David Cushman — HFS Research[00:21]
Welcome, Arnab Chakraborti, the new Chief Responsible AI Officer at Accenture. Welcome, Arnab. First question: where does your role sit in this vast universe that is Accenture, and how did you end up in it?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[00:37]
Fantastic, David. Great to take that. It’s a privilege for me to play this role for Accenture as the Chief Responsible AI Officer. This is a new role that has been created at Accenture and directly sponsored by our CEO, Julie Sweet. This sits within the broader Data and AI practice at Accenture, which is a huge focus area for us, and I have been part of the Data and AI leadership team from the very beginning, for the last 12 years. This came about in the last 7 to 8 months, when I was asked to take on this role and build it out as a business focus for Accenture.
David Cushman — HFS Research[01:20]
What are the risks of using GenAI in the enterprise? How are these any different from the kind of risks we’ve faced when previous technologies have swept through our businesses — the arrival of the web, or bring-your-own technology, or early applications of AI, mobile phones, and so on? Why is this so different and special?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[01:33]
I think this moment is different, for a few reasons. Number one, the speed at which this technology is evolving is very different from what we have seen in the past. The second big difference is that this technology is much more democratized, which means it’s in the hands of millions of people. And that’s important, because you need to manage the technology very responsibly when it’s in the hands of millions of people. The third dimension is that this technology has a huge promise of transforming organizations back to front, front to back. Sixty percent of organizational tasks and activities can potentially get impacted by large language models, and that is a huge opportunity for value creation. And with that comes significant risks as well. The number one risk I see with generative AI and LLMs is the confidence with which it gives out outputs. When you give outputs with such a high degree of confidence, there can be hallucinations, and that can erode trust with customers and employees. The second big area of risk I see is around bias. Generative AI gets trained on large amounts of public and internet data, and there could be bias in that dataset that propagates into your models and leads to discriminating actions against certain portions of protected societies. And then the third big risk I see today is really about security. How do we make sure these AI systems are not vulnerable to security attacks? How do we make sure there is no infringement of IP and no infringement of private data going out?
David Cushman — HFS Research[02:50]
That’s quite a long list. So to what extent do you think the existing enterprise governance capabilities that have been built up over many years, over many technologies, are fit for purpose for the challenge of GenAI?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[03:20]
This is a big opportunity and a challenge for organizations right now. We did some research about 6 months back, and we asked organizations how important it is for them to address the big risks that generative AI brings. Ninety-six percent of the C-suite executives highlighted that it’s one of their topmost priorities to address. Then the second question we asked was: what is the level of preparation and operationalization of responsible AI within your enterprise, end to end? And that number came down from 96% to 2%. That’s the gap today — from talking about responsible AI and its importance to actually operationalizing it into real practice, from 96% down to 2%. That’s the gap that has to be covered. We have done over 1,000 GenAI projects in the last 8 to 9 months, and the learning from that is that organizations have got the principles, policies, and standards for responsible AI in place to govern. But that sets the guidelines, and some of it is theoretical, some of it is paperwork. Converting that paperwork into tangible actions that organizations need to take is the journey organizations are going through right now. At Accenture we have been on that journey over the last 2 to 3 years. This is not a new topic for us — we’ve been on this journey with our own responsible AI compliance program across 750,000 employees over 120-plus countries. This requires significant senior leadership sponsorship, but also practical execution: setting up the right processes, the technology, the tooling and the infrastructure, as well as training and enablement of talent across the organization.
David Cushman — HFS Research[05:15]
There’s an awful lot to take in there. One of my concerns, though: you’re saying there’s basically 2% of enterprises actually equipped to scale GenAI. Is that actually the experience you’ve been having?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[05:45]
That’s exactly right. If you look right now at how many organizations have actually scaled GenAI, you can count them in numbers. One of the main reasons they are not able to go beyond the pilot to scaling and industrializing it is because they don’t have all the guardrails and governance processes established.
David Cushman — HFS Research[06:12]
It seems to me what you’re saying is that governance must stop being a thing that’s done by committee and become something that is actually built into the whole process of delivery, particularly around GenAI. I’ve seen similar — we’re seeing that a lot of enterprises get stuck in this POCs-and-pilots piece. But some are overcoming that; that number’s increasing. The number of companies battling their way through these challenges is increasing, but I guess in order to make it repeatable and scalable, it needs to be baked in. So what are the steps we should take to fill those gaps — that 2%-to-96% gap? How are we going to close it so that we can really enable the responsible use of GenAI at scale?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[07:18]
It’s a journey; it’s never done. I have the privilege of sponsoring that for Accenture. We’ve been doing this for the last 3 years, as I said, and there are 5 things we have learned from that. It starts with creating your own standards — not waiting for the regulations to come to you, but creating your own standards and principles for governing AI within the organization. That’s where it starts. We created our own AI policy and AI standards much earlier, even before the European Union AI Act draft came out, and we created that in line with our code of business ethics and our values.
David Cushman — HFS Research[07:44]
That has been a challenge for enterprises, because there’s been a concern — ‘I don’t build anything in, in case the carpet gets pulled from under me.’ But it seems to me that actually it’s kind of priced in: when this legislation comes along, it’s pretty much in line with expectations, nothing’s surprising anyone. So is that what you’re suggesting — we should just get on and do it?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[08:11]
My advice is we should actually take a self-regulation angle, where we are being more proactive to set up our own policies and standards based on the industry, our values, and our code of conduct. That can play a big role in influencing and informing what regulations policymakers put in place. So that’s step one. Step two is very fundamental, David: it’s about knowing what AI is happening within your organization. It starts with that fundamental question, and when we ask it, we get different answers. If you have 5 different answers to the question, it’s clear you don’t have a good handle on all the AI you are developing yourself or with third-party suppliers. The next question is: what’s the level of risk associated with that AI? The third big step is how do I test my AI systems that are being developed — for bias, for explainability, for transparency, all the risk dimensions we talked about — so that I can be confident that, as that AI gets ready to be deployed, it has gone through rigorous responsible AI testing? You need a framework, tooling, and a system to do that. It cannot just be a questionnaire-based assessment; it needs to be very surgical, using technology at the heart of it. Once you’ve done that, it’s important to understand that this is a very dynamic technology. The data is shifting, the models are drifting. How do you create an oversight and monitoring mechanism to observe how your AI is performing over time — whether there’s data drift or model drift? As soon as you see those, there needs to be surgical intervention to address it. So that’s the fourth step. And then the fifth step, which organizations often miss and which needs to be factored into your AI governance and responsible AI journey, is: how do I understand the workforce impact? How do I understand the environmental sustainability and carbon impact of my AI models? And how do I understand the security and privacy impact? These have to be baked in as part of your responsible AI, so that you are being responsible by design, not just on the technology.
David Cushman — HFS Research[10:24]
It’s almost as if you have to get this in place in order to have any chance of succeeding with your POC, or pilot — particularly the pilot step. There seems little point in proceeding with a pilot if you haven’t got all of those governance elements in place, which is quite a lot to ask an organization to do before it’s got a proof of value. So how have you been handling that in your engagements?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[10:53]
Don’t wait to start — first of all, you need to make a start. Say, ‘Let me get my principles, policies, and standards in place.’ A lot of our clients actually still have a lot of traditional AI going on; it’s not all GenAI. Establish that for your traditional AI. Make sure you’ve got your foundation in place. Make sure you have a risk assessment framework in place even for your traditional AI — that’s how we started 3 years back. There was no GenAI back then. Then you need to look at what uplift you need to do on your principles, your risk assessment framework, and your testing framework to take care of the GenAI considerations. And that, to me, is a lighter lift, because you’ve already laid the foundation. So it’s more the uplift you have to do. Then, the way we are suggesting to our clients: once you have this foundation in place, let’s think about the use cases of GenAI that you want to pilot. Make sure you actually have them go through this governance mechanism and the responsible AI foundation you’ve established. Two things happen with that: your initial set of use cases go through it, and the learning coming out of it informs how you now take that framework, approach, and tooling to the hundreds of use cases coming down the hopper over the next 6 to 12 months. And that’s how you start building the muscle.
David Cushman — HFS Research[12:03]
Arnab, I think this is going to keep you very busy for quite some period to come, I’m sure. It’ll be interesting to see whether your role is repeated in enterprises — whether we start seeing the emergence of responsible AI officers in the enterprise. Have you seen any examples of that so far?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[12:29]
Yeah, I think it is. First of all, this topic has gotten a lot of attention with the boards. The boards, with their responsibility and accountability, are asking the CEOs: what is your action on the topic of AI governance and managing the risks of AI that come with it, and how are you making sure it’s embedded inside your organization? So this is a huge board priority that is coming down to the CEOs. My belief is that the organizations that are going to be leading in their industry around AI will need to put a huge amount of focus and investment into this, and that starts with having the leadership in place to do it. It’s not a plus-one. It is the job. It is such an important priority. So I believe this is going to happen with the leaders in the industry. They’re going to be appointing — like what Accenture has done, we are probably the first of our kind in our industry — and I think we’re going to see that pattern in our clients’ industries as well.
David Cushman — HFS Research[13:33]
Do you think the roles of the Chief AI Officers and Chief Responsible AI Officers will replace any of the elements of CIO or CTO jobs, or, as you suggest, are these additional to the requirements and responsibilities of the business?
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[13:59]
Yeah, I think it is a function of the maturity of the organizations. Depending on where organizations sit in their maturity level around data and AI, and the focus and impetus they put on it, I see this as having the potential to become a dedicated role — the way we have done it at Accenture — instead of having it as a plus-one layered under the organization. So I think it’s a function of maturity, and the next 12 to 18 months is where we are going to start seeing these new roles.
David Cushman — HFS Research[14:29]
It’s a fascinating area — thank you for your time. I’m wondering: roles often come and go, right? The Chief Digital Officer was playing for the moment; maybe there aren’t so many of those about as there once were, but it waved a flag that this was important to the organization. I suspect we’ll see a bit of that over the coming 12 to 18 months in enterprises that want to illustrate to the market, at least, that they are taking this seriously. So if nothing else, you’re maybe blazing a trail for that. So thank you for that. We’ll be watching closely, and we really appreciate your time today. Thank you very much indeed.
Arnab Chakraborti — Chief Responsible AI Officer, Accenture[15:12]
Thank you, David. Great to be here with you.
