There has been so much written about and talked about Artificial Intelligence over the past few years. It is going to be invasive, take away jobs, change the way we live, etc. I was reading an article in the Fortune magazine – 25 ways AI will change businesses. Many interesting ways predicted in the way we hire employees, change production lines, monitor health and so on. It got me thinking on what changes, if any, would we see in Data Networks, specifically in the method that the data packets traverse the Internet today and make it possible for so many applications to run. I am talking of the way we access Google, Amazon, any website on the Internet or chat on Whatsapp, use Instagram, Facebook, etc. Today, all of this is possible because of TCP/ IP. At some point in the future, TCP/ IP will give way to a more evolved protocol for transport, no doubt. Sticking with AI, how can the data packets traverse the network intelligently? Or is the network already intelligent and there is not much more in terms of intelligence that can be built in? Here are my thoughts:
Intelligence in design
It all starts with design. Think of our road network (well, maybe not, if your mind goes to Bengaluru road network) – there are multiple items to be considered when creating a plan. Current traffic, estimated traffic growth, political compulsions, population and industrial growth, etc. Data networks are very similar. Designed by humans, they are constrained by all of the above parameters + the cost involved + the capabilities of both,
- the designer and
- the devices to be used
Humans are super at learning. We replicate as much as we can, where we can and design is no different. Houses everywhere nowadays are square or rectangular boxes. Designs are “inspired” by other designs. Data network designs are similarly inspired – one enterprise’s network will replicate another and so on. Therefore, I think this is one of the first areas that AI can make a difference in. If all the intelligence is in replication, then this must be easy to automate and artificially be intelligent about. Machine Learning is expecting machines to learn from patterns, much like humans do. Machines can do a good job of learning design parameters and apply them, with suitable modifications (intelligence) to suit design parameters, can’t they?
I argue that machines can do a better job than humans at this, simply because of the number of parameters to be considered. Listing some that come to mind:
- Existing bandwidth requirement
- Estimated bandwidth needed
- Number of sites to be connected
- Proposed increase in number of sites (tied to business growth)
- Vendor equipment available and their capacities (can number in the hundreds, easily)
- Setup costs and maintenance costs
- Time available
- Ease of maintenance and setup (based on vendor and equipment history)
Network design is a specialized job, performed by professionals, who are usually experts in network equipment available from specific vendors. Not every enterprise can afford these professionals and there is no standardized practice for design that stipulates adherence. AI can change all of that. All of the above items are simply input parameters for the AI program, which will chew on it and spit out proposed designs. We want to retain a human element in all that we do – well, the final decision can be based on political factors, which are likely out of scope of the Artificial Intelligence at this point in time! Some things are best left to humans, for everything else, there is AI.
The next time, I will delve into some other data network elements that can change with AI. Till then, do pitch in with your thoughts. I’d love to hear from you!
Mithun's Musings 