Published 07 February, 2020; last updated 10 December, 2020
There do not appear to have been any greater than 10-year discontinuities in telecommunications performance, measured as:
Radio does not seem likely to have represented a discontinuity in message speed.
This case study is part of AI Impacts’ discontinuous progress investigation.
Fiber optic cables were first used for telecommunications in the 1970s and 80s.1 While previous telecommunications technology sent information via electricity, fiber optic cables instead sent information via light. Though electric signals travel at around 80% of the speed of light, while optical signals within a fiber only travel at roughly 70% of the speed of light, fiber optics have other benefits which add up to a considerable advantage.2
Bandwidth-distance product, usually given in bits*kilometers/seconds, is both the most apparently relevant metric of progress in telecommunications4 5, and the one that was suggested to us as discontinuous. We also considered data transfer rate (measured in Mbps) for transatlantic cables, as a metric which more closely tracks the performance of cables that were actually in use, with the Atlantic serving as a distance constraint. We found separate data for bandwidth-distance product across all technologies, in fiber optics alone, and crossing the Atlantic, so we consider each of these metrics.
We used a tool for extracting data from figures6 to extract data from Figure 8.2 from Agrawal, 2016,7 shown in Figure 1. We put the data into this spreadsheet. Figures 2 and 3 show this data without a trendline, and the log of the data on a log axis with a straight trendline.
Figure 1 below shows progress in bandwidth-distance product across all technologies on a log scale.
If we treat the previous rate of progress at each point to be exponential (as Agrawal does, with two different regimes) then optical fibers appear to represent a 27 year discontinuity.8 The following 2-3 developments are also substantial discontinuities, depending on whether one breaks the data into multiple trends. As shown in Figure 3 however, the log of the data fits an exponential trend well. If we extrapolate progress expecting the log to be exponential, there are no discontinuities of more than ten years in this data. This seems like the better fit, so we take it there are not discontinuities.
Argawal’s data also does not include minor improvements on the broad types of systems mentioned, which presumably occurred. In particular, our impression is that there were better coaxial cables as well as worse optical fibers, such that the difference when fiber optics appeared was probably not more than a factor of two,9 10 or about six years of exponential progress at the rate seemingly prevailing around the time of coaxial cables.11
Figure 4 below shows bandwidth-distance product on a log scale in fiber optics alone, from Agrawal, 2016.
We chose to model this data as a single exponential trend.13 Compared to previous rates in this trend, there are no greater than ten year discontinuities in bandwidth-distance product in fiber optics alone.14
There was a notable temporary increase in the growth rate between 1996 and 2001. We speculate that this and the following 15 years of stagnation may be a result of heavy telecommunications investment during the dot com bubble.19