The idea was to draw a parallel between the impact of electrification and what we’re now seeing with AI — a comparison that Andrew Ng, one of my favorite researchers, often makes.
I found a fascinating study that dives into the large-scale adoption of electric motors, and here are the five main takeaways:
1️⃣ It took 20-40 years to reach the full-scale effect of electric motors on productivity.
2️⃣ The main reason it took so long was the slow pace of electrification at factories and plants, with price regulation playing a role as well.
3️⃣ So, why did factories hesitate to adopt electricity, even though it was clearly superior to steam power?
— Significant investments had already been made in steam engines, along with the entire organization of factories around them.
— Because of this, it was primarily newer and fast-growing industries (like tobacco at the time) that benefitted from electrification. However, these productivity gains barely registered in overall economic growth statistics because established industries waited for equipment to age out before making the switch.
— Retrofitting electric parts onto old steam-based setups also didn’t fully deliver; production was still organized around a central steam engine, limiting returns.
4️⃣ Once power could be supplied to individual machines, it allowed for a complete reorganization of production processes. Assembly lines could follow natural flows, and any issues in one part didn’t mean stopping everything.
5️⃣ Electrification also enabled more decentralization in production, unlike the centralized steam-powered setups.
So, here’s my billion-dollar question: how will a similar scenario play out in the case of AI?
Which industries stand to benefit most because they aren’t burdened by sunk costs in legacy equipment, skills, and processes? And how much are outdated workflows, built around older software that lacks capabilities like speech and image recognition or text generation, holding back productivity today?
Learn AI/ML through practice.
Join our AI/ML Simulator for Product Managers and complete four projects around the most common AI problems.
No coding or math expertise needed.