How AI Revolutionizes Established Tech — Part 1

Ever wonder what happened to technologies that not so long ago were part of our daily lives? As Artificial Intelligence (AI) enters more and more areas of our existence, the question is whether older technologies can keep up with these changes or whether they are destined to be forever forgotten… In this brief series, we look at a few of these tech and how they managed to adapt and embrace AI.

Part 1: AI Rules Wireless

For nearly two decades, Bluetooth and WiFi Routers have been at odds with each other, avoiding interference and managing spectrum as the wireless ecosystem kept expanding. But how were they able to accomplish that?

Bluetooth adopted a strategy that served them well for a while. Specifically, they shifted operation to unoccupied bands by detecting transmissions and changing frequencies rapidly. These are known as frequency-hopping techniques, and such tricks allowed Bluetooth devices to escape the very real and threatening technological annihilation that seemed unavoidable as demand soared for more radio spectrum resources.

But wait…What happened as more and more people streamed videos and scrolled through social media on their smartphones? The sensible approach that regulators took to manage wireless for allocating different frequencies for the exclusive use of specific radio users became impractical.This meant that wireless technologies could no longer have exclusive frequencies, and in spite of frequency hopping, the sharing of available bandwidth was no longer feasible.

The inefficient use of spectrum due to unused portions combined with the surging demand in wireless data transmission called for more advanced solutions.

The spectrum-management problem had to be rethought and teams all over the world turned to AI to design methods for sharing spectrum to increase overall data throughput.

Closer to home, DARPA (the US Defense Advanced Research Projects Agency) launched a competition with some attractive prizes to train AI models that could learn to allow autonomous radios to collectively share wireless spectrum while transmitting far more data than would be possible through assigned bandwidth frequencies.

At the core of the solution lies the idea that if there are too many radios trying to send signals there won’t be enough available spectra, so sharing a spectrum band simultaneously is the only way to go. The teams competing in the DARPA challenge developed scenarios where independent radios broadcasting together would access the same frequencies and AI systems would work out the sharing of frequencies among networks.

The success criterion in the competition would be how many successful tasks such as phone calls and video streams were completed while sharing frequencies with other networks. This meant that radio networks would access the same frequencies simultaneously, but each network would use an AI system to figure out the sharing of those frequencies. The real winners in this case are the AI-managed radio networks that collectively accomplish more tasks than they would if each radio was using an exclusive frequency band.

AI techniques have excelled in a few areas where the rules of engagement were predetermined such as playing chess, or where variations on a problem such as speech recognition called for more exceptions to the rules. But spectrum management did not fit with any of these.

In the end, managing the wireless spectrum turned out to be a reinforcement-learning problem with rewards for success such as when data is transmitted successfully and penalties for failure, such as dropping of transmission. The AI system accumulates points during the training period, remembers these successes and tries to repeat them during the testing phase. Furthermore, as if this was not enough, it turned out that training multiple AI systems together was even more successful, similar to sports teams where players train and play together.

The takeaway from this experiments of allowing AI to take over some of the wireless problems is that if we can predict, even for a moment, what frequency might open up, just enough to transmit more data, without grabbing more than needed, a fundamental change can occur in conserving the spectrum environment without wasting precious hertz. If only humans could emulate this approach and allow themselves to collaborate, to capitalize on every opportunity, and then know when to step aside until thenext opportunity opens up. In any case, AI could potentially open up a new era of wireless communications.