AI, Mass Evolution, and Weickian Loops
The early stages of technology development around a new platform tend to be simple wrappers. Back when microprocessors emerged we saw single-board computers, simple wrappers that abstracted away some of the complexity; when personal computer operating systems arrived, some of the first applications were just wrappers on operating system functions themselves, like Norton Utilities.
We have seen this play out over and over on the Internet itself. Many of the most popular early applications were just wrappers on existing Internet and Unix applications. Mail wrapped around Unix mail; Twitter can be thought of as a variant of the Unix function “finger” for .plan files. Simple wrappers are often very popular and can be financially successful, so being a wrapper is not pejorative. Rather, it’s just that as ecosystems evolve to exploit new platforms, the preponderance of new and widely-used applications tends to evolve away from these wrappers.
Application evolution itself changed, of course, with Internet apps and services. For the first time, there was a more direct two-way conduit between what users did with applications, and what developers could see about what they were doing. This led to a more coupled system, and more rapid application development, where many features were created in reaction to real data about how products were used, as opposed to instincts, or interviews, or surveys, all of which were popular before it became possible to get that information directly.
Over time, as developers get a better sense of what is possible, and as users evolve new needs based on what they see, applications tend to change dramatically, becoming more sophisticated, and less like simple wrappers. The iterative process continues and can be slow, but the trend remains, that of abstracting away from simple wrappers toward more complex applications, fueled by user data. And, at the same, the data flows and loops more seamlessly from usage back to development, leading to more directed enhancements.
We see all of this with the emergence of a new class of AI applications, often built on top of chat AIs, generative AI, and so on. The first few attention-getting applications and services have largely been wrappers, whether around marketing (click here to build a blog post for teenagers promoting XYZ), image creation (here is a prompt to create pictures of happy skiers), or office activities, like spreadsheets and presentations (create a six-slide deck for a typical hardware product go-to-market plan).
At the same time, the feedback loops are changing. No longer asynchronous, like in the PC era, or narrowly synchronous, like in the Internet era, the feedback is now fully-duplex and low latency synchronous. You can see this in the Discord server of the popular and controversial generative AI service Midjourney, where people can see one another’s image creation prompts, and the results, and then they rapidly iterate in realtime on top of each other’s prompts, moving rapidly to fringe ideas, as well as much higher levels of sophistication and verisimilitude. You are seeing, for the first time, real-time, mass evolution, as people and groups are de facto creating increasingly complex applications right in front of you.
These two changes—the shift away from wrappers to more complex and richer applications and services, and the increasingly fully duplex nature of the evolutionary pressuring forces—changes the nature of this phase of AI app development. It is a futurist cliché, but the pace must be swifter than ever, given the ease of iterating on spoken and written language, as opposed to the formal grammars of closed programming languages and compiled applications. It is faster and more democratic, given the collapse of de facto coding costs, as well as the disappearance of just year-ago barriers to participating in this iterative process.
This leads to interesting and important questions. For example, and we can push this analogy too far, we know that in biology over-fast evolution leads to instabilities; we know that slow-evolving species tend to do better than fast-evolving ones, in part because the latter respond too readily to transient stimulus, rather than exploiting their ecological niche.
A fascinating example of this slow/fast evolutionary paradox can be seen in the ecological literature. The Lepidosauria is a subclass of reptiles, containing the orders Squamata (snakes, lizards, and amphisbaenians) and Rhynchocephalia (tuatara). Originating almost 250 million years ago, the two groups split, with the Squamata evolving slowly, but the Rhynchocepahilial evolving quickly, in biological terms. Today lizards and snakes are everywhere, but the Rhynchocephalia sub-branch has only New Zealand’s threatened tuatara left. Fast evolution left it fragile and perpetually over-adapted to conditions that kept changing.
We will likely see the same with AI apps & services, as the rapid pace of evolution, coupled with a shift away from mere wrappers, causes various local and non-local destabilization. Things will appear more transient than ever, in part because of this connected mass evolution, a souped-up variant of what sociologist Karl Weick called double-loop learning with tight coupling.
This wave has hardly begun, and the next phase of AI apps and services is coming faster than anyone expects. This will come as a surprise to people not thinking in terms of wrappers, substrates, the slow-fast paradox, and Weick-ian feedback loops, but it will be where most of the wealth creation, human flourishing—and even societal disruption—happens.
===
Reference Materials:
Midjourney is Making Fake Images Go Mainstream - The Washington Post https://www.washingtonpost.com/technology/2023/03/30/midjourney-ai-image-generation-rules/
Herrera-Flores, J.A., Elsler, A., Stubbs, T.L. and Benton, M.J. (2022), Slow and Fast Evolutionary Rates in the History of Lepidosaurs. Paleontology, 65: e12579. https://doi.org/10.1111/pala.12579
Weick, K. E. (1969). The Social Psychology of Organizing. United Kingdom: Addison-Wesley Publishing Company.
“Society's Technical Debt and Software's Gutenberg Moment,” SK Ventures.
