In the backdrop of supremacy for the digital assistant and the progress in development of driver-less cars, the stakeholders in the enterprise try to gauge when the progress in the artificial intelligence would let them transform their machines in the data-center to intelligent assistants.
Anirban Kundu is SAP Program Director and Global Enablement lead for SAP S/4 HANA Embedded Analytics. I really appreciate his guest blog here on my website.
The transformation above looks like a page out of the next AI thriller when the machines become intelligent and can sense, respond and learn about the events they are set up for and start communicating with us predicting the next best decisions we ought to make in an enterprise setting. So, the question is – Is this possible at all and if yes how far are we from that day?
In this blog post Anirban presents a possible path for the making of the Intelligent Enterprise and break down the making of the intelligent enterprise into a journey and propose a “reference architecture for the intelligent enterprises”.
As a working definition, we would measure the intelligence of the machine based on its ability to respond with least latency and highest accuracy to events and situations for the scope it has been commissioned for. Subsequently its ability to predict events and generate response which can be automated to deliver the best outcome would qualify it as an intelligent machine.
In our simplified model, we would take the analogy of development of intelligence in humans to outline the steps which can be independently implemented as parts of the reference architecture. The model of intelligence starts with the ability to process the various signals in the environment and first step is to become “context aware”. Let’s say the child touches a hot plate and instantaneously it responds by taking the hand away from it. This is the step 2 when the “response mechanism” gets instilled and we see the basic form of intelligence manifest in the “sense and respond” paradigm. Here is a simple model which illustrates the basis for the reference architecture.
Such incidents pleasant and unpleasant stacks up in memory and in time the child can store “patterns of response” given a situation. As the child grows up the range of interactions with the environment gets complex. In time, it becomes evident that the child would need learning, supervised initially when the parents and teachers can guide him to respond more effectively to complex situations. As maturity and intelligence sets in the child could now handle situations independently and moves towards a model of unsupervised learning to respond to scenarios untrained for.
Now that we have the model, the questions arises so what might be the challenges in implementing this model for our enterprise. For starters let consider the following simplified representation of the information flow across a global mid-sized corporation.
The intelligence of the system would be efficient if we define a model which can identify the impact of the event triggered anywhere in the value chain and can be consolidated with the least latency and is available centrally.
Consider an event in this setting. An auto manufacturer identifies a machine breakdown bringing down the assembly line and consequently stopping the production process. As the production process is stalled, there might be urgent orders which would be impacted and may carry contractual obligations, requiring fulfillment via an alternate source. While the production line is stalled, there is a built of the “work-in-process” inventory and rendering your model mix planning and production schedules invalid.
Above is a simplified representation of the process illustrating what all scenario might need intervention at the instant. Critical operational handling of the event and the firefighting response at the plant floor brings back the production process online again.
So, what did we lose during this unforeseen event. Well typically isolated operational events gets masked within the layering of the corporate reports. In the figure below you would find just a snapshot of the reports and the plans which are prepared for the executives for decision making.
A quick look at the maze of planning process across departments soon makes you realize that the impact of the operational event which occurred above is practically untraceable. This is biggest challenge of the enterprise systems where the “sense and the respond” performance analysis is lost in the maze of reports, KPIs, dashboard and plans across the lines of business.
In absence of the effective traceability of the events, their response across the enterprise makes it nearly impossible for the systems to move to the next steps of learning, be it supervised or unsupervised.
How do we then solve this challenge? Based on the “Sense-Response” paradigm defined earlier for the enterprise the key measure would be the “Event to Impact Cycle Time” as illustrated below.
Finally, we increase the accuracy of the response by taking advantage of the rapid advances in AI. The technology advancement is propelling the classification of the images making machines more human like, with the sense of vision. With the advances in the natural language processing, soon our machines would be able to communicate with us the optimum response proactively like our own private digital assistant.
Note: As from the example illustrated above you could imagine the reference architecture is more of a whitepaper than a blog. In interest of keeping this blog simple I would request if you have further interest in this topic, please comment or send me a note indicating whether you would be interested in the detailed report and details about your role -end user, consultant or learner. Here are the previous blogs which may be of your interest as well.
Economics of Platform Commercialization
How embedded analytics unlocks the value of digital enterprise
Disclaimer: Some images are sourced from internet and is for illustration purpose.