I’ve been thinking a lot about where AI will have the most impact. Will some industries be more impacted than others? Will certain functions be quicker to adopt AI than others?
Penetration of new technology isn’t usually uniform. Forming a view of where it will happen faster puts you at an advantage for spotting opportunities and getting your timing right. Most emerging economies skipped the PC revolution and went straight to mobile. We had a bunch of exciting products launch to address that — think UPI in India.
AI is useful because it is a 24/7 machine. It doesn’t get tired, and it makes no judgement on the task it is given. This means it’s useful because:
Customer support is an example of a repetitive task that is expensive. Applying AI to customer support helps automate the repetitive stuff. It creates two advantages: 1) lower cost for companies to provide support, and 2) better quality because your support team spends time on the complicated stuff.
Drug discovery is an example of an application where there are too many inputs. Scientists discover drugs using a combination of academic research, trial and error, and experimentation. AI is helpful because it can trawl through more information than a human can. In doing so, it reduces the number of options on the table for the scientist to test and improves drug discovery. Incidentally, Insilico Medicine was the first company to be granted a license by the FDA for a drug discovered and designed using AI.
AI’s speed of penetration will be driven by the following:
One of AI’s biggest short comings today is the hallucination problem. It can sound very right but be very wrong.
There’s a bunch of people working on this — I have no doubt it will be solved with time. However, it does mean that AI will be more useful in some industries vs. others. You do not want people asking ChatGPT for medical advice because if they do, and it gets it wrong, it could lead to a terrible outcome.
On the other hand, if you are writing a marketing email, it’s absolutely fine to get ChatGPT to write your first draft and then iterate.
Jobs where a large portion of the work is done in front of a computer will be impacted first.
When you think about actions AI can automate, it’s obvious that tasks that can be completed using a computer alone (”knowledge tasks”) are easier to automate.
On the flip side, consider a factory. It’s going to take a while until AI can run and optimise a factory. Even if the software existed, plugging it into the hardware in the factory presents a whole new set of challenges.
On a similar note, I suspect we will see regulated industries move slower than others. Consider an industry like healthcare. Before deploying AI in a user facing environment, you will need to get it approved by a bunch of regulators. There is important nuance here — a regulated industry might adopt AI very quickly if a qualified individual remains in the loop. Legal services is the best example of such an industry as we will see below.