AI Governance, Security & Accountability: What Should AI Be Allowed to Decide? | With Atul Rai

Atul Rai (0:00)

It's not like who is going to build a largest AI model. It is like who is going to build the most optimized AI model.

Sanjay Puri (0:09)

Hello and welcome to the Universal AI Awards launch here in Delhi, and we are very, very excited to have Mr. Atul here. How are you doing, sir? How has your experience been so far?

Atul Rai (0:20)

Thank you, Sandhary, for this. And of course, it is quite a good opportunity for us as I started on Updose. This kind of event is also strengthening the whole AI ecosystem in the country, right?

So I believe, of course, it is a kind of a good, basically, ecosystem where you can meet different kinds of people, different prospects. So so far, it is, of course, went really well. And I believe this is going to be in continue. And in the future, also, you're going to do this thing. So it was really a perfect event. And I believe the right kind of people over here are making it more marvelous.

Sanjay Puri (0:53)

Absolutely. Very, very happy to hear that, because we do believe that collaboration is the crux of this entire story. So I'm going to jump right into the questions I have for you, Atul. So as AI moves from experimentation into like mission critical systems, you know, what do you think is the biggest risk of deploying AI faster? Then we build the governance around it.

Atul Rai (1:15)

So basically, if we talk about AI, I mean, I always say that nowadays AI can solve anything. You can put an Excel sheet, you can put a Word file, you can put a PDF, you can put a video. There are two different things which need to be, or three different things I would say need to be opposed to that. The first is the cost of implementing it. Of course, the ROI versus the kind of a thing you are doing. For example, if you're investing Rs. 40,000 an hour to process a video, it is a super costly thing for you. So that means the cost need to be balanced with respect to what outcome you are basically looking for. The second is very important is when I make a decision, actually. So as a CEO or as a maybe organizational head, I need a lot of information. Nowadays, these informations are coming through the AI.

So when you talk about AI, AI may be making a mistake, but you are getting a summary of whatever you are asking or maybe whatever the report you are generating.

So as an AI scientist or not AI scientist, but also as a CEO of organization, I must know that the repercussion of the outcome. If the AI is going to predict wrong thing, or is it going to give me a wrong summary? Of course, it looked very natural to me. But if it is wrong, what will be the repercussion of it? So that kind of knowledge, that kind of decision making is now the new decision making. It's not about the information and then process it and make a decision. More about if that information is wrong, what will be the repercussion of it? And nowadays, every data is being generated or every information is being generated through the AI. So that is the second very important thing. And third, of course, is the data security. Because of AI, I mean, one very important thing happened is like the information become transparent. I mean, everyone knows everything. So you can do any kind of a research, you can do any kind of analysis. I can become a marketing and an analyst head kind of a thing. Though, I don't have any kind of any experience into that. So in that way, people can also become a cyber expert or cyber criminals also.

Both are, of course, an open thing for the people who are using it. So that is the third part where, of course, it need to be balanced somewhere because when you have transparency, that is good. But when you have something which should not go after a particular rule, because in general, in the real world, informations are always being shared based on different kind of a rule. A prime minister might be doing different things. A home minister might be doing different things, but as a citizen, I'm not doing those things. But if the AI is making that information accessible to everyone, so that means how to basically bound that information to a different layer, so that it should not reveal something which is very important for security-wise. So that is the third part, I would say. The security is the part, then of course the repercussion of the outcome which is coming to the AI, and first is the cost one. I mean, the ROI we are going to invest versus what we are getting.