Learning about learning
Why some energy technologies learn quicker
Classes start next week at my university. A new semester marks an end to my extended and well enjoyed break, the return of students from across the nation and globe, and a fresh slate for students’ learning. What we are learning, whether that be inside or outside of the classroom, depends on who you ask.
Something I have been writing about recently and have come to appreciate is how people, technologies, and industries come to learn.
People
Knowledge is power as they say, but so are the skills to diagnose, research, and understand problems and collect the information on how to solve them. In the post above I describe how information transfers, which are largely free via the internet, could be priced using micro-transactions via a subscription based model. The idea would be to think of education like a heating bill, where over a period of time we use heat, and are subsequently charged based on how much we used.
On the supply side of this transaction, experts would be paid for their efforts in generating and communicating this knowledge. So rather than needing to become large enough to generate ad revenue, each instructor may serve as a miniature school or university.
“…supply of information is difficult to come by because communicating effectively is difficult. This is why experience is valuable, even though virtually all of human knowledge exists online. So a platform that connects experts to information seekers is valuable, and may even look a lot like colleges and universities.”
If there is a threat to the current method of education and information transfer, this might be it. Why register for classes at a state university (and pay thousands of dollars to do so) for education you “coulda got for $1.50 in late fees at the public library.”
And if there is a certificate that comes along with it, all the better.
Technology / Industry
They say that experience is the best teacher, and so is the same for technologies and industries. Last year I wrote a post about energy learning curves, and how producing technology becomes cheaper over time.
Also included in this post is what it means to optimize under a deterministic vs stochastic future. In short, a deterministic future is one that is decided. There are a set of inputs, and then according to a model those inputs correspond to a set of outputs that are determined exactly. Under a stochastic scenario, there may be uncertainty in some of the inputs. That is, if you are a farmer, the weather may not be as you expect it to be. And moreover, how you optimize planting your crops isn’t the same under uncertainty as the average case, even if the distribution of future scenarios is balanced. That is, there are equal number of “better” or “worse” scenarios than the average.
In the post and linked Volts podcast with Doyne Farmer (which I highly recommend!) there is gap in understanding why some technologies learn fast while others stay flat. In the post I speculated that in addition to the need to be a commodity, uncertainty played a role in understanding the value, and hence the stochastic optimization is an apt comparison to a deterministic one.
Following up on this, (Malhorta and Schmidt 2020) investigates the relationship between learning rates, complexity and customization. They find that complexity is needed in either the design of the tech or the manufacturing process in order to see high learning rates. So you might have a relatively simple device (like a solar panel) but if the manufacturing process is complex, the cost to produce tends to come down with experience of mass production.
Additionally, there may be design uncertainty. Like in the case of wind turbines operating in dessert or colder climates, where there are “platform designs” such that wind turbines going to a dessert are different than cold climates, but for a series of cold climates they are relatively mass produced.
Malhorta and Schmidt have a grid to explain the tradeoff between complexity and customization.
My speculation that uncertainty matters wasn’t directly investigated in the paper, but I think that although they are distinct concepts, the degree of customization and uncertainty aren’t too far apart. For example an oil well (which wasn’t discussed in the paper) with unknown value is inherently customized. Although the process of retrieving oil is common, how much, when, and what price to sell at, is, custom to that well.
Regardless, knowing that customization and complexity matter in learning is valuable. The results of this paper can be helpful for informing energy policy for fast learning technologies. For example if you wish to develop an export industry and bring down costs of a technology, you might try for a Type 1 tech. Germany and Spain have done this via Feed-in Tariffs for solar PV. Additionally, you might use these results to try to move some technologies like nuclear power towards something more commoditized like an SMR.
So while I enjoy the last weekend before starting classes by finishing up my move, I hope that you’ve enjoyed learning about learning. And come to appreciate that the transaction costs and search costs for learning may be expensive. But so may be the costs for not learning.
*Note if you wish to listen to the Volts podcast which discusses the paper by Malhorta and Schmidt you can find the link below*