In our personal life, our smartwatch is an example of a DSS. It helps us take decisions more intelligently on our lifestyle – diet, exercise, sleep. Another example is the customer feedback ratings that we see on almost all online platforms. These kinds of systems support us in choosing the right restaurant, or the right electronic product for example.
These examples should help us understand the role of DSS in decision-making. They provide additional data or information that helps us make better and more informed decisions.
In the past, these were merely smart systems. Such systems performed tasks based on a pre-determined algorithm programmed into them by a human. While the system did not require too much intervention from a human after it was programmed, it was not capable of learning and evolving on its own either. Today, we have intelligent systems (popularly termed Artificial Intelligence) that are capable of learning and evolving as they get exposed to data on an ongoing basis.
Using AI As A Decision Support System In Modern Businesses
A few examples will help illustrate this:
Predictive Maintenance of machinery
Maintenance is typically carried out on machines that break down. This is suboptimal and is really called repairs. On the other hand, we have preventive maintenance, in which the company periodically shuts down the machine and undertakes maintenance whether it is required or not. A more cost-efficient mechanism is to use AI algorithms that constantly monitor machinery using IoTs and evolve to predict exactly when a machine is likely to fail and therefore has a need for maintenance. This saves a lot of money for the company because unnecessary downtime, parts replacement, and manpower costs are minimised.
Risk Prediction in the financial industry
Companies use AI-based algorithms to identify loan applicants that are most likely to repay their loans on time. This helps the company reduce its NPAs. But it also helps them decide how much loan to provide to a person, or to price their loans at different rates for loan-takers with different risk profiles. AI helps a company hyper-personalize loan distribution.
Disease identification in the healthcare sector
The world of Radiology is being transformed with the advent of AI algorithms that scan MR, CT, and X-ray images to identify potential abnormalities within the person. These algorithms have learned to look for such abnormalities by parsing thousands of patient images – from both healthy individuals and individuals with an abnormality. AI algorithms then suggest to the radiologist that there is a suspicious artifact in the image. The Radiologist can then look at the images more closely and make a decision on the future course of treatment if required.
In all the above examples, the process is similar as far as an AI algorithm is concerned.
A seed algorithm is developed by engineers, scientists, algorithm specialists, and software engineers. This seed algorithm is then presented with thousands – if not millions – of data points to learn from. The algorithm parses all this data and changes its algorithm if required (the AI engine evolves).
This evolution of the algorithm continues even when the AI engine is placed in a production environment – and therefore continues to get better.
This engine is then presented with real-life situations and can make recommendations that humans can then use to take decisions.
Today this is the most common way in which AI engines are being used. However, we already have situations in which an AI algorithm makes the decisions itself, and implementation is left to the human or to another piece of machinery. We are beginning to see examples of this kind of algorithm in autonomous vehicles, industrial process automation, fly-by-wire aircraft, and dozens of others. In a sense, they are moving from Decision Support Systems to Decision Making Systems.
Advantages:
AI Engines do not skip data points or add additional data points that don’t exist (with humans, we have errors of omission and commission)
If an algorithm works, it is by and large repeatable – unless it evolves of course
They learn far faster than humans can ever hope to do
Disadvantages:
Bias can creep into an AI Engine across the complete chain – the algorithm may be biased when the human writes it first, there may be bias in the kind of data fed to the engine for it to learn from, and the interpretation of the data that is used to reinforce the learning may also be biased.
Another big danger is that the human expert may abdicate his/her role. An AI-driven output is used as a decision in itself – rather than as an additional aid in supporting decision-making.
At some stage, it is likely that the algorithm has evolved so much that it is not recognisable against the initial version. The users then lose a sense of control or understanding of why the AI engine is working the way it.
What Next?
Whilst AI engines do have their own strengths and shortcomings – we must remember that a tool is only as useful as its user. The reality is that organisations will continue to invest heavily in these kinds of tools – both at operational and strategic levels. All of us must gain a better understanding of the power of AI and how we can leverage it to the advantage of the company, its customers, and its employees.
AI is here to stay – it’s time to join hands with it.
This article is written by Bhaskaran Srinivasan, Chief Academics Officer, UNext Learning