Invention, Innovation, and Diffusion - or the Twin Gaps
Why persistent elbow grease matters more than sparks of genius
If you’ve been around Product for a while you probably came across variations of these ideas, and have developed an innate sense of them. It pays to put them forward consciously and explicitly, especially in these days of “AI will speed everything up!”
Some things don’t change, or at least change very slowly. Human nature is one of them, and this fundamentally affects technology and Product work. This should help you understand what you’re seeing in practice, and hopefully in talking about this with execs.
Invention, Innovation, and Diffusion
These are the three stages of technological progress. Problems usually arise when people underestimate the time to progress between them. Let’s start by defining them, and talk about why the gaps matter.
Invention
Invention is the development of something new. A new way to do things, a process that didn’t exist before. Think of steam engines or Bell’s transmission of audio signals over copper wires.
Innovation
Innovation is finding a new application of technology. This is a new way of doing things, but the focus is on a different application rather than inventing a completely new way. Thing of putting steam engines on wheels to move trains, or codifying visual images over audio signals to create fax machines that operate on telephone wires.
Diffusion
Diffusion is the adoption of the innovation across society. Building up a network of railways takes time; the initial effort is to move people and goods between two terminal points, and as that diffuses we see whole changes urban landscape and the creation of commuter societies. The value of faxes is often used as an example of network effects, where the more people have one, the more useful it is.
A key observation is that few things are “true” inventions. Nothing exists in vacuum, and we build on what came before. It also takes a lot of tinkering to make something useful, which leads us to the next section.
The Twin Gaps
When you move between the above stages of technical development and adoption, you’ll encounter two main gaps that you’ll need to cross.
The Capability Gap
Just because a machine can do something, doesn’t mean it can do it well. Obviously “doing well” is context dependent, and will mean different things to different people. The invention of movable type together with the printing press improved over the use of wood-block printing. There were still a lot of improvements along the centuries — small inventions and innovations — until we could see something like affordable newspapers and mass-market paperbacks.
The Adoption Gap
This is directly related to the definition of “well.” The first cars required maintenance relying on mechanical knowledge, and not useful to the average person, even those might have afforded the car itself. Decrease in direct price, increases in reliability, and the availability of support services (mechanics and gas stations) slowly tipped the balance. It still took decades to reconfigure cities for the changed patterns of transport.
Related concepts
In tech circles, “crossing the chasm” is an old concept that speaks about the difficulty most companies run into when they saturated the enthusiasts and early adopter crowd, and are now trying to reach the early majority. This is when companies used to progressing at the speed of technology, suddenly hit a wall when their customer base is progressing at the speed of people. I have yet to meet the executive that likes to hear “take a deep breath and slow down,” even if that is what’s needed.
Today’s AI Gaps
Much of the AI industry today is in these early stages. The various inventions and innovations which led to ChatGPT’s launch in late 2022 took a while while to build up. But the public at large is still facing those two gaps, and will continue to for a while due to the very nature of AI as probabilistic technology.
First, is how well do AI systems handle cases. I’ve written before about that sense of magic when you first ask a model something and get a coherent, cogent reply, or see it figure out an answer after ‘thinking’ about the inputs. But then you scratch the surface, and it’s often superficial (or outright wrong).
Which begs the question, of what use is a technology that’s only 80% correct? Or 95%? or even 99.9%?
The use is in the use-case. For some applications, like suggesting alternative marketing slogans, getting it right four out of five times is good enough. For others, like giving official tax advice, even one-in-a-thousand wrong answer is unacceptable. That’s why when I work with clients (often government agencies or regulated industries), I always start with the impact on someone’s life — what happens when, not if, the technology goes wrong?
Past the gap in capabilities, lies the adoption gap. Just because the technology might make less mistakes than a human — is it accountable? Does society accept this? When it’s your life on the line, how much will you trust the AI lawyer or doctor — or car mechanic, for that matter — as primary care giver?
Despite the efforts of billionaire CEOs to ramp up their stock prices by over-promising capabilities, society will move in a different pace, commensurate with the capabilities and humanity’s capacity to adapt. The intervening period will be… interesting.
Final note
This article is looking at AI from a product manager and an historian’s perspective about the forces of innovation and the diffusion of technology. There are, of course, many other considerations that need to be applied to AI systems, like the environmental costs, legal and moral issues, societal impact in its application, lack of governance, etc.
While I fervently wish these would be addressed sooner, the current political climate is mirroring other trends in history, both broadly and specifically around AI, and makes it unlikely. It seems like affairs will progress as previous such socio-technical revolutions did. But that is a subject for another time.
For now, there are steps you can take to better plan for closing those gaps with your products (whether AI or not). These revolve on selecting the right technology for the use-case, and accounting for adoption. No matter the speed of technical advancement (real or imagined), the speed of diffusion is order of magnitude slower.
As always, good Product Management is always about the long game.