Saturday, March 31, 2018

Impact of Blockchain, Deep Learning et al on next generation solution architectures

We are facing yet another onslaught of technologies that will revolutionize societies and work. People are predicting that the new technologies of block chain a.k.a distributed ledger and AI a.k.a Deep learning and predictive analytics will change societies, disrupt businesses and render all of us slaves.

However, as much as new businesses will be formed around these strategies, its not necessary that they will prove as disruptive as people claim in each and every case. Societies with more people than jobs have already figured out that the social contract cannot be wished away except in authoritative societies. Democracies require people to win elections, and that means that to whatever limited extent, the will of the general population still applies, unless you have an oligarchy that is operating . If people cannot put food on table, all progress is merely academic.

Also, in a market economy, it does not really help if you can only sell what you build to a handful of people or corporations. Look at phones like Vertu that most people have not heard about. It is much easier to sell something that costs ten dollars to a million people in today's frictionless economy than sell a ten million dollar item to one person or corporation.

What does this all mean? In my mind, it means that the new disruptive technologies with narrow applications will still need to fit in broader solution architectures. It also means that they will not be used where their application cannot be justified in terms of cost or complexity of deployment.

Let us start with block chain. It is a distributed ledger which provides a non-repudiable history of transactions. So it assumes that we have the context of a distributed transaction database. With that, it is logical that main-stream database vendors are adding this capability to their primary offerings to extend the transactional history beyond the firewall. The use cases where block-chain makes sense, is therefore limited to scenarios involving multiple parties that do not trust one another and need to transact. There may be many legacy applications where this fits, but getting technology to a point where people trust the technology enough, is going to take time. At least three to five years in my opinion.

Deep learning on the other hand has 3 primary use cases. It can be used for classification, for prediction and for generative scenarios. Again, not an easy scenario to pull off. You need to train the networks, and then deploy them. There could be embedded intelligent agents that update themselves from a common training pool, but it will take resources to deploy.

Bottomline, I think these are great technologies in the hands of the architect, but the hype around them is in many ways unwarranted.