While the second coming of AI hype from generative AI has been even higher than the first hype cycle in the mid-teens, the story is similar: more automation for simple tasks (and some more complicated tasks), making individuals more efficient, and reworking the economics of a business.
Vertical Expertise Will Become Less Important
As digital transformations assemble end-to-end workflows to replace our human and paper processes, work will be transformed. Laying out a business process will be the critical first step in any commercial enterprise, for two reasons: (1) it will be tied directly to financial costs, and (2) it will dictate where your human labor and machine labor will be allocated.
Machines will do the “vertical” work, that is – tasks that are repeatable, benefit from speed of delivery, and require attention to detail that is highly consistent and invariable.
Humans will do the “horizontal work” – tasks that require creativity, accuracy, and/or require lots of attention to variance.
What will it actually do?
The introduction of generative AI didn’t change the fundamental model. AI still can’t obtain, en masse, very high levels of accuracy in complicated tasks (i.e. an LLM can’t perform higher than 70-80% accuracy at many objectives), but it is becoming more efficient. Remember – it’s still just a machine built by humans, that can handle more tasks, to make one person money (instead of many). The impact on concentration of wealth is a post for another time.
It will replace tasks instead of people, changing what they do to be and become more efficient. For domain specific tasks where 95%+ accuracy is required by customers, these will remain human-focused (the irony being, many humans perform at less than this clip – thus the importance of hiring the right people).
For example, instead of replacing a senior analyst at a knowledge work firm (i.e. law, accounting, medicine, services) – 25% of their tasks may be automated 50%, leading to 12.5% increased efficiency.
Junior staff, on the other hand, will be more at risk – if 50% of their tasks can be automated at 75% (they are more invariable), then nearly 40% of their overall work may disappear.
This means general purpose algorithms will be difficult to use in domain specific context, but may help with general tasks, and make the training of those domain specific models far easier, faster, and cheaper.
What’s the Impact?
For Investors: For investors, the thesis around AI has to balance the accuracy of the algorithm so that it is high enough to be useful in a domain, but also transferrable elsewhere. Otherwise, you’re building a very expensive service business that creates an algorithm every time a part of a problem is encountered (instead of productizing to service a problem completely).
Consumers: If we utilize AI properly, it will represent an enormous shift towards consumer power, and an increase in consumer surplus (without getting complicated, this is due to microeconomics and MR=MC). Due to the inevitable (and here, very beneficial) march of capitalism, higher quality products like legal or medical advice, will become cheaper, better, and be delivered faster for nearly all consumers.
On Architecture: Because of this, architecture will be the most critical function of any modern business – knowing where/when to implement AI. If a business is simply an output algorithm (the measurement being its stock price), then those with the best “architectures”, or data models, will win.