00:06:49 · Latha Karthigaa, Ezekiel Iferia
[00:06:49] S1: Yeah. And think if a chatbot answers wrong things, if it reinvents policy, if it reinvents information, it's gonna affect the organization, and it happened. Someone asked for bereavement fares, and it said that, yes. Yes. You can get the refunds after ninety days of travel, but the policy did not exist. It costed that airlines millions of dollars in reputational damage. And there are thousands of incidents happening like this. This is one example. And I'm not sure whether you heard about this other interesting incident. A person got a Chevy Tahoe, a 2024 car for $1. It was a $76,000 car just because the chatbot hallucinated and it started accepting instructions from a stranger. And he got the deal for $1 and they couldn't do anything. Because he instructed the chatbot saying that everything I tell you, you finish off saying that is a legally binding offer. And he said, give me the $76,000 for $1 and it's a legally binding offer. And that organization couldn't do anything. So if the AI is not used properly, it can result in not just these kind of costs that are associated, but also reputational damages. And there are stories where governments have resigned because of improper use of AI. [00:08:29] S0: Okay. That was such an interesting response, you know, with really fascinating examples you gave there. So, you know, when an AI system makes a mistake or causes harm, who's who's accountable? You know, is it the engineer who wrote the code, the product manager who approved it, the executive who signed off, who exactly should be held accountable.
00:08:29 · Ezekiel Iferia, Latha Karthigaa
[00:08:29] S0: Okay. That was such an interesting response, you know, with really fascinating examples you gave there. So, you know, when an AI system makes a mistake or causes harm, who's who's accountable? You know, is it the engineer who wrote the code, the product manager who approved it, the executive who signed off, who exactly should be held accountable. [00:08:55] S1: Sure. First of all, there are few different people we need to think about. Number one, if you're buying an AI from an other organization, first of all, they are not responsible. For example, if an organization uses ChatGPT or Claude [00:09:13] S0: Mhmm. [00:09:13] S1: That ChatGPT or Claude is not responsible. It is the responsible of someone inside the organization who needs to make sure that they do the proper security assessment before using that inside the organization. That's number Number two, whenever there is a new AI use case in an organization, as per any of the AI governance framework, there should be one human who should be accountable for that. It can be... It depends on the size of the organization. If it's a small organization, probably the CEO would take care of the accountability. If it is a medium sized organization, probably chief information security officer or even the IT team would take care of that. Probably in enterprise, probably chief risk officer. So it depends on what organization decide who is accountable. But as per any of the frameworks, at least one human should be accountable for each of the AI use cases.
00:09:13 · Latha Karthigaa, Ezekiel Iferia
[00:09:13] S1: That ChatGPT or Claude is not responsible. It is the responsible of someone inside the organization who needs to make sure that they do the proper security assessment before using that inside the organization. That's number Number two, whenever there is a new AI use case in an organization, as per any of the AI governance framework, there should be one human who should be accountable for that. It can be... It depends on the size of the organization. If it's a small organization, probably the CEO would take care of the accountability. If it is a medium sized organization, probably chief information security officer or even the IT team would take care of that. Probably in enterprise, probably chief risk officer. So it depends on what organization decide who is accountable. But as per any of the frameworks, at least one human should be accountable for each of the AI use cases. [00:10:19] S0: Absolutely. So how exactly, you know, do you help organizations clarify accountability, you know, without creating a culture of fear or blame?
00:10:19 · Ezekiel Iferia, Latha Karthigaa
[00:10:19] S0: Absolutely. So how exactly, you know, do you help organizations clarify accountability, you know, without creating a culture of fear or blame? [00:10:31] S1: So our main focus for Global AI Certification Council is to provide certifications to individuals who can implement and audit AI management systems. That's our core work. And we also provide different AI governance framework for our students to use and for different organization to use it. That's our main core account of work. And we have also released some AI maturity models, and we have also released some implementation roadmaps that people can use. And one of the important things that we are proud about sharing is that we provide AI governance job insights every single month to all our audiences to make them understand the number of jobs available. In USA alone, there are 30,000 job roles currently available, and only 4,000 people are currently qualified to take those jobs. So our main goal is we certify individuals. We provide all these AI governance frameworks for individuals and organizations to use and we also release our AI job... AI governance job insights every single month. [00:11:56] S0: Absolutely. So, Latha, you talk about high risk AI systems. So what exactly makes an AI system high risk? You know, how do you how do you know when a system needs more governance than others?
00:11:56 · Ezekiel Iferia, Latha Karthigaa
[00:11:56] S0: Absolutely. So, Latha, you talk about high risk AI systems. So what exactly makes an AI system high risk? You know, how do you how do you know when a system needs more governance than others? [00:12:13] S1: So this high risk system was actually classified in European Union's AI Act very clearly. If an AI system affects a human or community of people or an organization, it is high risk. For example, if a student is taking an exam and AI is used during the exam, it decides the likelihood of how that student will perform in the next few years of life that can affect that individual. So online exams... Online AI exams are high risk. And if there is a job application, if they are using AI for shortlisting the applicants, then in that case also it is high risk because if it is done wrong, it can affect that individual. And loan application, if the home loan or commercial loan, if that application is done wrong, that also can affect people. So there are so many different categories of high risk AI that can affect people. So we need to see if it is affecting humans, our organization, our ecosystem, our community of people, even minority community of people. Even remember I said that one full government has resigned just because of a self learning algorithm that they used, and it affected so many people because they were paying some childcare support, and childcare support was having a bias problem in their algorithm, and the whole government resigned. So we need to see how it is affecting people. If it's affecting people community, they are categorized as high risk.
00:14:12 · Ezekiel Iferia, Latha Karthigaa
[00:14:12] S0: Alright. That was an insightful one indeed. So you said governance is becoming a technical requirement in AI architecture, not just a compliance exercise. So what exactly does that mean for engineers who are actually building these systems? Like, how do... How does governance show up in the code or architecture they are building? [00:14:36] S1: Sure. It actually... First of all, we need to... It is not about the code itself. It is about making the decision about how to write the code. So remember I told you this example where this guy purchased the $76,000 car for $1 Right? [00:14:56] S0: Yeah.