Life often presents a choice: settle for a reward now or wait and explore for something better.
These decisions are rarely easy—they challenge the boundaries of our instinct and intellect.
This time, I found myself wrestling with a fundamental choice: should I go for the locally optimal reward or hold out for the global optimum?
Choices are always tough.
Choices are meant to be tough. Had they not been tough, then they would not even have reached my level of cognition. My inbuilt neural and muscular pathways would have kicked in already to deal with it.
Anyways...
Here I was, trying to figure out what the right call was.
To answer this, I wondered if there is a general path to progress. Does humanity optimize locally or aim for global optimum?
If local, then what is the point at which we draw the line between local and global? I have read and seen the Marshmallow Experiment, in which small kids are given a choice between having one marshmallow now or delaying it by 15 minutes (or 10 or 5, depending on the experiment) and getting two marshmallows.
Delayed gratification is widely celebrated as a marker of future success, especially by those of us who are reading this blog. We have always been delaying gratification for the next time. But this experiment doesn’t answer a more practical question: how long is too long to wait? Is 15 minutes the right delay? What about a day? A year? A decade?
For the kids in this experiment, it was clear. Wait for 15 minutes, and your reward doubles.
Often, for the choices we have to make, we have to take a bold step in the face of uncertainty, with no one to tell us hey kid, wait for 1 year, and your reward will double!
I started with one question: whether to get a locally optimal reward or wait and explore until I hit the global optimum. Now I have two: local vs global, and what is local, really?
Well, this is what happens when I start to get into a rabbit hole.
Anyways, let's try to answer the first question. We can learn something from how humanity and nature have managed progress.
Human innovation is a story of a series of locally optimal moves, each one solving the immediate problem at hand, while never really answering "what is the big picture".
Consider the evolution of communication systems. The journey from smoke signals to smartphones was not a single leap but a series of incremental improvements:
- Smoke signals to messengers: Humans faced the problem of sharing information across distances. Messengers provided a reliable method for transmitting messages over long distances, albeit with significant delays. However, messengers were not always faithful to one's messages, often introducing their creative reinterpretation in between.
- Messengers to handwritten messages: The challenge of preserving information across time was solved by written communication. However, this introduced the need for literacy and reliable transportation networks. At the same time, communication systems needed an expansion in their distribution capacity. One could not communicate with 100 people by writing them messages. One can hire only so many scribes.
- Handwritten messages to the Printing press: Mass production of written content democratized knowledge, solving the problem of limited distribution. Yet, it still lacked instant accessibility. Printing even popular books like the bible was highly costly.
- Printing Press to the Internet: Instant global sharing of information addressed accessibility but brought new challenges like information overload and cybersecurity concerns.
- Now to AI-assisted communication: Personalization and efficiency improved dramatically, but ethical dilemmas and dependency on algorithms emerged as new issues.
Transportation follows a similar trajectory:
- Walking to animal domestication: The need for faster and less tiring travel led to the domestication of animals, but this introduced dependencies on their care and resources. Also, long-distance at this time meant 10s of kilometers per day. There was a dire need to expand this 10 to 100 to collect more food and trade with other humans.
- Sailing ships: Long-distance trade and travel were enabled, yet navigation and weather posed significant risks. Further, ships had a big problem; they could only travel on big bodies of water.
- Steamships and trains: Reliability and speed improved, road transportation over longer distances was enabled, addressing earlier limitations, but the infrastructure required (rain tracks, factories, steel) was costly and complex. Further, train tracks can't be built from every place to every other place; trains were limited to "important" activities, activities that were important for the masses. Personal travel still needed solving.
- Combustion-engine vehicles: Personal and faster transport became accessible but it increased reliance on fossil fuels and environmental concerns.
- Electric vehicles: Reduced emissions provided a partial solution to environmental issues, yet challenges like charging infrastructure and battery disposal remain.
Each step solved immediate problems while creating new ones, driving continued innovation.
All innovations worked for certain time horizons. They did not solve the problem for eternity. For a few, it was 1000s of years (messengers, written messages, travel via animals); for a few, it was 100s (sailing ships, trains, printing press); for others, it was 10s (combustion cars, internet).
Several trends are noticeable: solutions tend to be localized, addressing immediate problems rather than long-term challenges. The lifespan of solutions is shrinking—in other words, the pace of innovation is accelerating.
One thing is clear: local optimization may not be globally optimal, but it’s far from futile. Incremental progress addresses immediate needs and lays the foundation for future innovation. No one can predict the distant future with certainty, making it impossible to globally optimize from the outset. We can only extrapolate from current trends and adapt as new challenges emerge.
I began with one question: should I go for the locally optimal reward or hold out for the global optimum? Along the way, I found myself asking another: what defines “local” in this context?
While the second question remains unanswered for now, one lesson stands out: local optimization is not the enemy of progress. It’s a stepping stone, not a roadblock, on the path to global solutions.
That’s all for this entry, folks! We’ll dive into the second question—“what is local, really?”—some other time. Until then, goodbye!