AI engineers are in high demand – but what is the job really like?

They’re some of the world’s  most in-demand tech jobs. . .

These are synthetic IDs. As good as a genuine ID. We can see that it didn’t understand the phrase.

Took it literally. It even read my handwriting. Is this what people mean when they  say that AI is going to take our jobs?

. . .

but what does a day in one of  these roles actually look like? From fighting scams and fraud, developing  large language models from scratch, to designing chatbots, they’re all  working with one technology: AI. This is Most Wanted.

I’m going to show you some cool stuff, Nessa. Absolutely cool. That’s a mask of your face.

Yeah, it’s an exact replica of how I look like. Rajat Maheshwari is part of the Cyber &  Intelligence Solutions team at Mastercard. In his role, his team develops tools to  manage customer risk and prevent scams.

Part of his role involves getting in the mind of  potential fraudsters by creating fake identities. These are synthetic IDs. As good as a genuine ID.

You can have these identities, you can have  the face mask, you can have the fingerprint, and you are essentially replicating someone else. That person can do anything with these. This was done by a Japanese artist who did  the masks for James Cameron’s movie Avatar.

Sometimes we have to think like bad actors to come  up with the solutions which can stop these things. The world has evolved from these masks,  and now the deepfakes are coming in. The intent was not to break the technology, but  the intent was to help the solution providers to enhance the level so that they can  stand against these attacks as well.

Wow, what do we have here? Please have a seat. Over at Amazon Web Services, Joel Garcia and his team have made a game out of simulating  real-world security conditions as well, which they hope will help clients improve  their incident response processes.

What we have here is a project called Chaos Kitty. Why Kitty? Well, I have cats at home;  always destroying my furniture.

Oh yeah, agents of chaos for sure. What exactly is chaos engineering? We’re going to intentionally inject  some failure so that we can learn.

We can see that there’s many  colorful lights and all these bricks that we use to represent  what we have in our AWS cloud. When it’s red, something is wrong  with the security configuration and when it’s green, it’s all good or compliant. We added on a Gen (Generative) AI assistant.

What we have here is a typical  company security policy. Typically, without a Gen AI assistant, they will have to look through and  study this policy document very deeply. We have fed this document in, along with the Gen AI assistant, so it knows all this information.

I could go in there and ask questions. It's going to give me some best practices. Alexa, fix Chaos Kitty.

Remediating Chaos Kitty environment. You’d be able to leverage AI and Gen AI  to actually help fix the challenges and then they’d be more focused on the  other areas that could be improved. To be able to converse  naturally with an AI chatbot, a large language model, or LLM, is needed.

LLMs are AI models, pre-trained on vast amounts of data, which can understand and  generate human language responses. Popular models include ChatGPT-4 and Gemini. The generative AI market is expected to  grow over $1 trillion in the next decade.

A team in Singapore is developing a large language model that’s catering  to Southeast Asian languages. Leong Weiqi speaks 14 languages fluently,  and he’s an AI engineer and linguist involved in developing the SEA-LION (South-east  Asian Languages in One Network) model. We’re going to look at an  example of informal Indonesian.

We are essentially asking the model: our friend,  because of his work, he’s just panicking, and he’s working really hard every day;  how can we help him best manage his work? Yeah, but in that context, they’re  saying, they’re using an idiom, right? Exactly.

They’re saying that his beard is always on fire. That idiom might not be so  understandable to certain models, right. So let's see how they deal with this.

For this demo, we have four panels, and each  of them corresponds to one language model. So, on the left, we have SEA-LION, our model,  and we have three other models on the right. Yeah, we can see that it  didn’t understand the phrase.

It took it literally. Exactly, like it’s thinking  about setting someone on fire. It's also in the same tone.

Exactly. It’s casual, informal, still colloquial. Whereas this one is still sticking  to a more formal kind of response.

That’s really what we want to achieve as well. Now we are seeing that some  of these models out there, they are not able to handle  multicultural contexts. That’s understandable because they’re building  these models for a particular audience.

For us in Southeast Asia, we need  to operate within this region, handling our languages and cultures. So, this is why we decided to build SEA-LION. Vincent Oh works as a Senior Specialist  Solutions Architect at AWS, and his projects involve leveraging generative AI based on human  prompts to create personalized experiences.

StoryGen was a project that we did with the  National Library Board of the entire Singapore. We wanted to reinvent the future of libraries. We used generative AI and AWS technology  to create a new experience whereby young children and adults can put in a series of inputs, their own selection, and a brand-new book  will be created on the spot for them.

What you call the prompts that you’re actually sending back to the large language model. This is just the very beginning of the  art of the possible and it’s going to be amazing how people will leverage Gen AI to  unleash their extended level of creativity. Part of those new jobs that are  created as part of AI is a role or skillset which is actually prompt  engineering, which didn’t exist before.

And when you look at prompt engineering,  you don’t need to be technical, you just need to understand how to  put those prompts, as you would, for example for a detailed search, to maximize  and leverage the power of a large language model. There is a growing demand for AI  specialists, and here at AI Singapore, the rest of Weiqi’s team, hailing from all  over Southeast Asia, including Vietnam, Philippines and Thailand, are working  on making SEA-LION more inclusive. You are the one who brought  everyone into this team.

More or less, more or less. My dream is that eventually we have everybody from all the countries in Southeast  Asia being represented here. AI Singapore's apprenticeship program  aims to grow the pool locally, even those considering mid-career switches.

In fact, some of the SEA-LION team graduated  as AI apprentices after switching disciplines, such as Tai who studied finance and  Weiqi who was previously a pharmacist. We’re not really alone trying to  tackle low-resource languages. Ultimately, all the stuff that we  do is fully open source as well and is really shared with the public in  general so that everyone can benefit.

Everyone is working on different  aspects of the large language model. We get to learn about it with each other. It does feel a bit like a mini-ASEAN.

There's so much to do in this field. And AI is one of those industries where  you can make a mid-career switch. Absolutely, absolutely.

And I’m a good example, Nessa. Rajat started his career in the  mobile and semiconductor industry and made the switch over to AI in 2014. I have seen that journey  in the past 20 years or so, where AI has really progressed and it’s  becoming an integral part of our lives.

When you ask people to use a digital  ecosystem, there are side effects. According to the Global Anti-Scam Alliance, more than $1 trillion are lost to scams  every year, affecting 2 billion victims. Rajat’s work at Mastercard includes  working with banks and governments to apply intelligent AI systems to  predict whether scams are taking place.

Let’s take the example that  you are giving me $100, Nessa. That looks like a perfectly  legitimate transaction. But if 1,000 more people are  giving me $100 in the same day.

. . That will sound alarm bells.

Exactly. We train the models to detect the  behavior, to detect the patterns, and then do the risk assessment and  save you sending the money to me. Welcome to the MasterCard experience.

Whenever a person taps the  card, our AI models kick in. Then we give a risk assessment  to the issuing institution, what do we feel about this transaction. You can see the screen here showing that  the transactions are getting declined.

Where do you see the state of  AI in the next couple of years? The integration with the LLMs, that's  essentially the large language models. Large language models form a big part of the user  experience solutions that Joel works on for AWS.

Take reading invoices for example. If you want to introduce something else into  your business, a new form, a new process, then you’ll have to reevaluate the  technology or re-factor that in. With Gen AI, because of the large language  model, it can actually parse these.

Parsing refers to the process of  breaking down a user’s input into smaller pieces and analyzing each  piece to determine its meaning. I’d like to know, if there’s a proper signatory. It even read my handwriting, so it knows from my  terrible handwriting that it’s Michael Garcia.

Did you see yourself ending  up where you are today? Even at a young age, I knew that in some form or way that I was going  to be involved with technology. What did you start out studying?

I graduated computer science. I went through a variety of jobs. I even founded a company before  startups were fashionable.

Today, with AI, when you’re thinking  about redefining experiences, you’re not going to get to it on the first [try]. This is a super longshot ambition. You know that Southeast Asia is not 10 languages.

There are like hundreds of dialects:  Balinese, Javanese, Bisayan. So, we hope to get those  represented as well, eventually. How soon can we get there?

Our hope is that it will trigger  similar movements across the region. Outside of Southeast Asia, the  interest is also very strong. We have similar situation in India, in Africa,  where there’s a bit of underrepresentation.

If I am able to save someone's life-long  savings, then that makes my day. Is this what people mean when they  say that AI is going to take our jobs? It’s going to make something more efficient  so that people can actually concentrate on what matters more, either for  the business or for customers.