Did steam power and electricity go through the same slow, invisible payoff phase that AI seems stuck in now?

This explores whether steam and electricity went through the same 'where are the productivity gains?' phase that AI seems to be in now. The short answer: the corpus doesn't hold historical studies of either one, so it can't line up the two directly. What it does have is a framework built for this exact question, plus several current measurements that show the pattern happening in real time.

The framework is the productivity J-curve from Brynjolfsson, Rock and Syverson Why does AI productivity appear weak when investment is high?. Their argument is about 'general-purpose technologies', meaning inventions that reshape many industries at once. Getting value from one takes large investments that official statistics barely count: retraining staff, redesigning processes, building new software and ways of working. During that build-out, productivity looks worse than it really is, because effort goes in and nothing countable comes out yet. Later, once those hidden assets start producing, measured productivity overshoots. Electrification is the usual historical example for this kind of lag. Factories first swapped a steam engine for one big electric motor and saw little benefit. The gains came decades later, after factories were rebuilt around small motors on each machine. That history comes from the wider economics literature, not from this corpus. But it's the reason the J-curve is framed as a recurring pattern across general-purpose technologies rather than something unique to AI.

What makes the corpus useful here is that you can watch the dip side of the curve in today's data. Executives report AI gains bigger than their measured results show, and the likely reason is that revenue lags behind operational improvements Do AI productivity gains feel larger than they actually measure?. That gap between felt and measured gains is what an unmeasured-investment story predicts. Economy-wide estimates are still small: about 1.1% from worker surveys versus Acemoglu's 0.7%, and the difference comes from method rather than real disagreement What is generative AI's actual aggregate productivity gain today?. Narayanan and Kapoor add a further reason the gains may stay hidden. AI speeds up the 'doing' part of knowledge work, while the work of deciding and delivering stays the same or grows Does AI really compress all layers of knowledge work equally?. So faster tasks don't automatically show up as more output.

The twist worth taking away: the historical parallel cuts both ways. The J-curve says the early slump is partly a measurement artifact. It also says the later surge partly is, because growth looks fastest when the hidden investment finally pays off, not when it was actually made. So whether a GDP figure says 'AI is disappointing' or 'AI is booming', part of what it reflects is where we are on the curve. If you want primary sources on steam or electricity specifically, this collection doesn't have them yet. The J-curve note is the closest way in.