Generative AI is the Answer: What Was the Question?

Tom Godden (00:10):
So we often say generative AI is the answer. What was the question? Generative AI is really an exciting new technology. It's a transformative technology. In fact, we believe it's probably on par with transformative of technologies like the printing press, electricity, personal computing, internet, and even the cloud.

Phil Le-Brun:
You say it's new; in some ways it’s the culmination of 60 years worth of development, what with the cloud enabling low-cost compute, and then advances in things like statistical techniques, the transformer model now being used for generative AI, all of this has become possible. So there's been this buildup of new breakthroughs, which has brought us to this point now where we can do some absolute incredible things in organizations.

Miriam McLemore:
The thing that I would add is data, right? We've all been worried about data. We're buried in data. And generative AI gives you a path out to actually use that data, use it productively for some things you just described, whether it's improved chat bots, call center, content creation. There's a new announcement we just made on HealthScribe--you've sat in the doctor's office and watched the doctor type on the computer. Let’s get the doctor back in the room with the patient. So there are some great new capabilities, but it is about harnessing data.

Tom Godden:
And we get all excited with generative AI about all of the foundational models, the large language models. People want to run to the end and start using generative AI. But it's your data that is going to make the difference. If you don't have a strong data foundation, you are really going to struggle to be able to do anything beyond well clever parlor tricks with generative AI. You really as an enterprise, as a business, need to get to that predictability, that contextual information, and your data is going to be the difference maker in that.

Miriam McLemore:
What I love about that is data has been regulated to the back office as a heavy lifting task, and now the executive leadership team needs to lean in to that data strategy.

Phil Le-Brun:
And we're at this point now where we've been on this journey as a company to democratize machine learning and artificial intelligence, but often that's for the folks who sort of get the technology side. This now levels that playing field. If you look at some of the McKinsey data, they're suggesting 75% of the benefit from generative AI is going to come from four areas: customer operations (things like call centers), sales and marketing, research and development, and then software development. You talked about chatbots, for instance. This ability to have a conversation with a brand, me as a customer to have a conversation that gets me what I need, but also from the company's point of view, the money they'll save and the friction they can take out the customer journey: that's going to become a competitive advantage.

Miriam McLemore (03:20):
I was just with one of our customers and they are working really hard to escape the velocity of their day-to-day business, of what is keeping them busy, but not looking ahead, not thinking big. How do we as leaders break that model? You're looking outside-in, you're anticipating change. We have to be different leaders, and that does address the bureaucracy that is holding you back.

Tom Godden:
Well, this is a transformative change just like cloud computing, just like digital transformation. And I think a lot of our old practices serve us well here. You got to get the culture right. You got to get the organization right. You got to look hard at the processes or mechanisms, as we like to call them, and get those in the right place. Otherwise, you're going to buy that super fast race car but not have the pit crew and the driver trained and ready and able to use it. And it's not going to go anywhere. You'll get some benefit probably, but we're talking about being transformative because that’s the concern. The competition is not sitting still, right? Your competitors are not sitting still. And there is a first mover advantage on a lot of this. So be transformative or even go so far as being disruptive within your industry, and it's going to take all those pieces.

Phil Le-Brun:
Scott Galloway, a professor, talks about how you shouldn't be worried about the impact of AI on your company. You should be worried about the impact of those who understand how to use AI to transform their company. And Miriam, you have the saying “think big, start small, scale quickly.” And I think that's what we need to do. You can't fall in love with the hype, but small thinking is a self-fulfilling prophecy. So how do you really think big about how you can transform your company, but start now, overcome your inertia, and learn how you can use this. Get your hands on the fastest way of learning what's going to work in your organization.

Tom Godden (05:27):
It comes back to generative AI is the answer. What's the question? Only do it because it's adding value. Use it because it's the right tool. Oftentimes we see “just” data analytics or, what is laughable, “regular” artificial intelligence, “regular” machine learning, which seemed so advanced six months ago. Sometimes they're better purpose-fit for the problem that's needing to be solved. And you don't need all of the lifting and effort with generative AI to get those things done. Be driven by value. Don't just do it because all the cool kids are doing it.

Miriam McLemore:
But lean in.

Tom Godden:
Well, get going. Be impatient. There's a first mover advantage. This is transformative. Again, this is on par with some of the other large disruptions we've seen, and you got to get moving. You got to have a play on it.

Phil Le-Brun:
There's nothing stopping you from getting going right now. You can use large language models right now through AWS SageMaker JumpStart. All of this stuff is right there. It’s a pay-as-you go model, switch on, try it out. If it doesn't work, switch it off.

Miriam McLemore:
You can build a data strategy right now. You should have done it yesterday.

Tom Godden:
It'll be good for you no matter what, whether you do generative AI or not. And get your cloud house in order. If you haven't already been robust and mature, get a Cloud Center of Excellence and the team set up to be able to go after those things. Do it because it's just a good idea to do it, but do it because it's also going to be required. It's table stakes, foundational elements for generative AI.

Phil Le-Brun:
And don't just leave this to the CIO. No disrespect to the CIO, but this is a business challenge. I mean, us as business executives have to understand finance, have to understand people management. I think there's a need now for business executives to dip their toe into the technology and data water and figure out what do they need to know. To your point, Tom, this isn't magic. This is going to be a combination of the technology, business process change, and people change. How do you bring that together? You can't leave that just to the IT team.

Tom Godden (07:26):
And as part of that, we’ve got to look holistically at responsible AI. You should have a responsible AI program in place if you're already doing AI and ML. But generative AI, because of its generative nature, is going to force you to really double down on that and to understand how you're mitigating bias from things, how you're preventing hallucinations from occurring with inside your systems, even toxic results, unless you have the right structures in place. We talked a lot about this, the Galloway quote, it's people understanding how to use generative AI, but it's that human oversight that's going to be so imperative. We're not quite at a place where we're prepared to completely remove the human from the equation. Now maybe you do in production, but you've tested and tested and tested with human oversight on how it's operating, and it's a living, breathing thing. So unlike lots of software that we used to write, you test it, it works, put it in production, you go, “great, we're going to move on. We'll come back and test it later. Maybe again, someday.” No, this is going to be something where you're going to want to run use cases and tests on a regular basis. I mean daily, maybe even hourly in some things to make sure that you're in control.

Miriam McLemore:
I love your point, and Phil, you say this often, you've got to be careful with your addiction to prediction because you can use this technology to say and reconfirm things that you believe. You have got to be careful and listen to the data, leverage this to open new pathways and consider new approaches.

Phil Le-Brun:
I think it comes back to a lot of what we talk about with data: over 75% of the issues we see with data are people, organization, and culture. And it starts with leadership. As a leader, are you role modeling the behavior? Are you questioning the data? Are you asking the right questions? Are you just trying to confirm the decision you already made? But I think this is such an interesting, exciting time if you're a business executive now, the potential you have, but also the obligation you have around responsible AI. And it's not even just the ethics and bias and the such, it's the implications of what you're doing. I think we're doing the right thing with Amazon Bedrock, which is bring the model into your environment in a secure environment. Use your data to train that model rather than taking your data and putting it into a public model. But also, we know that there's no single foundation model that's going to solve all your problems.

Tom Godden:
And that's where Bedrock's real strength is going to come in. The ability to access these language models through an API, and the ability to then pivot and move if warranted, if needed, or to be able to access a new one for a new use case, but to be able to do it so rapidly and quickly, just like all of the other AWS services. Spin it up when you need it, spin it down when you don't.

Phil Le-Brun (10:28):
And if you want to build your own model, well, don't jump into that. Bless your heart. Bless your heart, bless. But don't jump into that. I mean, you could be spending 10, a hundred million dollars, but if there's a real business case to do that, firstly, learn about what you really need to do with the technology you have. But then you've got the infrastructure in the cloud. You've got things like AWS’s Trainium and Inferentia to drive the cost of inferences and training down. So almost regardless of where you're going in the future, you want your data strategy set and you want to be in the cloud. You don't want to try this at home.

Tom Godden:
Let's talk for a minute about the vanity metrics around the foundational models. The latest stats are, the biggest foundation models have over 500 billion parameters. Sounds really cool. Great. I would like five. Why can't I have 800 billion? But I think what we're also seeing is sometimes you don't need that much size. In fact, it can create more spurious results and answers. Having a purpose-built--even a public, open source one—but one that's purpose-built for the use case that you're trying to do, that's tuned with your contextual business information, most likely has better results and efficacy than these vanity metrics that are amazing to talk about. 500 billion parameters sounds absolutely amazing, but may not be what's needed to solve the problem.

Miriam McLemore:
Yeah, the right data for the right problem. And again, as you said, start with the problem. Work backwards from a business value that you can drive, lean in, pick a place to begin. It's an exciting time, but it's going to take a minute to figure out your rhythm and what adds value in your approach. I am amazed at the customers that are already leaning in, making some incredible pathways that we're all going to copy. And that's one of the great things I think, and at AWS sessions, is leaning into other customer use cases, learning from those that have tried things. You don't have to do it all yourself.

Tom Godden:
We talked about Code Whisperer. I see a lot of people also very interested in Contact Center. It's a target rich environment. You have a direct relationship with the customer, so you got to be careful, but it's also lower risk from maybe trying to come up with that next new therapeutic in the healthcare industry. Please do it, we need that kind of advancement. But now we're getting really high risk; really, really complicated. I'm also seeing some people look at their intranet and if yours was like mine, it's where information went to die. You had a great search engine on top of it that found next to nothing in it, and now you have a great opportunity to take and unlock all that information within your organization, but also a great way to start to bring this to life so people can see its potential and do it in a low risk kind of way. That adds a lot of value. Go do it. Be impatient.

Miriam McLemore (13:35):
That's one of the big announcements that we've made is around training, right? Because how do you learn how to do this? And so getting out there, getting your team trained, getting your executive team trained. We have a number of offerings that can help our customers train their organizations on places to start, what the tools are available so you can make your own decision for the right approach for your company.

Phil Le-Brun:
Learn and be curious. I mean, we've got the Executive course from Training and Certification. It’s a real straightforward, what is generative AI? We've got the Coursera course now, which is fantastic. If you really want to get into the nuts and bolts and some of the things you were talking about, about that balance between amount of data and parameters and compute and finding the right balance. So it's all out there. A lot of this is public domain. Do it now. Start learning now. It's never too early.

Tom Godden:
And that training's going to help you bring people along because, let's be honest, this is a transformative technology, but it can also be disruptive. Some people rightfully are very concerned about, what does this mean? Not only for my job. I've got rent, a mortgage, kids to send to college. Do I still have a role in this new, massively exciting, transformed world? What's it going to do to society? And I think helping them see their role, helping them understand what role that they're going to be able to play and support them through that training, is going to become even more vital in this than in other transformative evolutions we've seen.

Phil Le-Brun:
Yeah, break your silos down. I mean, you talked about bias, Tom. The best way of mitigating against bias is to have a team that is representative of your customer base. Plus also, we know machine learning in general, generative AI absolutely, is going to cut across the organization. It's going to work despite your organizational structure, not because of it. So get rid of your bureaucracy. I guess you can use generative AI to get rid of some of it, but it's back to what makes-

Tom Godden:
Build me a new org chart?

Phil Le-Brun:
Yeah, automate PowerPoint.

Miriam McLemore:
Tell me who should be in charge.

Phil Le-Brun:
That’s where some bias is going to creep in. That's right. But use it to really understand your competitive advantage. You look at companies like Autodesk who are using generative AI, now they're reducing the weight of some of their designs back to 40%. What a great sustainability benefit. But they've really identified, “where can we use it to make a competitive difference to our organization?”

Tom Godden:
Do it because it adds value. Don't just do it because the cool kids are.

Phil Le-Brun (16:11):
What I find interesting is often there's this tension between, “do I save money because times are tough” or “do I innovate?” and I don't think there's a choice anymore. You need to do both. And the reality is there's so much money wasted in organization. So I think 94% of CXOs in one study showed that their own organizational structure is preventing them innovating. All of that bureaucracy. How long does it take you to make a decision? What we sort of tongue-in-cheek call the “bureaucratic mass index.” How much time are you actually spending doing meaningful work versus waiting for a decision? How do you drive those decisions down? So I don't think it's “do you innovate or do you save money?” I think you do both. Drive out cost of undifferentiated work, free up that to innovate, and it becomes a virtuous cycle. And even use machine learning, generative AI, to actually drive out some of that cost and bureaucracy in your own organization.

Miriam McLemore:
What we have seen and say to our customers is, constraints actually drive innovation better than when we have everything at our fingertips. Getting between a rock and a hard place makes you get creative about “how do I get out of this spot?” You can leverage tough economic times to think differently. You don't have a choice. But I also think, as you said, with generative AI, one of the great values is going to be productivity and saving of some of that undifferentiated lifting. I was at the Coca-Cola company for many years, and generating content, generating new sites, new experiences, new images for our consumers and customers, point of sale material, it will be a game changer for martech.

Tom Godden:
We've seen this play out in other transformations. The real change isn't always just the technology, but it's your willingness to apply the technology in a new way. We saw that with electricity, changing how we laid out factories and operated factories. We were able to run factories safer 24 hours a day. So again, the technology was the enabler, that initial enabler, but the real transformation occurred when we rethought the process. So as we look at this and we look for this new balance, we got to go back and look at our processes and go, why am I doing this? Does generative AI allow me to think of doing this in a completely different way? Don't just automate your past with generative AI. Use it as an opportunity to rethink these things and do them completely differently.