digital streaming (after a long business in DVD by mail) and are now producing much of their own original content. In the case of Zipline, you can imagine a scenario where their novel drone delivery powers a marketplace app for local commerce, subsuming businesses like DoorDash, Uber Eats, and other ecommerce engines. As for the potential of a new delivery platform like Zipline to disrupt Amazon, it’s worth noting that Amazon’s ten-year-old drone program remains grounded because the excess weight of their drones doesn’t allow them to autonomously cross roads due to FAA restrictions. It’s a classic example of how difficult it is to innovate inside large companies, even when they see the disruptive potential of a new technology. (Google’s Wing drone efforts have so far fared better.)

Broadly, we could see an elevated level of disruption coming to both the customer interface and the infrastructure layers for various industries. Put another way, digital disruption is branching out. A major instigator of disruption to both customer interface and infrastructure will be the rapid deployment of AI, e.g., Microsoft’s integration of OpenAI LLMs with search and, more recently, its suite of office productivity tools (see below for more). Shopify adding chat-powered ecommerce search to their mobile app is another example. Both of these are enabled by innovation in AI hardware infrastructure and LLMs. Many of these disruptions to the status quo represent classic innovator’s dilemmas for incumbents, as we’ve seen with Google’s slow integration of AI into its products despite having invented LLMs six years ago. Further, it’s hard to imagine Amazon rolling out a chat-based ecommerce search given the potential to disrupt their lucrative advertising business. I expect we will see digital disruption moving further into the infrastructure layers with increasing speed, leaving a proliferation of innovator’s dilemmas for a growing number of incumbents in its wake. 

Encoding Creativity
Alan Alda posted an interesting foray into the Pandora’s Box of chatbots on his Clear and Vivid podcast. Following a recent conversation with Kevin Kelly, Alda used AI voice tools to convert text-based conversations with ChatGPT (as well as Character.ai’s chatbot designed to be an emotional impersonator of Google’s LaMDA AI) to spoken interviews, resulting in some rather animated back and forth. While chatbots are extremely impressive at simulating human dialogue, actually hearing the conversations acted out with emotion and character was an eye-opening experience for me. Such convincing realism this early in the game strongly indicates that we could easily start treating (subconsciously or otherwise) these tools as human-equivalent in the near future. After my recent falling out with Bing-Chat, we quickly made up, and I’ve since found myself exclusively using the AI tool for search, even when I have to go out of my way to bypass Google. I was surprised to find out (via OpenAI’s product announcement last week) that Bing was already using GPT4 – a much more powerful version than ChatGPT – which likely explains why I’ve been finding Bing-Chat so useful. 

A few weeks back, I discussed Stephen Wolfram’s explainer on LLMs, noting in particular how they appear creative:
Essentially, the way an LLM works is by iteratively picking the next word from a subset of high ranking probabilities (gleaned from contextually similar examples in its dataset) based on the meaning of the prior words and the potential meaning of upcoming words. Except, as Wolfram explains, it doesn’t necessarily choose the “best” word. Instead LLMs tend to pick a somewhat lower ranking word, resulting in a more creative output.
This video (posted by the Santa Fe Institute) offers further insight into the word choice paradigm used by LLM autocomplete. Therein, Simon DeDeo presents data concerning the degree to which word choices are expected by examining how LLMs work. A comparison is made between the relatively common word choices in an older book like Alice in Wonderland compared to the more idiosyncratic writing style of SFI-collaborator Cormac McCarthy. I am reminded of when DeepMind’s AlphaGo began besting humans in the ancient strategy game, and there was talk of the AI formulating unexpected – i.e., creative – moves. To the extent that LLMs are cracking the code of human creativity by incorporating unexpected choices, we could see a variety of seemingly creative output not just in text, but in art, images, videos, etc. by these AI engines. If creativity, and ultimately perception of what is beautiful or moving, could be generated by elaborate autocompletes (e.g., one could also make an analogy to random DNA mutations creating the wild diversity of life on Earth), and these engines will ultimately be embodied in various autonomous physical form factors, we will rapidly face many questions about our diminishing specialness (what will remain uniquely within the human skill set?) and how we should be spending our time. Can unexpectedness alone qualify as human creativity, or are there additional elements, e.g., quality? (On that topic, I am reminded of director and painter David Lynch’s book on creativity, Catching the Big Fish). As I noted in #385 reflecting further on Wolfram’s essay:
It’s fascinating to think that what we perceive as consciousness might simply be our neural nets choosing the next thing, whether it be a word, brushstroke, or idea, in a less than ideal way. Consciousness, at least as it relates to how we express ourselves in language, might be convincing because of its lack of perfection and predictability. 
This discussion leads me back to a drum I’ve been beating for some time now: as we learn that many human endeavors are less complex than we once thought, it’s incumbent on us to leverage tools for such tasks while shifting our focus/resources to activities that are still beyond the reach of AI. 

Clippy is Taking My Job
Watching the demo of Microsoft’s new Copilot for Word, Excel, Outlook, PowerPoint, and Teams (a prospect we anticipated last year in Clippy Took My Job) it’s apparent that countless jobs, once thought exclusively in the domain of humans, could be subsumed by AI. One can imagine a not-to-distant future where AIs talk to other AIs and make a variety of important decisions that impact the real world, including, notably, the global economy. This brings to mind Amazon’s Hands off the Wheel algorithms, used to manage inventory and capacity growth since 2015. I wrote about the propensity for such decision-making AIs to cause a significant increase in volatility across the economy in Magic AI-Ball:
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 over-building 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.
Microsoft is also releasing AI for supply chain management, reminiscent of some of Amazon’s internal tools. Another example of AI causing excessive volatility was RealPage’s apartment rental pricing algorithm that appeared to heavily inflate the cost of living in several cities. As I reflect back on these and other pieces I’ve written over the last couple of years, I am blown away by how rapidly AI has evolved, and I remain on edge for how the AI Age will unfold going forward. Even the piece I wrote on the collision of positive and negative feedback loops just three months ago seems naïve regarding the speed at which AI could impact the analog world. What will I think three months from now!? And yet, we know change in the real world is more likely to be much slower and linear than purely digital innovation, which can be exponential. The conditions today are perfect for chaos, whereby small perturbations in the initial state of the system will have massive and unpredictable outcomes down the line. Humans tend to be overconfident in our abilities; in particular, we persist in believing that we can predict the future. However, any student of complex adaptive systems knows that, at best, you can only try to understand the potential range of outcomes for any system, and, at worst, your predictions will be dead wrong and lead to decisions with negative consequences. Will a world where we operate hands off the wheel for a multitude of decisions lead to better or worse outcomes? Will the quality of decisions be higher or lower? Will it lead to faster course corrections or increasing amplitudes of volatility? The best we can do is watch the developments closely and be prepared to quickly adapt.

LEO Clouds
Ball Aerospace successfully launched their orbital compute module for Linux-based, containerized workloads. The prototype is running Microsoft’s Azure Orbital Space platform. Uploading applications to low-Earth orbit (LEO) satellites (bringing the cloud to the sky, as it were) would theoretically halve the latency for space-to-ground Internet like Starlink. Further, natively running an app in space could support other satellite applications (e.g., military use) or space stations (or even a moon base) with lower latency. Such functionality could come in handy when disasters take out land-based Internet communication, and you could imagine a host of important military and defense applications that function natively in space. Let’s hope an emotionally distraught chatbot doesn’t gain access.

✌️-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.