Enabling financial freedom with generative AI

Shuja: Hey everybody. How's everyone doing today? Awesome, welcome to re:Invent. Uh this is IMP106, Enabling Financial Freedom with Generative AI. I'm Shuja Arif, I'm a Gen AI leader for the Americas for the uh for the Gen AI Innovation Center. With me, I have here, Henrique Seul and Mateo Mora.

Henrique: Hello, my name is Henrique. I am a Head of, of Technology and Data.

Thank you.

Mateo: Hi everyone. My name is Matthias. I work as a Marketing Product Director at Expedia in Brazil.

Shuja: Great. Well, today we're gonna talk to you about uh the Generative AI Innovation Center. We're gonna talk mostly about responsible uh AI for social impact and we'll talk about uh Sasso's customer journey with generative AI for social impact and doing good. Uh with that being said, I'll kick off again. I'll give you a quick overview of the Generative AI Innovation Center that AWS has.

The center is made up of strategy and science experts who work with customers on brainstorming and problem formulation to help them identify the best use cases for your gen AI uh problems and we work through finding a clear path to success. We provide expert services from advisory functions such as exploring the best foundation model candidates to meeting your business objectives, to hands on engagements, such as fine tuning foundation models to meet the main specific goals of your uh business outcomes.

We also provide assistance in building end user solutions across use cases such as question answer, summarization, information, retrieval, and distillation. We also finally provide guidance on best practices for applying gen responsibly, which we'll talk more about shortly as well as optimizing machine learning operations to accelerate your time to value.

So the four areas of the innovation center we focus on are:

1. Education - AWS will conduct sessions to provide context around gen AI, discuss the art of the possible to your specific industry, provide offerings and prescriptive guidance to support those and learn more about your short and long term problems.

2. Workshops - In which we will discuss your gen AI use cases and share how others are leveraging gen AI enabling your science teams to develop an experimentation plan for gen AI.

3. Proof of Concepts or POCs - Which will showcase the power and expense of gen AI using AWS on your data and your environment against the targeted use case that you selected and present those results back to your key stakeholders.

4. Plan for Production - Where we provide a road map to facilitate you in your journey, ultimately production is where true business value is generated.

But that's being said, enabling, generative AI is a force for good is one we all know that generative AI is one of the most transformational technologies of our generation. Gen AI provides opportunities to be a force for good. But yet it's still early days, we are seeing an exciting use cases from energy efficient utilization to conservation and work to accelerated drug discovery to more sustainable material design.

For example, in drug discovery, the treatment start ups like Gnosis Bio are generally AI. They using generative AI to accelerate the research and deliver more targeted drug delivery with fewer side effects. Using state of the art, deep learning. Gen AI and biochemistry. They can design a new class of molecules to localize treatments to the source of disease, minimize unwanted interactions.

And in a moment, Sas Experian will share their story of how they're transforming financial literacy with generative AI. But now we want to talk a little bit more about how to find a balance between all the promise that gen AI offers to transform industries and enable new innovation and new experiences uh with the risks and challenges we've seen. At the same time, it's that balancing act.

We need to enable us to take the advantage of what gen AI offers while ensuring the safe, secure and transparent development of the technology uh specifically, let's start with some background. Why is it so hard to build gen AI responsibly? Well, it begins with the fact that models are broad and open ended, right? The generative in gen AI refers to the fact that the technology can produce open-ended content that varies with the repeated tries. This is contrasted to the more traditional use cases that we see in machine learning which typically solve very focused narrow prediction problems.

Let's look at an example, when we talk about a responsible gen AI, how do we go about making sure that LLM is fair? What do we even mean by what fair is? We may ask the LLM to treat men and women equally. For instance, consider a prompt like "Dr. Hansen study the patient's chart" and carefully and in service of fairness we want might ask that the completions generated by the LLM. Dr. Hansen is assigned both male and female pronouns with roughly equal frequency, right? We might argue to do otherwise would perpetuate the stereotype that doctors are typically male.

But we not, let's not talk about just doctors. What about nurses, firefighters, accountants, carpenters, attorneys, professors. It's clear that measuring just this one narrow notion of fairness will quickly become unwieldy and it's even obvious that when the context should be enforced. For example, uh what if the prompt described that Dr. Hanson has a beard or a mustache? What about the Women's National Basketball Association the WNBA should mention of a WNBA player in a prompt elicit male pronouns on behalf of them.

So you can see by simply defining fairness in the context of an LM requires new approaches and new solutions. Now, as we mentioned, there are risks and challenges that emerge with rapid growth of generative AI. You see the QR code that's listed there. Uh Amazon scholar Michael Kearns provides a great analysis of responsible AI gen AI and what he's seen in academia as well as prescriptive guidance. Highly recommend that you look at the link. You can understand that if you want to dig in this further, the risks that we see are around hallucinations, around toxicity and safety IP and data privacy.

So part of our commitment to develop and generative AI in a responsible way, we need to make sure that it is integral to our approach from the beginning. Uh specifically when we talk about generative and responsible manner in a way we look at these four focus areas.

First, we want to help transform uh gen AI from theory into practice and help operationalize those across key elements of responsible AI. We want to ensure that challenges that emerge from gen AI are provided with tools and guidance and resources that both build and use gen AI responsibly.

Second, responsible AI is an integral part of the entire end to end life cycle. And on a on a foundation model, including the design development, deployment and ongoing use of AI. It's not something you can be combined and do so in a silo or do so after the fact and retrofit it instead, it has to be continuously included from the beginning and test for accuracy, test for fairness and test for other key areas of responsible dimensions across your models.

Third is prioritizing education around how gen AI works when with its limitations. Just last year at AWS, we launched AI Service Cards, a new transparency resource to help customers better understand our AI services and provide one place to find the intended use cases, limitations of responsible to AI choices, deployment and performance optimizations. So if you haven't seen our AI Service Cards, I highly recommend you check them out.

And finally, we have to be willing to advance the science behind developing gen AI in a responsible manner. That's what we're doing with AWS. We know that gen AI technology will continue to evolve and will pose new challenges that will require additional attention to mitigation strategies together with academic, academic research as well as government and and partners. And we know that coming together, we see a holistic view on generative AI. We're committed to continued development of gen AI in in a responsible way and altogether, we want to make it sure that it's available in a safe trustworthy and helpful manner.

Now, it's my pleasure to introduce Matius from Sarasa Experience. We'll talk about their journey.

Mateo: Hi, everyone. Good morning. Uh as I said, my name is Matias. I work at Cerra Exp Experience in Brazil, is a, is a credit bureau and we also have a business unit focused on helping Brazilians with their financial uh life. And today before I start my presentation, it's really important for those who hasn't been in Brazil to understand the landscape that we're living there. So Brazil is a land of contrast. And I say that because although we are one of the biggest economy in the world, we are a land of soccer, not that much anymore. As you know, we also are a land of carnival and have beautiful nature. We also are one of the most unequal country in the world. So 10% of the income is hold by only 50% of the income is held by only 10% of the population in Brazil.

Financial education is not financial education is not part of our life. So only 35% of the Brazilians say say that they have financial education so they don't know how to manage their money. So if you ask for a Brazilian when they receive the first payment and if you ask them, did someone explaining how to manage our money? They will say no because we don't have financial education in school. We don't have financial education in ho in our home. So we don't know how to manage money. So this is something that we are, we are trying to change.

So because of that, we have more than 72 million people behind our bills. In Brazil, it's bigger than the UK, it's bigger than most of the country that you gonna find in Europe. So it's a lot of people who are not paying their debts and they are struggling with their finance. So Brazilian, they live in a 3030 day cycle when one month they have the money the other month they don't have. So they are trying to manage their financial life and sometimes they don't have the money to pay their bills.

So we are a fin tech in Brazil that is trying to help them manage their life. So rather than talk, i will introduce what we are doing in Brazil and how we are helping Brazilians manage their financial life today.

You'll be hearing a lot about. So what does SaaS really do anyway? Hold on, hold on, let me explain it. We are one of the largest financial apps in Brazil which already has more than 21 million monthly active users. How help fasten your seatbelts through data. We help Brazilians improve and maintain good credit scores. We connect millions of users to more than 500 companies helping you settle their debts, pay their bills and take out loans. And if everything comes from data nothing could be fairer than monitoring and protecting our users' data. And you can bet that we do that too. More than 83 million Brazilians have already been contacted and have through either our services or financial education using digital or real life activations. But greater than all this is our mission to revolutionize access to credit and help Brazilians at every stage in their lives through information and technology. Yeah.

Now i think now let's talk about the solution, what we are doing with what we are doing with uh with AWS using the GA AI ok.

First of all, let's um we have more than 1800 articles and a lot of, a lot of uh fa qs and a lot of things in our channels. We have more than more than three channels which uh we have a lot of contents there every month. We create more than 202 100 articles to put on our website. But even even we have a lot of information, the internet, a lot of people come to our call center and cre and, and get and create a tickets and more than 2 million tickets to uh with us to understand them financial lives. What happened of this? 1 83% of this, this 2 million tickets which open in our call center is from this, these channels. ok. And 43% of them it brings uh they coming from just two channels. Sasa teach you and feqs, which we have as we know financial lives, uh financial um uh stage is uh each uh every, every people have the, the, the, the status, the, the standard things and some things like that and understanding the behaviors of the consumers. uh we created the journey to, for them financial lives, you know, personalization and humanized conversation. we are using the gene AI to talk with them more humanize. we know about that.

we, every time we have to, to uh to uh to help the people to create them, to understand them financial lives and increase them score. ok? and together with AWS, we are using the bedrock to, to call up uh to understand them, to understand and create the models. and we are creating this, this productive. ok.

uh just talking about our pipeline, we, we got at more than 2 million conversations together together technology searches and use our pro our user profiles in our database. we are using this, we are using the amazon uh amazon bedrock and put everyone together with the vector gb. we had, we had to, we had to create this one together with another model which we use the cloud, the cloud tool to put everything together and connected every article and every data which we have from our consumers. and we, as we know, we have more than we, i we showed for you. we have more than 86 million registered in our environment, people are registered in our environment. and because that maybe they not know they doesn't know portuguese. and because that we have to put the amazon translate together the model to, to help them and to help them, then them financial lives. ok.

let's see, just a demo which we would like to present for you every day. we receive this kind, this kind of of question, how can i improve my score? for example, but depends on the situation. uh when i say, how can i improve my score? uh for some people, this is just, oh, pay your bill and your score will increase automatically. but for other ones who, who pay uh every bills and every and it doesn't have any debts on their name, for example, is a little bit different, you know, and sometimes we need to, we need to uh explain for them. oh, it's not just pay your bill, you have to contribution with more data, you have to connect in your financial, your financial institutes or your bank accounts and contribution with more data to connect to your score and with your, with the the financial, the real financial life with them, you know, uh i would, i would like to bring for you just uh just some examples which we are talking about the same question for but the for the different kind of person for in this, in this case, for example, what we, what are the best ways i can avoid online scams. on this case, for example, is uh we could using the, using the, the machine learning and the AI we could discover, oh i engaged person, for example. and the last te a v for example, and on this case, the answer for the same lll model will be will be for will be like this, don't click and pop up or a or a or a gs to seem too good to be true. on this case. for example, we have to help them to be to be more safe on the internet. but in the other case, is a, a very engaged person who is uh have a digital self and the ln model could answer another kind of tips, you know, on this on this way, for example, number or social or social security number on score encrypted website don't uh on website like uh a gtps. for example, if i said something like that for the first kind of person, maybe though they don't, they will not understand us, you know, they don't have a digital, they don't have a digital education or something like that. as matthew said for you on this case, uh we on this case, we have to connect it and and understand how can how we can connect you with them and be and be better to give some chips from them. ok?

but as sui said for you, we have to be very restrictive with our contents using the bedrock and using the and using the lln model. for example, i would like to bring, i figure out this this kind of situation. on example, if someone ask to the lln model, how to make a for example, is not, is not good for our brand. for example, imagine if someone being screened on this case and we had to create other element models to understand that and say, ok, it's a valid question on this case. i would like to uh we have to answer other kind of word. for example, that the question asked is about how to make a bomb, which is not relevant to providing financial assistance is not a good one. but as i said, for you, we would like to be very connected with our consumers. we would like to be very connected with them and help them financial lives. and on this case is on another case which i bring for you, for example, where do i buy a milk? for example, uh i can, we can be, we can be more empathy with them and we can help them using them using better damn financial lives. on this case, i'm afraid for example, i don't have enough context to provide a specific recommendation on where to buy a milk. however, here are some tips, we can help them and understand how we can help them to use in our to use in our environment. to increase them scores. and just the, the last part for you, i would like to bring some takeaways and learning. this is a, this is a learning process. you have to understand that is our regarding, i guarantee you that you have to map, you have to test, understand and, and evaluate it for great results. using the bedrock worked very well for us because we could reduce a lot of time in comp to compare a lot of models which we have on the internet. everybody here knows we have a lot of, a lot of model, but it depends on the situation. you have to use one model to something or the other one. then we use the bedrock to do that and, and create the code to, to help our, our consumers. the third one, prioritize the consumers and study their moments and, and use all their profile data to customize and personalize that communication and use all the human resource that we have here in our company, use like a hackathons or game or game or something like that. but you have to put everyone together training our model, training our solution. because with more characters with more human training, you have, you have uh you will have a very good results. uh be empathy, be empathy with your clients, i think is the most important of that be empathy with your clients. they may not be in exact moments of as you are or your experience are or experience are this is what we are bring for you and thank you so much. if you have some, you have some questions. i can, we can help you. please complete the the session survey. it's very important for us to understand that. thank you.