According to RootMetrics, a mobile industry analytics firm, San Francisco and other Bay Area cities have among the worst wireless networks in the US. How is it that at the center of technology innovation, businesses and consumers must put up with close to the worst mobile networks in the United States? Part of the answer can be found in the archives of the California State Assembly.
The trouble with old infrastructure
In 2016, the telecommunications giant AT&T asked the State of California for permission to dismantle its existing copper-based, landline telephone network. In the bill, AT&T explained to lawmakers that 85% of California households no longer use landline phones.
AT&T argued that the large environmental and financial costs of maintaining the copper-wire network hinder innovations and improvement of other aspects of its communications infrastructure; i.e. fiber optic and mobile networks.
The Transatlantic Cable first lain in 1858 (unsuccessfully), to be successfully relain in 1866, could mark the beginning of an interconnected world —in the modern sense. Photo: the Great Eastern steamship laying cable between Newfoundland and the UK in July 1865.
The Plain Old Telephone System (POTS) evolved out of the telegraph network of the early 19th century and was the first (electric) global communications network. In 1865, sending a message (a letter, a note) from New York to London took about 6-8 weeks. In 1866, when the transatlantic telegraph line was completed, an 80 word message took 1 minute to travel the Atlantic ocean.
Later in the century, the telephone network was invented and designed to work on the same principle —send electrical impulses over copper wires, filtering and amplifying the signal at points along the way. In the 20th century, great advancements in manufacturing and computing allowed humans to do great things with two-strand copper wire networks (paired copper wire networks). Fax, Cable TV, even the internet came into existence thanks to the existing and greatly evolved copper wire communications networks.
However, as the 2016 California State Assembly bill evidences, AT&T is not interested in the nearly two-century-old technology anymore. Nor are any of the other telecoms. The California Assembly was not the first to see hear requests to dismantle or abandon the old POTS network —and will not be the last. The question is: Why are AT&T and other telecoms keen to dismantle their plain, old telephone services?
When infrastructure technology can’t grow with you
Increasingly, our modern economy depends on technologies like GPS, mobile computing, and wireless communications. Consumers and industries alike demand fast, flexible and adaptable communications networks. Vast, yet often overlooked sectors of our modern society such as banking, securities, shipping, depend on high-capacity, high-speed, highly reliable networks. Copper just can’t keep up —it’s weighing us down (literally and metaphorically speaking!)
In the 1970’s and 80’s fiber optics where on the very cutting edge of communications technology and a prominent feature in futuristic-looking photography. 1988 is the year that the first fiber optic transatlantic cable was lain. Today all of our undersea cables are made of glass, as are the links between most cell towers.
POTS is an evolutionary dead-end. There are physical limits to how much information can be transmitted over existing copper wire networks.
No amount of upgrades to the system will increase the boundaries imposed by physics or lower the mechanical difficulty of maintaining it.
However, much of modern communications was built on the old copper wire system. For example, the 911 Emergency Response system evolved with the POTS system. As part of this system, 911 operators can locate people who call from landlines, but not VoIP phones or cell phones —not because the technology does not allow for it, but because the systems have not been set up for it. As a result, when telecoms propose to switch to a new system entirely (like fiber optic) members of certain communities (example), special interest groups and some industries my oppose such changes (like paired wire manufacturers who did not invest in fiber-optic tech back in the 80s).
Since switching to the modern system takes so long, especially when bogged down by the cost of maintaining the old system, telecoms are forced to charge more for legacy services or incentivize the purchase of otherwise out-moded services (bundled voice and internet packages). Why? As more and more consumers choose mobile networks as their sole voice providers, telecom companies must figure out a way to pay for the old, heavy system.
Maintaining a legacy system like the old phone network can be very expensive and worse: become a downright hinderance. How can we tell when a new technology or system will replace the older, established technology? Easy: look if there are any new adopters of the legacy system.
New technology enables rapid growth
Consider the example of telephony and mobile networks in sub-Saharan Africa. The Pew Research Centers has published a number of studies on the state of communications networks in Sub-Saharan Africa. According to its 2018 report, traditional landline telephone market penetration is roughly 0%, while cell networks have a market penetration of roughly 80%. The Pew studies also show us that mobile networks have grown faster in Sub-Saharan Africa than they have in the US.
Between 2002 to 2014, cell phone networks in 7 Sub-Saharan countries grew more than they did in the US —the world’s largest economy.
A graphical comparison of 5 Sub-Saharan African statesGraphic courtesy of Pew Research Center
While in 2002, only 12.2% of the population in the Sub-Saharan countries had cell phones, 78.4% had cell phones in 2014. This constitutes roughly a 66 percentage point increase. In the same period of time, the US was only able to increase mobile networks by a mere 25 percentage points.
Unencumbered by a legacy system (neither landlines nor legacy mobile networks), the 7 African countries in the study were able to build a modern communications system that scales well and that can continue to evolve, delivering customers solutions that legacy systems could never have.
The Take-Home lesson
These numbers say a lot about the power of new technologies to do exponentially better than legacy systems. The African countries in the Pew study can continue to improve their communications by taking advantage of a new technology without having to consider replacing an existing communications network. Societies which are now developing communications networks for the first time have the advantage of choosing more modern technologies rather than having to go through the whole evolution of communication unnecessarily (telegraph to phone to data, etc).
Back here in the US, so many of our everyday tools, i.e. cell phones, social media, Google Maps, depend on the fast, cellular and fiber optic network. As a result, telecommunications companies are anxious to cast off the burden of the legacy POTS systems, favoring faster, cheaper-to-maintain, and highly scalable optical and wireless networks
However, due to the vastness of the POTS network and the number of social institutions that have evolved with the legacy copper-wire system, this process will take a while to complete. Companies currently using POTS networks have the unique opportunity to come to dominate the market by being at the forefront of the changeover in the US. Those that do, will outlast those that wait too long or allow social pressures to stall their efforts.
A Similar Vein – Legacy ASR Systems
You can see a similar progression in the newer world of ASR —Automatic Speech Recognition. In a sentence, ASR technologies are what data scientists use to gather business and academic insights from large voice data sets like those found on smart sales enablement platforms such as 8×8 or Invoca, products like Siri, and call centers.
Much like the US’ cell networks, ASR technology hangs on to remnants of its past. Whereas products like Nuance Dragon Transcribe were revolutionary in the 1990s, the costly and time-intensive methodologies that gave birth to the industry 25 years ago have become insufficient to do the smart, AI-driven analytics that drive modern industries. The parallel here is clear: in the same way as the POTS system was cleverly upgraded, adapted and modified to enable internet connectivity, and television (AT&T U-Verse), ASR technologies were tweaked, duct-taped and wrangled into the Siri and Google Home products that we know and love today.
Advanced Research Projects Agency was responsible for funding research which led to the creation of many technologies that are an integral part of the present day.In the image, a virtual reality system.
In the 1990’s the chipsets that allowed engineers to bully internet over the aging copper wire network finally became cheap enough to make DSL available to the average consumer. This was a brilliant intermediate step that brought broadband internet to the homes of billions of people in the around the world —but it was just that, an intermediate step.
Similarly, hybrid ASR systems work well for the current uses cases or “general” purpose transcription, but fall short when data scientists, businesses and researchers need high accuracy and faster performance to analyze their specialized voice data sets. Furthermore, just like POTS require costly maintenance, hybrid ASR systems burn more computational, time, and monetary resources to do their job.
A radical evolution in ASR
The developments in fiber optics from the 60s, 70s and 80s inspired the imagination of engineers and the public alike. It was clear from the 80s onward that fiber optic cables would replace copper wire. It was also clear that advancements in several fields would be necessary to make the switch complete. Similarly, at the same time—in the 1980’s and early 90’s—mathematicians and computer scientists developed powerful learning algorithms that would revolutionize machine learning and therefore ASR. However, these developments could not be reliably implemented until critical advances were made in computational power and access to large amounts of data.
The creation and general adoption of the internet in the 90s and early 2000s (made possible by hybrid fiber optic/POTS networks) gave us the big data we needed. Finally, in the second decade of the 21st century the powerful computing needed to make deep learning ASR feasible came into existence. Now all the key pieces were there. End-to-end deep learning ASR frees engineers from scaling problems, enabling them to accurately process voice data in real time. This allows companies to learn much more from the voice data they’ve been unable to access (for nearly a century).
Imagine how many hours of recorded radio exist in an archive somewhere. ASR technologies allow us to search vast stores of audio data and find actionable insights for business and science. For example: Are my call center operators staying compliant with disclosure regulations?
One of the game-changing aspects of end-to-end speech recognition systems is their ability to rapidly train custom models. Rather than relying on the general models of early hybrid ASR giants like Nuance and IBM Watson —which ask your data to fit their systems, end-to-end ASR technology enables custom models that fit your data, with all of its unique quirks and business-specific language. This means that rather than investing in a static, monolithic ASR system, you invest in technology that continually adapts and improves to fit your needs.
What tech evolution means
Every once in a while, a new technology appears that radically changes human society. The telegraph-phone-copper wire network of the 19th century was such a system. The idea of connecting the world at near-light speeds was good, and we have found faster, more flexible, more responsive ways of doing so: mobile and fiber optic networks.
The same evolutionary process is evident in the world of ASR. Rather than typing into a computer, increasingly we talk to it with Siri, Alexa, etc. The technology that makes such a conversation possible, ASR, has taken another step in its evolution now allowing us to probe voice data, look at large amounts of call data to extract actionable information—something essentially impossible even a few years ago. Deep learning ASR allows companies to
- Stay compliant with regulations by looking at 100% of their calls
- Track sales trends in phone call data
- Efficiently understand your customers attitudes
- Provide meeting summaries automatically
ASR represents a major step towards communicating more naturally and without impediments. Watching deep learning ASR make this a reality in the years to come will be fascinating.