Qualcomm had a 5G update this month, and things appear to be going very well. With over 45 OEMs launching phones and over 50 operators having already deployed 5G technology, things seem to be moving far faster than 4G did.
Driving this rapid move is the fact that 5G is a bundle of technologies that include security, spectrum sharing, spectrum aggregation, standalone mode, and wireless network virtualization. Rollout speed is also being enhanced by new cell site technology, which was showcased at the event by Rakuten using Altiostar’s unique solutions, which took the time needed to provision a cell site down from 3 days to 8.5 minutes.
With the old way, you could do ten cell sites a month; with this new technology you do more than 50 a day (granted you’d need to take into account travel time but since much of this would hook to existing cell site power, and you could ship it to the site for installation at speeds that are truly amazing).
But throughout the presentation, whether they were showcasing Samsung’s powerful 5G flagship, the XR2 Extended Reality solution or advances in their Always Connected PC initiative with Microsoft and Lenovo, there was this sense of massive numbers of computing capable devices connected over low latency, optical grade, virtualized wireless network.
Collectively I think this could form the foundation, particularly when you include the distributed data centers and ever-smarter connected sensors, to create the first General Purpose AI at scale (which is expected in the 5+ year timeframe). While I’ll grant you this same concept formed the foundation for Skynet in the Terminator movies, it still represents a powerful possibility.
General Purpose AI
Right now, AIs are developed for very narrow applications, say facial recognition or object identification. You define the problem, figure out how to feed the system the needed data, train the core system, and then distribute that training to inference engines, which then can do the specific job they were designed to do. If you want this AI to do something else, that’s a shame because it won’t. It is designed for a specific purpose and that both makes the project far easier and far less flexible than is ideal.
In contrast, a General Purpose AI, which only exists in concept right now, can move from task to task like we do and make complex decisions based on widely disparate information sources. Let’s say you wanted to isolate likely coronavirus carriers with a system designed to identify drug traffickers. With the old dedicated AI, you couldn’t get there, but with a General Purpose AI, you’d change what the system was monitoring, create a logic tree, and you might even have the AI develop the skills that would then pass to the inference engines. All of this would happen in seconds and minutes, while the old way might take weeks or even years to get to the same place.
You don’t just need to change what the inference engine is analyzing. You need to incorporate different elements like skin temperature, visible attributes of the virus (sneezing, coughing), and determine proximity to known carriers.
Wrapping Up: Why 5G Makes This Possible
For a General Purpose AI to function, it has to be able to reconfigure rapidly the entire infrastructure surrounding the solution in realtime. With 5G you not only get greater bandwidth and security, but the networks are also largely going in virtualized. This means you can make major changes in traffic flow, what the networks connect to and prioritize, and even make changes to the 5G sensors distributed throughout the ecosystem already possibly serving other purposes.
For instance, you might repurpose the Infrared facial recognition cameras in an airport to also measure facial temperature to help flag those with fevers. And suddenly, a solution that was developed for drug interdiction and to capture terrorists is changed to identifying coronavirus carriers so they can be removed from the general population helping to contain the spread of the virus.
This highly flexible nature of 5G will be critical to the creation of a General Purpose AI, which will need to make use of that flexibility to provide a wider range of functions than we currently get from our focused AI solutions today.
Now clearly, this is only part of the solution, we need the brain too, and a future version of something like IBM’s Watson will be equally critical to the effort. But without this massive realtime network flexibility, I doubt we could put 5G on the critical path to General Purpose AIs at scale.