Creating Trusted AI for Business (Ana Paula Assis, Marc Al-Hames) | DLD 24

All right, so good afternoon. It's afternoon already? No, it's morning. Good morning, everybody. What a great crowd, huh? Yeah. Fantastic. A lot of conversation about trusted AI. Do we need regulations? Don't we need regulations? We're going to go a little bit deeper on that. But let me do just a little bit of reflection. It's a new year, but the conversation is the same. And I don't think it's a bad thing. Remember, we entered 2022, it was all about the metaverse. Do we talk about the metaverse now? No. Not really, right? Now, we started 2023 with the explosion of ChatGPT, and we are still talking a lot about AI and generative AI in particular. So 2023 was really the year to see the art of the impossible, right? Everything that we could do with AI, a lot of experimentation, a lot of POCs, a lot of pilots running around AI. If we look into 2024, 2024 will be about how you scale AI, how you make sure that you drive value from AI, and how you manage risks and create trust. I think those are the key elements that are going to be the proof points of whether this technology is just a hype or something that is here to stay. So let's start with the value points. We have already identified very clear— clearly, basically 3 use cases where there is demonstrated value for AI. Digital labor, basically automating business processes, infusing AI in applications to accelerate the execution of tasks. Customer service, the famous chatbots, right? I mean, how you can address a client's request, answer a client's question in a way that is faster, that is more accurate, that drives higher customer satisfaction, and finally IT operations. And in that part, I would include the development of code. Now, I've been talking to a lot of clients in the beginning of the year because I really wanted to see, well, what are you seeing already, right? You've done a lot of experimentation in the past few months. Give me some of your conclusions. Let's look at the application development component. Most of what I'm hearing is not that the code generation, the automatic code generation that is generating the biggest value. It's how I do better documentation, how I automate test procedures, how I use AI to create synthetic data to run my tests. So we are starting to fine-tune really where we can get value from the technology that probably is not what we initially thought it was going to be. Now let's talk about scaling. Scaling AI has to consider, I would say, 3 major aspects: skills, and we are starting to see skills to be a big barrier for scaling AI, data quality, We just released a report where one of the key items that clients are mentioning as a hindering factor to expand AI is data complexity. So how to make data more accessible, and finally, how you truly create a platform. I was with a client at the beginning of this week in Dubai, and he was saying, I operate in in tens of countries around the world. My business is a multi-industry business because I do telco, but I also do payments, I also do services. How do I create a consistent platform that wherever I am operating in the world, I have the same look and feel, I have the same compliant requirements, and I meet the legislation of the countries that I'm operating in and the industries that I'm operating in. So this concept of a platform, I think, is going to come very strongly in 2024. And by the way, that's the approach that IBM took to develop its own platform, Watsonx. And finally, the topic of safety, managing risks, and ensuring that you build trust. That's all about governance when it comes to the enterprise. That's all about making sure that you understand what are the principles of ethical AI adoption, that you follow those principles in a consistent way, and most importantly, that you do it also with the support of technology. If you have to have the human in each step of the lifecycle of AI, it's going to be impossible to scale AI. So you have to have a platform that supports the automation of those processes so that you can meet the requirements that you will need to have regardless if you have a regulatory environment or not. It is the society, it is your client that is going to demand that you have a responsible usage of the technology. Regulations are probably going to create the standards, but the demand is going to come from the market. So with that, I'll invite my colleague and to ask a few questions so that you can go a little bit deeper. Martin, great to be here with you.

Ana, it's great to have you here. Actually, I've seen you with Thomas Saueressig of SAP, our friend from the DLD community, back in autumn on the DLD on AI. And I thought we should have more of you here. Little did I imagine that I am supposed to ask the questions.

Wow, fantastic.

So this will be fun. You touched on enterprise AI and everything is developing very, very fast right now. If you would go out to the street and ask people what is AI, their answer would be ChatGPT. That's what people understand. There's much more. Aleph Alpha. There's enterprise AI. From your perspective, you're dealing with a lot of enterprise clients. What's the key difference? It's not the models, huh?

No, the models, they are available pretty much to all the sides, right? You can do it from a consumer standpoint, you can do it from an enterprise standpoint. The large language models are available in different ways. They could be proprietary models, they could be open source models. What is different when it comes to the enterprise is that, well, one of the things that we learned as we were experimenting with generative AI is that sometimes it makes things up, right? It doesn't necessarily give you the right answer. So businesses cannot cope with that. They need accuracy. They need something that can be demonstrable, that you can have the evidences of how the algorithm reached to that conclusion. So the stakes are much higher. You have to handle privacy of data, right? If you're working with a— with one of your clients and they provide any type of personal identifiable information that you have to account for related to GDPR rules, for example, you have to make sure that you are responsible with that. So the stakes are higher. You have to make sure that you understand all those aspects before you really release your applications out there.

You talked about privacy and earlier in your keynote, you also talked about data complexity. But if I look at the enterprise environment, we as a publisher, we publish our content and then every model can just insource it, use it, get better. And there will be— there are actually lawsuits. I would imagine the typical enterprise customer would not feel well about handing you all the data and say, look, do whatever you want, but give me back a good model.

Yeah. No. So that's why I think that when it comes to the way that we work with the corporations, It's really creating your own instance. I mean, of course you have access to the platform, but the data that you use on the platform to train your applications, it's your data. We are not going to take advantage of the data to provide other services to other clients. So that clarity is important. When we look at our own models, the models that IBM is training and it's creating that are called Granite, We are using really our own data to do that, right? And in a way that is absolutely verifiable to the point that if a client eventually faces any type of claims or lawsuits, we are going to indemnify that client because we believe that we have total evidence that that data is trustworthy. So that is how you really implement a safe environment for companies to use their data with a guarantee that that data is not going to be leaked or used for other purposes.

Do clients believe that? You must meet a lot of clients. If you talk to them on a scale from 0 to 10, 0 being total fear and 10, let's start yesterday. What's the mood in the companies?

It's very interesting. I think we had one year of a lot of explosion and experimentation. Now that these things are going into deployment mode, go live, actual usage, and we start to have really the discussions around, okay, let's put it on a contract. This is now where all these topics are starting to come up, right? How do you ensure protection on copyright? How do you ensure that, to your point, the data that has been consumed by the platform is not going to be improperly used or misused? So all these things are happening right now. And in my view, it is going to be very important that the companies, the corporations, make a very thoughtful decision on who they are going to partner with, because they will have to have partners that provide those services and services.

This is where you advertise IBM, huh?

Yeah, I'm going to advertise IBM now. Yeah, I mean, we have a lot of experience in dealing with clients' data in our history, right? This is something that we haven't started now. Actually, artificial intelligence is something that we have been working since the '50s. But this topic of making sure that the data is responsibly used is part of the DNA of the company. And that's really how we approach all these implementations with our clients.

Thank you. Will the regulatory bodies, European Union help? You're working for a US company, but you're responsible among other countries for Europe. Is the regulatory framework that we're setting up helpful, harming?

Yeah. And I think there was a comment here previously in one of the panels. We welcome very much the risk approach that was taken by the European Union on the AI Act, because you have to regulate the risks of the technology and not the technology itself. So I think the CEO of Aleph Alpha was saying, this is math, right? You cannot regulate math. It's a little bit like that. So you really have to understand what are the usages where AI could really create harm and regulate those usages and not the whole technology itself. The other aspect that we are very proponent of is that you really create the regulations around providing the liabilities and the responsibilities and not so much licensing, because if you go with licensing, then you're going to stifle innovation. And finally, you have to have a very clear differentiation between the responsibilities of those who develop the models, who develop the platform, and the ones that deploy and the ones that use the platform. So having that clarity on roles and responsibilities, I believe, is going to simplify and create a more, let's say, practical framework around regulation.

I'm not sure the CEO of a mid-sized to small company with a 1,000-employee headcount would understand what you just said. If you would be in a sales pitch and you would say, look, we apply European regulation, we bring in all the power and experience of IBM, this is why you're safe. Yeah.

So we do that really providing a platform that automates a lot of those regulatory requirements. Later today, I think we're going to be presenting a demo of Watsonx.governance. I encourage all of you to go see it because you're going to have clarity by looking at how the product operates, how you can really have a visibility of how the models are running in your organization, how the test has been done. You create a fact sheet of your model like You look at this bottle, what are the components? You're going to have a fact sheet on your model on what data it was used to be trained, what does it do, what are the thresholds and the weights of your algorithm. So all these things will be there. So we provide this as a service, right? I'm not going to come here and say, oh, you have to implement this or that. No, here is the solution. It's an as-a-service model. You contract from us and we are going to provide you those capabilities.

And then coming—

did I sell it to you?

Yes.

Yes. Okay.

But our company is bigger, so I understood it.

No, no, no. Yeah, yeah, yeah.

But coming back to the regulatory question, if you have all this in place to convince me as a small company with all the transparency, coming back to the regulatory question, do we then even need the regulatory body? Because you just pitched that you have it all implemented. No, you need the

regulatory body because you need to create the standards, right? The regulations, they ensure that you have— you level the playing field, right? What I would— the first recommendation I would give to this small company CEO is try to identify what are the applications that you're going to develop that can really create harm to the society or to your client, right? I mean, you are developing a solution that could be used to provide, to proliferate misinformation. DLD, you guys are in the media space. Yes. Right? You have responsibility to ensure that the information that you are providing is accurate. And misinformation is one of the highest risks that are part of the risk level of the regulatory framework from the European Union AI Act, in the sense that we have to We have to safeguard the democracy, right? This is the basis. The basis of democracy is trustworthy information, and AI cannot compromise that. So the first thing that a CEO has to do is identify the levels of risk of the use cases that they are operating in, and then they will need to adjust to the regulatory requirements based on that, because not everything is going to be high risk.

Yeah. You're not only responsible for Europe, but also for Middle East and Africa. Anyone from Africa or Middle East in the audience? Oh, Steffi, we need to work on that. We need more.

We need more.

Oh, they missed the most important piece. If we would jump into a plane and fly to Africa or Middle East, AI, a topic over there?

Super topic. We just released our AI Adoption Index report. We do this every year. Interestingly enough, if you look at Middle East, if you look at Asia, the adoption level is 2 times higher than pretty much any country in Europe. We need to watch out for that.

We're losing out.

We're losing out. We need to speed up, we need to accelerate. A lot of work, and I see tremendous potential for AI in Africa. To solve some big humanitarian social issues, right? So for example, we have 2 laboratories, 2 development labs in research labs in Africa, one in Kenya and one in South Africa. They have been instrumental to us to develop a tool that is called 360 Fairness that ensures that AI is not biased. And the team there contributed a lot to us in terms of creating the parameters to implement unbiased AI, right? Right. So tremendous opportunity. We are working with Melinda and the Gates Foundation around using data to better track focus of malaria expansion in the, in the, in the, in the continent. So I'm very, very hopeful and I expect a lot from that region in terms of adoption of AI. We saw that happening with mobile, right, with, with, with mobile phones. in the areas of payment, in the areas of creating really new business models using mobility. Now I think it's the time for AI.

Really good to hear. And for the next DLD, we really need to get the Middle East and South Africa to be here.

Absolutely. I'll bring them.

So next year we do a session together. That's even better. One last question because we run over time. Make a prediction 5 years down the road. How much of IBM's revenue will be related or in some way has touched AI?

Well, I'm not in a position to make— to give guidance for the company, but what I can say is that this is the biggest bet that IBM has made in the past year. And this is going to be about how we provide the technology to our clients, but a lot about how we transform our own portfolio using AI.

Thank you so much for the conversation.

Thank you, Mark. It was great to be with you.

I make a bet for you because you cannot give the guidance, 80%. Wow.

Let's see. Thank you.

Thank you so much.