AI Industrial Revolution

I have a dream. One day, we can stop waiting in line to get things done. That was my thought waiting at the DMV line registering my car. I looked at the office layout, there are 20 windows active and all the staff are working hard but still not enough to service all the incoming requests. Was it labor intensive to enter data into the computer and hand out some plates? And if it is, are there big optimizations missing in the office? There has to be. I am pretty sure that in another hundred years, we will be much faster than this. What would that look like? Let me take a step back, and look at how humans unlocked efficiencies in the past.

Industrial Revolution: Better Cheaper Faster

Consider the value of a bullet, I reckon it is worth at least one warrior. It is a one-time use item, but it is more efficient at disabling your enemy than a human. It would be pretty hard for me to convince a tribal leader if they have never seen a gun before, but they can be convinced. With a magazine of bullets, you are effectively leading a dozen warriors from your pocket! And that is the ultimate example of human efficiency. So with all the hype around generative AI now, how do we think about its effects on productivity? They are silver bullets for what exactly? To examine this holistically, I searched some past revolutions for clues.

In the first industrial revolution, humans developed specific tools to help increase their productivity. Imagine a craftsman and the toolbox they carry, how did those tools get made? What kind of mind is invovled in designing a hammer? How many iterations of prototypes were done to get the hammer to work. And just out of curiosity, did someone get rich from selling this tool?

Now imagine standardization across this trade. Craftsmen could now make a decent living and we need more of them to make more things. But making things by hand even with hand tools is just not scalable. Wouldn’t it make more sense to have an assembly line for this? We would build a conveyor belt where each craftsman specializes in just one small part. The specialization along with the assembly process unlocked the next level of productivity. Overtime, the craftsman becomes a factory man and the skill set that is required changes, from making one product by yourself to managing a complicated process to manufacture a batch of products. Because with more scrutiny overtime, we have built machines that can replace humans at building small individual parts, and we have an assembly line process of assembling the parts together. So to extrapolate this out into the future, a factory should really be automated to remove almost all human involvements, becoming an alien dreadnought factory. And to be frank I kinda saw that not in a factory, but in a shipyard at Rotterdam. A once vibrant harbor city filled with people and services for humans, now a mega shipyard with containers. No more human services because they didn’t need humans!

Ok so how can we replicate that process with services? Well, it certainly isn’t a copy-pasta job. In manufacturing, we can scale up because the output is identical and predictable. But in services, what we want to get done is usually different on a case by case basis. Both the inputs and outputs can change from one service to the next. For example, I paid $750 for a tax service and it took the accountants two weeks to consolidate my expenses, review my qualifications for certain tax benefits, and put together the form for me to review. How can we utilize this AI thing here so that it costs 100x less? Can this process be generalized to all services? I think we already have some clues on this.

The internet - scalable information Take the search engine for example, information lookup becomes trivial. So I could just look up relevant tax information I need, then put together the tax return myself. Have you tried to do this entirely by yourself? It is quite hard work and time consuming. You will spend lots of time reading before spending more time figuring out which column to put what. I had used turbotax for two years before finally giving up. There are too many edge cases they don’t cover and I am spending too much time manually typing each trade record. Search engines give you the information, but you have to extract the data, see if it addresses your needs, then do the work of applying the information to the job. Surely if you are a tax accountant and this process must be less laborious right? Possibly, but they can’t process too many clients if the turnaround time is two weeks.

so even without AI, we have streamlined some of this labor by structuring our data. Think SQL and strongly typed programming languages, these are akin to machines for the mind. Because they are logical and internally consistent, processes can be automated, standardized, and scaled. They haven been in production and it works. TurboTax is a mega business and it handles a large volume of tax returns. They streamlined the process such that it can handle most of the simple cases. So we have technology for batch processing and it is efficient, but it can’t handle exceptions. We need to manually constrain the inputs; the machines are simply not adaptable enough to handle unusual requests. And that is why I am waiting here in the DMV line, because registering a new vehicle is an unusual request as oppose to renew my registration.

When I was young, I didn’t realize how much it takes to configure a machine. Whenever we need to design, architect and build a service or product, it is incredibly laborious development work with many teams spending months on end specifying all the details and ways it can go wrong. I can’t help but think that there is gotta be a better way to design software for products or for services. Why do we need humans in this process loop? Because here is the conventional wisdom, a human can identify, think, build, and reflect on the issue at hand. Its ability is much more generalized than a machine. A judgment call is what a human can do.

So about this gen AI thing, is it capable of replacing a human?