I recently read the book Machines of Loving Grace: The Quest for Common Ground Between Humans and Robots by John Markoff, a technology and science reporter at the New York Times. This is a good book that goes over the history of the development of automation in the 1950s and 1960s, and takes you to the current day where new robotic developments from Apple (Siri) and Google (driverless cars) put us in the another age of rapid change.
The book delves into the differences between goals of
- Artificial intelligence (AI - creating a human-like intelligence, like a robot) and
- Intelligence augmentation (IA - sometimes called Intelligence Amplification).
- It also touches on the question of how these systems will affect employment and the role of humans in the years to come.
Network automation spans a range of definitions, from simple scripting to the more advanced notions of translating policy and intent into low level networking, probably best exemplified by Cisco’s Application Centric Infrastructure. The future may become more interesting as more complex forms of intelligence will simplify network design, operations, monitoring, and troubleshooting. There are foundational pieces that are starting to be put in place, such as advanced telemetry for gathering even more data about the state of the network.
Some people have stated that computer networks are the nervous system for an information infrastructure. If you take that analogy further, you wonder whether the data gathered by the nervous system (not the data transmitted by the network, but the state of the network such as congestion, throughput, link status) can be utilized to reconfigure the system for better operation.
I asked that question about a closed-loop feedback at the 2015 Open Networking Summit session on Data Center: SDN Solutions to a variety of speakers including those from Arista, Big Switch, Cisco, and VMware. The consensus seemed to be that autonomic feedback isn’t quite there yet.
But I’m looking forward and wonder if we can make significant progress in the future and whether network automation will fall into AI (automatic behavior) or IA (making people do their jobs better), or whether we have started to touch on these capabilities and what effects they have on employment, skills sets, and enterprise data center or campus operations.
We currently haven’t yet taken the road toward AI. Making architectural design decision requires skills sets based on significant training and certification, as shown by certifications such as CCIE and CCAr (Architect). But what about more day-to-day like work? Have systems already been successful in extending monitoring? Systems like Arkin and Kentik have combined network visibility with big data to provide insights into network operations that plain human analysis finds hard to do at scale.
Systems like Cisco Application Centric Infrastructure have gone a long way to reduce the need for understanding every CLI command and changes the conversation to understanding higher abstractions, while at the same time having complete control over the packet forwarding and the data plane. And at a low level, Multi Link Aggregation (MLAG and variants), as used by Arista, Big Switch, Brocade, Cisco, HP, and Juniper help eliminate headaches related to the Spanning Tree Protocol. We take those things for granted now, but if you consider the manual work required in the past to prevent the network from going haywire, these are early forms of intelligence augmentation.
I believe that in small ways, we have already achieved forms of intelligence augmentation via various forms of network automation and core technologies. We have not yet achieved intelligent automation, but in recent years, we have make great strides toward that eventual goal and we may see AI capabilities, too. Network administrators need not worry about their jobs going away, and IT managers need not think about reducing training and certification investments. Indeed, perhaps an emphasis on higher level training is what the doctor ordered.