The future’s already here, it’s just not evenly distributed, and it doesn’t look like we expect it to
When scientists first started talking about Artificial Intelligence in the 1950s and 1960s, a lot of the discussion centred around how to best create AI that would think like people do. This view of AI has dominated our imagination ever since.
Think of HAL in 2001: A Space Odyssey, Deep Blue and other chess-playing computers, Skynet and the rise of the robots in the Terminator movies, and all the current discussion about the singularity. All of these are pictures of Artificial Intelligence doing what human intelligence does – just doing it faster, or better.
Over 50 years later, except for the chess-playing computers, we’re still waiting for this form of AI to take off.
Because we’re not seeing this type of AI, AI has been a failure, right?
Well, not really. There’s actually tons of different types of AI in practical use right now. An article in Wired UK outlines many of the current uses of AI, and it’s an impressive list: warehouse stocking robots, the google search engine, algorithmic financial trading, credit card fraud detection, and self-driving cars, just to name a few.
Even though we don’t have Skynet yet, we’re still interacting with AI throughout our day, often without even realising it.
The Long S-Curve of Innovation Diffusion
The story of AI illustrates a common innovation myth – that ideas spread quickly.
Around the time that computer scientists first started thinking seriously about AI, Everett Rogers was showing that most innovations follow an S-Curve as they diffuse through the economy. The Innovation Diffusion curve looks something like this:
Ideas are first picked up by people that Rogers referred to as Innovators, then Early Adopters. This is happening over the time period that I’ve labelled “X” in the diagram. Eventually, the new idea either dies off, or it takes off. Once the tipping point occurs, the idea then spreads rapidly throughout the market, until a saturation point is reached.
When people are innovating, or thinking about innovations, one huge mistake that they commonly make is to underestimate how long the idea will stay in the X range.
Here’s an example: email. The first email was sent around 1971, just a few years after the internet started. For a long, long time, email was only used by researchers, the military, and academics. It wasn’t until the late 1980s that universities started to make email available to students (that’s when I got my first email address).
By the early 1990s, the World Wide Web was built on top of the internet, and then email started to spread a bit more quickly. By 1993 or so, it was becoming relatively common among early adopters outside of academia. But even then, the question that you asked if you wanted to send someone an email was “do you have email?”. And in just a few years, suddenly everyone had email. By 1996 or so, the question was a simple “what’s your email address?” That was the tipping point. In another five years, it was “which email address should I use for you?, because everyone had a personal email address, one for work, and often a few more. Email had reached saturation.
If you thought that email started with the WWW in the early 1990s, X was only four or five years. But if you think of the whole story, X actually lasted about 25 years.
This is pretty common for new ideas. Xerography was patented in 1936, but the first Xerox machine didn’t hit the market until 1949. The technology didn’t take off for another seven years or so. X was about 20 years for this idea.
Even in the fast-moving internet age, X is often a lot longer than we expect it to be. Jeff Bezos had the idea for Amazon in early 1993. It took about two years to get the site up and running. In 2000, people were still calling it amazon.bomb, among other things. It didn’t take off until about 2002. X was about nine years for Amazon – and that’s one of the shortest time periods that I’m aware of.
The Reasons for the Long X
The unusually long period of X for new ideas is due to several things. Most of them have to do with uncertainty – we don’t actually know what the new idea is for yet. This happens in a few ways:
- We have to figure out how to make the new idea work: the best use of a new idea is often not obvious. In fact, because we tend to think in analogies, we often get this wrong at the start. In the AI example, the technology didn’t start to really take off until people stopped asking “how can we make computers that think like people?” and they started asking “we have computers that do some things that people can’t do well – how can we make use of this?” Going through this process takes time, and it requires a lot of experimentation. Many of these experiments will fail – but one of the critical things that we have to figure out is under which circumstances the new ideas work, and under which ones they don’t work so well.
- We have to fight against the hype cycle: the long X is a direct contributor to the hype cycle. The Early Adopters get excited about the new idea, and it gets oversold. Then the people that are threatened by the new idea fight back. When it doesn’t spread as quickly as expected, the excitement wanes and cynicism sets in. Eventually, though, through experimentation we figure out what the best use of the new idea will be, and at that point it is finally poised to take off.
Greg Satell explains this process very well in his post Why Trends Are For Suckers. This is what the hype cycle looks like – you can see the long X at work in it!
- Most importantly, we have to figure out how to create value for people with the new idea. This is the part that the Early Adopters tend to ignore – they usually like new things simply because they’re new. For everyone, the new idea needs to solve a problem. Avinash Kaushik explains the issue perfectly in 11 Digital Marketing Crimes Against Humanity:
When I look at winners and I separate them from the losers there is one thing that stands out. Winners have a sophisticated understanding of the holistic success of their digital existence. It comes from undertaking two simple steps: 1. Identifying their Macro and Micro Conversions and 2. Quantifying Economic Value.
I tend to talk about the need to create value more broadly, not simply economic value – but in either case, without clear value creation, the new idea will never take off. Again, it takes some time to figure out where how to create this value, and often the value being created isn’t the value that was originally expected.
The Myth of Quick Adoption
Our tendency to dramatically underestimate the true value of X in innovation diffusion causes all kinds of problems. If we’re early adopters, we expect new ideas to spread quickly. And yet, they don’t. If we’re threatened by new ideas, the long X can give us a false sense of security. As it becomes clear that early predictions are exaggerated, we become complacent. But eventually, once all the experimentation has been done, and people have figured out what the new ideas are really good for, and how to create value with them, the threat begins to bite.
I’m not sure of any way to move through the innovation diffusion curve more quickly. It is by its very nature slow, experimental, unpredictable, exciting, revolutionary and wasteful. It is part of what makes innovation both exhilarating but also frustrating.
Being aware of the myth of quick adoption is the first step towards figuring out how to deal with it.