Technological Lock-ins & Future Gazing
Which tech tree are we on? And where could we have gone?
Sometimes essays should be short and sweet. We’re experimenting with a new style of essay for the next few months, one that is light and meandering in its scope. This series will act as breadcrumbs for future, longer pieces.
In 1989, the economist Brian Arthur asked a simple question:
To what degree might the actual economy be locked-in to inferior technology paths?
Brian did not have a clear answer to this question. Studying the history of technology we've often wondered if there were situations where we were locked out of a superior technological path because of 'externalities' — other dimensions that we never accounted for.
Brian uses the example of the U.S. nuclear industry. An overwhelming majority of reactors in the U.S today are Light Water Reactors (LWRs). This design was adapted from a compact reactor used in the USS Nautilus — the first nuclear powered submarine. A series of circumstances1 — such as the Navy's role in early construction contracts, political convenience and personal influences favored LWRs.
Soon LWRs dominated as learning and construction experience caused lock-ins.
This brings up a natural question: Are we living in one of these moments right now?
Consider deep learning today. There has been a lot of research and investment done on the “bigger is better” race regarding deep learning architectures and GPU requirements. This domain specialization has created important efficiency gains, however it makes it “even more costly to stray off of the beaten path of research ideas”2.
Our consilient hypothesis is that new developments here may require a completely different combination of algorithm, hardware and software.
But this demands a new way of looking at the field; unrestricted by the availability of current hardware limitations. This may also require looking back in time and figuring out early research branches that were ‘pruned off’ for one reason or another3. Ideas that were considered ‘ahead of their time’.
Palmer Lucky (@PalmerLuckey) , founder of Anduril had something similar to say in a recent appearance on Invest Like the Best:
“If I want to understand what's going on in the modern day, you want to go back to the future and say, what were people saying back then, what are the ideas that people aren't even discussing right now?”
Going down a historic deep dive, and future gazing from that vantage point might lead to some very interesting clues!
Light water: How the Nuclear Dream dissolved (https://www.osti.gov/biblio/6235611)
The Hardware Lottery - August 2020 (https://hardwarelottery.github.io/)
In their survey paper for Collective Intelligence in Deep Learning, David Ha and Yujin Tang mention Cellular Neural Networks (CeNNs) as a technology that became a subfield of AI Research in the mid-2000s, but then disappeared from the scene. https://arxiv.org/abs/2111.14377