Three Short Pieces
Analogies between corporate and human life cycles, an example of third-order thinking, and a common misunderstanding of evolution (with a discussion of machine learning!)
For the next little while, I’m trying something a bit different. Rather than one long, digressive post every month or so, I’m going to be writing shorter pieces more often. That’s a regular OR, not an XOR statement, by the way: there will still be long, wildly all over the place essays. They’ll just be tall trees separated by clumbs of shrubs and new growth. To drop that arboreal metaphor, you’ll likely see the ideas from the short pieces incorporated into the long ones.
What I’m Working On:
Studying LangChain’s documentation, with the goal of building a document reading app that will engage the user in creative, exploratory conversations. Other sites are charging for this service, but I think I can build one for myself that costs less and works the way I want it to. The goal is to have a way to improve my studying and learning - read the book, chat with the book, get the book to make its own flash cards.
Relatedly, I am reading and enjoying Ray Dalio's Principles: Life and Work. I'll have some thoughts about it and reflections to share in future writing, perhaps to demonstrate how my new app can talk to documents.
Working on a natural language to SQL workflow using LangChain (again). The idea is to help myself and my work team more rapidly iterate scripts for common customer requested data fixes. To quote a programmer: "I would rather write code to help me write code, than write code.” I have read a few pieces on impressive results achieved with this, but they’re mostly on toy databases rather than production ones. I want to create one that is general purpose and has a decent UI for ease of use.
Building creative prompts with OpenAI’s Playground. Try the first one, a Problem Restatement prompt, here [requires OpenAI account]. My ultimate, hazy goal is a web app to integrate creativity tools: problem definition exercises, the Phoenix Checklist, brainstorming techniques, mind maps, complemented with image generation and a chatty, creative partner to help you get the most out of your ideas. Very much Work In Progress
Studying Optimal Stopping Theory, and its associated problems. More on this later when I’ve wrapped my head around it and its potential applications.
Life Cycles, Corporate and Human: Bidirectional Analogies
I’ve been playing around these last few days with the analogies between corporate and human life cycles. Corporations can be thought of as having distinct life stages:
Birth / Startup - a company is founded and investors and its founders work hard and pour in resources of time and capital to grow the business and develop its main products.
Youth / Growth - with the product proven, the company enters the hustle and bustle of a competitive marketplace, now (hopefully) producing some ROI, but still focusing effort and capital into growing market share.
Middle Age / Maturity - the company at its peak. Its no longer a young upstart or innovator, but it is a force to be reckoned with in its industry, a company that has to be taken into account in strategic thinking by its competitors. The money is rolling in, and the investors are rewarded for their patience with dividends.
Senescence / Decline - the company in its declining years. Market share and profits are shrinking, and the focus of management is on returning maximum value to shareholders - selling off assets, spinning of subsidiaries, and generally trying to get as much value out of what remains as possible.
The idea of a corporate life cycle owes its origin, obviously, to biological ones. We map the stages of a human life onto a corporation. Of course, just like biological lives, lots of corporations die at the startup and growth stages - thankfully, this is becoming a disanology between corporations and humans as health care and economic conditions keep improving.
Now let's fip the analogy and make one not between human life and corporations (humans → corporations) but from corporations to human life (corporations → humans).
Startup - birth, infancy, childhood, probably the period of maximum outside investment in the new person’s life. Startup incubators (homes), venture capitalists (parents), and institutions (schools) have an interest in seeing the new venture grow and succeed.
Growth - more schooling, though at this point the new venture is expected to start putting in some effort of its own. The learning from teachers is less explicit, more and more is found out through trial and error, through experimentation and adaptation. Early career is included here too: finding out the market niche for your talents, skills, and personality. Debt has to be controlled as well - it can fuel rapid growth, but unless that is invested in future capital generation, it’s a drag on the future profitability.
Maturity - the 40s, 50s, 60s, and (maybe, with improving healthcare) 70s, the age of maximum growth and advancement. This is where, if things are going well, a person is earning the most money, undertaking the most impactful activities (having children, starting to make decisions at their workplace as a manager rather than just following instructions). Parents, schools (through alumni associations) and society at large also start to make maximum return on investment at this stage, through grandchildren, tax revenue, productive economic contributions, and investing in the next generation, similar to how founders of venture-backed firms move into venture capital in later life.
Ironically, at this most impactful stage, there tends to be the least societal attention focused here. Magazines prefer the “Top 30 Under 30,” (interestingly, it is rare to ever see follow ups written about these people…) to today’s established family men and women. As I come nearer to this point in my life, I think this might be deliberate - the mature prefer that young people not figure out just how good life can be when you’re older.Decline - Happens to everyone fortunate and blessed enough to live long enough. The market moves on, the capital wears out, the employees look for new opportunities elsewhere. This is when, if you’ve built up anything, you get concerned about legacy - how to look after your heirs, any bequests you want to make, and perhaps imparting last words of wisdom.
This is a limited view, naturally. Many lives, like most new companies, don’t pursue this trajectory. Companies can also, like governments but unlike people, be functionally immortal - they keep living until some external force kills them. I also can’t think of any human life equivalents of stock splits, mergers, or hostile takeovers (alien brain parasites hijacking a host organism?)
Third Order Thinking
Successful planning requires imagining the consequences of our actions, as well as the consequences of those consequences. While it's easy to predict the immediate effects of eating a doughnut, understanding that it leads to obesity is a more difficult concept to grasp. It's even harder to realise that this single doughnut could be forming an unhealthy habit, which may be tough to break in future. Distant events have lower probability than near term ones. The doughnut may just be part of an indulgent snack with coffee today, and you might still go for a run later on. Or it may be a sugar nail in your Type II diabetes coffin.
Just because second and third order thinking is difficult doesn’t mean it isn’t worth trying. And the payoff can be spectacular. I’ve been listening to the audio book of Peter Robison’s Flying Blind: The 737 Max Tragedy and the Fall of Boeing. In it, Robison describes the strategic insight an Airbus executive had, at the time Airbus was planning its A320neo. There was an internal debate at the company about how much of an advance the new narrow-body plane should be over its competitor, the 737. Engineers at Airbus wanted to produce a truly cutting edge plane, one that would be a major leap over Boeing’s aging though still very popular short haul flight champion.
Airbus’ Sales Chief, John Leahy, resisted pushing through with major improvements. Leahy understood that not only were major innovations in engine design, airframe materials, and electronic controls coming later in the decade, but that because of that knowledge airlines were unlikely to buy whole new fleets of (more expensive) advanced planes, when they could instead hold out for even more advanced models in a few years.
Leahy also understood that they were not the sole actor in the market. Boeing would watch what Airbus did, and if Airbus pushed innovation too hard, the veteran aerospace company could, and would, go back to the drawing board and deliver an entirely new, future ready airplane. Airbus was not in a position to match innovation for innovation with Boeing. The Seattle-based company could, despite nearly two decades of penny-pinching cost savings and substituting clever financial engineering for brilliant aerospace engineering, could still produce world leading designs. If Boeing chose to, they could bring the major innovations forward in time.
Instead, with Leahy’s insistence, Airbus built a less ambitious, still capable narrow-body plane, one that airlines would happily buy to expand their fleets. His insight paid off: rather than spend the cash and years of development to bring out something truly innovative, Boeing chose to redesign the 737 just one more time. This had consequences for two airlines, Boeing, the global aviation industry, and 346 human beings and their families.
I’ve been trying to practice second and third order thinking myself. The best means I can think to do so is to ask the following linked questions (with paraphrases):
Who are the other players in this game? (Whose actions do we need to consider? Who is in a position to respond to what we do in a way that could change what we can do in the future?)
If we do X, what are three (at a minimum) things they could do in response? (What potential Y’s could they do to our X?)
If they do Y1, Y2, or Y3, what would that mean for us, and what could we do in return?
Evolution, Fitness Landscapes, and Technological Development
Ask someone to describe evolution, and they'll likely talk about random mutation and natural elimination and the process by which the ‘fittest’ organisms survive and reproduce. Even sophisticated and knowledgeable people can misrepresent the evolutionary process. I recall Daniel Dennett, in his Darwin’s Dangerous Idea, describing how to understand the ‘fitness function’ as something like gradient ascent over a fitness landscape: there are valleys and peaks, and a fitness function is selecting iterations of organisms for those that move, thanks to random mutations and selection, up the peaks rather than down into the valleys. Different organisms fill up the landscape with different phenotype approaches that suit different niches.
But this understanding treats the fitness landscape as fixed and rigid, overlooking the intricate interplay between genotype, phenotype, and environment that characterizes the process of evolution. In reality, organisms don't simply exist in their environment; they shape it, and in doing so, influence their own evolution.
A new organism is introduced to an environment where it did not exist before (Darwin’s finches, for example, blown by a storm to the Galapagos islands). Previously innocuous differences in their phenotypes, the product of their genotypes interaction with their previous environment, give some an advantage: some are a bit better at catching insects, some havr an advantage digging for funguses, and still others are good at breaking open tough seeds. Some don’t survive, but the others find a niche and evolve into it. Those in turn influence the evolution of the new organism, with different phenotypes finding success than those that were favored by the environment before.
It's hard to keep this kind of interplay in mind. Its just much easier to think of the natural selection process as linear rather than dynamic, just as linear functions that map from one set to another are easier to grasp than nonlinear ones that affect each other.
A Common Theme
In the process of writing, I discover what I think. The common theme here is nonlinear causation: where an effect causes not just other effects, but comes back around and affects itself. In reverse order:
Organisms evolve in an environment they themselves affect and change, and are in turn changed by the changes they make to that environment,
Second and third order thinking can help us to understand and foresee how our acts can double back on and affect us,
And it can be stimulating to feed an analogy back onto itself to see what new insights it can yield.
Now to conclude this (actually quite long) collection of three short pieces with some thoughts on machine learning advances:
Call a technology ‘recursive’ if its development feeds back into itself. That is, developing the technology makes it easier to develop the technology further:
Electricity, harnessed into production and distribution networks, made it easier to conduct electrical experiments, and produce new and improved electronics. The lightbulb, a technology that exists because of electricity, makes it easier to work into the evening, or have laboratories and buildings too large to be lit by the sun alone. So you can make more progress mote quickly.
The computer advanced computer science, which in turn advanced computers. Even beyond proving theorems on machines, or checking long proofs, computers enhance our ability to manage and manioulate information, which in turn makes it easier to build more and more powerful computers. In his wonderful The Soul of a New Machine, Tracy Kidder relates that the computer whose development he documents, the Data General Eclipse MV/8000, was probably the last computer designed ‘by hand,’ without the assistance of software. After that, computers became both powerful enough and complex enough to help and be required in their own design.
The World Wide Web advanced the Internet. Not just in making it more available, but making it easier to exchange the tips and tricks for how to design effective web applications, build stable servers, store information efficiently, and a host of other applications that feed back into continuous improvement.
The rate of growth can be incremental at first, an arithmetic growth in capabilities. But then it can reach a point of exponential improvement. Though the function is perhaps best thought of as logarithmic rather than exponential: eventually, the steam runs out and improvement slows down. In most fields, at least.
The big question now is: will Large Language Models and their associated generative machine learning models advance artificial intelligence research? Paraphrased: is this the takeoff point for rapid, recursive gains in the powers of AI systems?
Its tempting to believe that it will be: I’ve noticed a major uptick in my own productivity, purely creative as well as practical, and I can’t help but believe AI and ML researchers will see the same, or larger, boosts to their own productivity. I can imagine generative AI, particularly research summarizers, boosting the development of the models, the rapidity of their deployment, and even the design of their hardware.
I’ll have more to say about these issues, as well as updates on my projects, very soon. I plan on posting another piece next week, expanded with some links I have found useful lately, as well as my usual digressive essays. Till next time …