Communications of the ACM - September 2018

years, the time until adoption by 90% of the population has progressively shrunk. For example, radio (starting 1925) took 50 years to reach 90% adoption and World Wide Web (starting

1994) 20 years.

Technology Jumping

When a technology passes its inflection point, many businesses respond by jumping to another technology that is still in its initial growth stage. The result is that the adoption rate continues to be exponential, even though the details of the technology change. Moore’s Law, for example, would not work without technology jumping. Ray Kurzweil argued that the doubling pattern of information technologies started in the 1920s, passed through four generations— electro-mechani-cal, relay, vacuum-tube, transistor, integrated circuits—and brought us to the fifth IT generation, silicon chips. 6 Kurzweil predicts the IT industry will make new technology jumps in the near future. There are many candidates such as quantum computers, GPUs, and neural circuits. Seba notes a similar process with solar panels and batteries.

How can we tell when a technology has reached an inflection point and it is time to seriously consider a jump? Andy Grove’s “10x criterion” for an emerging technology is a useful rule of thumb. He prepared for an inflection point by starting research projects in the emerging technology so that Intel would be ready when the new technology got on to its own exponential growth path.

Tony Seba used a similar approach formulated around cost functions of technologies. Adoptions increase exponentially as cost per unit decreases exponentially. For Seba, a jump point occurs when the cost curve of a new technology drops below the cost curve of the existing technology. A technology disruption becomes possible after the jump point because people have an economic incentive to choose the cheaper technology.

A more sophisticated approach uses an S-curve model, such as Bass model, fitted to the available adoption data.

Once the model is fit to the data, it can predict the inflection point and saturation limit.

exponentially decaying cost of batteries, which enable storage of excess solar electricity for later use during the night or daytime peak hours. Meanwhile, the cost of producing an electric car has been dropping, facilitated by improving battery technology and by significantly fewer moving parts than cars with internal combustion engines. Seba believes the switch to solar panels and electric cars will precipitate a massive shift away from oil.

In 1992, Andy Grove, Chairman of Intel Corporation, wrote about technology inflection points. 4 These are moments when a new technology emerges that is at least 10x ( 10 times) better than the current technology at doing its job. Grove thought those inflection points signaled impending economic avalanches. His response was to start research projects that would enable Intel to become good at the new technology and be competitive when the new markets arrived.

The common feature of the world described by these authors is exponential change, simultaneously in multiple technologies. For specific technologies, we can find useful trend lines, such as the periodic doubling of transistors on a chip or halving of solar panel costs. But these trends lines must be used with care: even though we can confidently predict the next generation of chips will be twice as fast as the current generation, we cannot predict the extent of adoption of the new chips or what people will do with them. Adoption depends on people choosing to embrace a technology; their choices are unpredictable effects of social context, norms, business practices, and receptivity. Consequently, John Seely Brown and Paul Duguid warn against using technology trend lines as means to forecast adoption. 8 More often than not, people start to see the impending disruptions and take steps to avoid them.

How do we navigate technology change when we cannot predict how people will use technology?

Underlying It All:
Exponential Growth

Let n(t) represent an adoption func-
tion: the number of units of technology
adopted by time t. Moore’s Law uses t
to index chip generations and says the
number of transistors doubles with
each generation: n(t)=2n(t– 1). Expo-
nential increase means that n(t)=eat,
where a is the growth rate.
We are very poor at appreciating
what exponential growth means.
Consider this old riddle: Every day
the rapidly growing lily pads of a
pond cover twice the surface area as
the day before, completely covering
the pond on Day 30. On what day was
the pond half covered? The answer
is, of course, Day 29. It seems to take
a long time to get halfway, then sud-
denly you are all the way.

How can you plan for investments, careers, and future markets if some exponential change has been going on and you do not notice it until Day 29, when it is too late? To help us cope we try various methods of prediction. If only we could see something coming on Day 15 or 20, we would have time to do something about it.

S-curves

Exponential growth is not a reliable extrapolator: at some point the resources to support the growth run out. For example, everyone in the community has adopted the technology; or transistors get too small to function properly. When we take saturation into account, adoption follows an S-curve: the graph of n(t) has an S-shape, exponentially growing to an inflection point, then slowing down and flattening at the saturation limit. The so-called Logistics equation was invented around 1845 as a mathematical model of the S-curve. In 1969, Frank Bass introduced a refinement that took into account the faster adoption rate of early adopters and gave a more accurate fit. 1