Recently, while scanning through the newspaper, my eyes got stuck on a small article in one of the internal pages of the paper. Normally, I skip such small articles as it doesn’t carry much importance compared to the other articles, however, the said article seemed fascinating and hence I took time to read the same. It was titled “India tops the charts in workplace AI adoption.” Fascinatingly the article was more interesting than the catchy title given to it. According to the BCG’s “AI at work 2026” report, almost 95% of the frontline employees in India use AI several times a week. This led my mind to think. Has AI become so instrumental in our day-to-day life? Apparently, the answer to the question is not a simple yes or no. It also includes a lot of may be in it.
Going back to a decade, AI was not even a topic of discussion on a table. If someone had asked me, what I expect the future is going to be, then my genuine answer would be flying cars, finding aliens, or time machines. While these things remain fantasies even today, AI or Artificial Intelligence took the floor subtly announcing that I am the secret super star of this show. With even a simple step being carried out through AI based software or programme, it is evident that AI indeed is the Master of the show.
The invention of AI and related technologies made a massive penetration in the world within a short span of time. It expanded to a scale where no work is now impossible to be carried out in a given time, provided sufficient data is available. Companies and organizations at large found that, while the initial cost or investment in developing an in-house AI based software or platform is high, it can bring down the operational cost significantly, resulting in operational profitability.
In a mere 1–2-year period, a large number of organizations including the banking industry embraced and explored the potentials of AI and introduced various technologies/ technological upgradations to enrich the sector.
As part of cost optimization and customer experience, the Banks are now-a-days leveraging heavily on AI tools. Data fetching, loan processing, and even the identification of potential NPAs and frauds are also being carried out effortlessly by organizations such as Banks through in-house or pre-developed and customized tools which are supported by automation or AI techniques.
For example, Banks are now having systems which can sanction loans in minutes than weeks. Instead of waiting for several days of manual verification followed by hundreds of documentations, the AI system can help one in reviewing the bank statements, analyzing the cash flows, and can also track the customer repayment history with basic and minimal data fed into the system compared to the long process involved in manual process. Chatbots such as those used in WhatsApp Banking etc. also makes the customers convenient to operate his/ her account without even visiting the branches or without going through tiresome procedures.
Thus, in Banking, AI and automation are making ground breaking innovations leading to operational optimization and customer satisfaction.
This brings to the second question, while we believe that AIs are capable of carrying out tasks more precisely and reasonably than humans, are they really resistant to errors? And the most realistic answer to the question would be a big no. And this is where the actual question comes in, we have spent decades auditing people, processes, and systems, but who audits the algorithms, the codes that are influencing the critical decision making in the Banks?
This is where the new discipline comes into place, Audit of AI or audit of Artificial Intelligence.
In the above given examples itself, suppose a customer reaches the Bank for availing a loan. He approached the concerned staff and submitted the necessary documents. The customer is a long-standing customer of the Bank and the staff is also well confident from the documents that he is eligible for the loan and is fully capable to repay the loan when its due. However, the Bank’s latest AI software for loan sanctioning, rejected the loan based on some relatively trivial reasons. While the action can be justified to a greater extent, stating that the call was a precautionary measure, what if the opposite had happened? What if, an ineligible customer loan gets sanctioned and it turned out to be a fraud in a future date?
While supporters can argue that the chance of such an error happening in an otherwise perfectly drafted code for a system is near to impossible, the fact is that, all models work based on certain programmes fed into it, and if the code or the programme is having a bug or is wrong, the output will also go wrong.
Although it’s true that tasks carried out manually by humans are susceptible to higher amount of risk, the same are more or less predictable and can be evaluated easily. However, when it comes to tasks which are automated, or where AI is used intensively, the chances of identifying what went wrong and where will be like looking for a needle in a haystack.
Nevertheless, if Artificial Intelligence is to be trusted with critical decision-making, then a structured approach to auditing it becomes essential.
Artificial Intelligence, while itself is something created by the humans, is like an iceberg floating in the sea. It holds much more than what is visible to the users. Hence, the first step in audit of AI itself should be understanding the model. Not just understanding the model for what it is, but also the decisions being influenced by the model and the rules being fed into the same. This will help the auditor to understand what to look for and where to look and the level of risk involved. Secondly, a detailed data audit can help the auditor to answer certain other questions, such as whether the data is accurate, complete, and free from bias, because if the data itself is biased, the outcome generated will also be biased in nature.
Now, think it like this, if the algorithms and logic behind the system are complex or loaded up with jargon will it serve any purpose? Which indicates that the assumptions and parameters used in the system should be reasonable enough so that the auditors, or anyone other than the programmers themselves, should be able to understand to identify any anomalies present. In addition to this, output testing also plays a key role while auditing AI. As we discussed earlier, there can be instances of false positives or false negatives, and hence stress testing with different scenarios becomes essential to ensure reliability of the model in place.
We might have seen the ‘Chitti Robot’ in the movie ‘Enthiran’ go rogue when a different programme was fed into the system, and how the professor turned it back to its original form and then dismantled the same. This brings to the last but not the least important aspect, which is, governance and continuous monitoring. Even if the system is highly advanced or believed to be fools proof, there should always be a human oversight for critical decisions, with proper accountability mechanisms in place.
If anyone is asked about which is the most fundamental and greatest innovation in human history, while there might be many answers at first, the loudest will always be the innovation of WHEELS. Why because wheels changed the pace and scale of human progress forever. Similarly, if someone asks which is the greatest innovation in the 21st century, the answer would be the introduction of Artificial Intelligence or AI.
Still, as we continue to rely more on these systems, it becomes equally important to question their accuracy, fairness, and accountability. The discussion is no longer about whether AI should be adopted, because its already there and is going to remain there for a long time too, but about how responsibly it should be used and how it can be governed. Because at the end of the day, it’s we the human who are going to be accountable even if the machines may learn and adapt.
