#2: Deep Dive - Why AI is reshaping the competitive landscape

AI Revolution: Strategic Insights for Business Leaders

In this series of 'AI Demystified for Executives,' host Andrew Psaltis provides an in-depth exploration of AI's strategic foundations and transformative capabilities for business leaders. We chart AI's evolution from the 1950s to the present, emphasizing milestones like machine learning, big data, and large language models such as GPT. Key episodes delve into the versatility of AI technologies, including GPT, DALL-E, and Whisper, as adaptable tools for various business applications, while highlighting real-world industry examples like JP Morgan's COIN and Lemonade Insurance. Discussions cover AI's role in customer service, document processing, and managing ethical concerns such as hallucinations and data privacy. Practical strategies for AI adoption include the importance of education, starting small with pilot projects, and developing a comprehensive AI roadmap. The series underscores AI's potential to democratize industry competition and revolutionize business practices.

00:00 #2: Why AI is reshaping the competitive landscape
01:57 Historical Perspective
10:13 GPT and LLM
18:26 Examples
28:52 Challenges and Considerations
36:46 Conclusion

Transcript

    Dragonfly Risings school of data proudly presents the AI demystified for executives podcast. This is the podcast for executives who want to learn how to apply aI to their business. I'm your host. Andrew Psaltis 

  I've been in the data and AI space for over 25 years. All of this technology. Gets me excited. But I find even more enjoyment in teaching others and helping them understand. How to apply it to their business.

As a reminder, this podcast is split into two parts. Seven minute. Monday's perfect for catching up on the week's most important aha moments. During your commute or while sipping your morning coffee or tea. And then 30 to 40 minute deep dives into the week's topics on Wednesday. That may be a bit much if you're driving. 

Be sure to subscribe to this podcast on your favorite platform, where you listen to others. You can also subscribe to my newsletter. 

So never missed an update. The address is in the show notes. Without further ado, let's dive into this episode.

 This month's theme is strategic foundations of AI for business leaders. And this week's topic is why is AI reshaping? The competitive landscape. This is our Wednesday deep dive session. Recorded live from my home studio on October 16th. 2024. Historical context of AI and business as something we really need to talk about to set the stage for how to think about why it's reshaping the competitive landscape. 

Historical Perspective

And this will also help provide a foundation. As we talk about more advanced topics in AI. So a brief history of AI in business, as we get going here. So in the 1950s to sixties, There was early AI research and lots of optimism as to what it was going to do. There was a Dartmouth conference in 1956, where the actual term artificial intelligence was coined. It was debated of whether they call it. Like artificial computation. I wonder how the industry would have shaped up if it was called that instead of intelligence. It's interesting, just to think about. In the seventies and the eighties. We had experts systems in business. A great example of this was the mycin system for medical diagnosis. This used artificial intelligence to identify bacteria. That was causing severe infections. Such as meningitis. And it was able to recommend antibiotics with the dosage adjusted for the patient's body weight. 

The systems then, though we're really rigid. 

Very rigid rules, difficult to maintain. It really was the engineers that were building it. In the nineties and then the two thousands. We entered the era of kind of data mining and business intelligence. And we had this shift happening towards a data-driven decision making business. A great example is Walmart's use of data mining for inventory management. They often only have a couple of items deep on a shelf. And that's intentional they're able to use a data mining and all of the data they have to have almost that real-time inventory management. So they could manage their inventory really well, based on the use of data mining. In comparison. The traditional grocery stores. They may have 10, 12 or more items deep. On the shelf and they're carrying all that inventory. 

So that data-driven, decision-making, Walmart's a great example. When you think about that, of using data to drive business, In the 2010s. This is where machine learning really started to become mainstream.

This was when the big data systems started to arrive on the scene. And with those systems to manage. Big data, we had improved algorithms. That drove adoption of machine learning.

A great example of this is Netflix's recommendation system. 

The 2020s where we are now, we've  entered into this era of the large language models and generative AI. 

This has had tremendous impact on various industries.  We're going to talk later in this episode about  use cases and how to really think about it in your business and apply it. So to look at some of these key milestones across this timeline.

 2012. We had the deep learning breakthrough and image recognition.  There was an image net competition  being able to identify images. In 2016 alpha go. Beats the world champion at go. 

And then in 2018 we had BERT, which is the precursor to GPT.  This is a large language model introduced by Google. In late 2018.  This really improved natural language processing. Then in late 2022. Chat GPT launched. And that  marks a shift. To consumer facing AI. Consumer facing, meaning that you as a business person, Are able to interact with. This technology. Directly. If you think about the difference between your  experience using chat GPT versus your experience with  Netflix  or Amazon? 

 You're interacting with a recommendation application. You're interacting with Amazon storefront. You're interacting with Netflix. In the case of chat GPT, you're now interacting with the underlying technology. Much different. 

You could do things now. That you couldn't before. We all can. And this is a  transition from. Narrow AI to  general AI . Narrow AI  meaning that it's designed for very specific task. So everyone remembers having to have spam filtering. We take it for granted now.  Wasn't that long ago that you were looking for spam filtering software for email. Very specific task. On the other side of the spectrum,  general AI. Flexible systems. Capable of handling diverse tasks.  That may be. 

Crafting an email for you. It may be generating an image for you. It could be looking at a financial document. All within the same system.  It's a really able to handle everything. 

The implications for this for business really is we now have more adaptable AI solutions. Because you could take the underlying technology. Large language models. Have it be used to generate emails. Turn around. And use that same exact technology to summarize financial documents. 

We'd be remiss if we didn't talk about the role of cloud computing. In democratizing AI access. 

The key moment. Sort of the introduction of GPU instances in the cloud around the 20 13, 20 15 timeframe. But the real game changer came. When cloud providers started offering specialized AA services. And pre-trained models around 2015 to 2018. 

The impact on business for this is thinking about startups. Eh, they have the ability to compete with larger companies by leveraging advanced AI capabilities. Subject matter experts. Have access to AI tools that were previously only available to large enterprises. And then the large enterprises. They have faster experimentation and deployment of AI solutions. 

Imagine a small online boutique competing with retail giants.  Using the same level of AI powered personalization. That really is the power of cloud based AI services. It may make you harken back to the days in the early two thousands. When cloud first started up here. It's that same inflection point now with AI. 

And from the impact and innovation. The time from idea to implementation. Has been drastically reduced. So in your business, you can think about now experimenting with AI applications. In days or weeks, not months or years. 

This is not without its challenges. 

It's opened up enormous possibilities. But it's also bringing challenges. Businesses need to carefully consider data privacy. The implications of that. Potential of course for vendor lock-in. 

And you'll start to see a trend towards edge AI. And hybrid solutions. These could offer the best of both worlds. The power of cloud AI. With the privacy and low latency of on-device processing. 

I'd encourage you to explore the AI services relevant to your industry. And if you haven't already. Take a look at what some of the services that are offered. By the cloud providers. Many offer free tiers or trials. Allowing you to experiment with these pretty powerful tools. Without any real significant upfront investment. Now let's move on to understanding GPT. And large language models.

GPT and LLM

The history is fascinating of how we got to where we are. But now we're going to go a little deeper into understanding GPT and large language models.  GPT stands for generative. The G pre-trained transformer. 

So let's break that down a little bit. Pre-training. Is learning patterns from vast amounts of data. You could think about this. As like your general education in school. Classes that you take. Across all different topics. 



You'll then hear the terminology. 

Fine tuning. This is taking the model. After it's general education. So after it's been pre-trained. And adapting it to specific tasks or domains. This is like getting a specialization. If you did an undergraduate degree, at least in the United States, you have a couple of years, often of  general ed classes? And then probably two years of your specialization. And then if you continue on and do advanced study, you continue to further specialize. So you think of fine tuning like that. And again, pre-training. GPT genitive. 

Pre-trained. The pre-training being learning patterns from vast amounts. So think about GPT. As  a digital polymath. With broad knowledge. 

 It is been pre-trained. On a huge amount of data. The internet. So some of the key capabilities, when we think about. GPT and these language models is natural language understanding. Contextual comprehension. It could do text generation. Creating human-like text. And if you have used chat GPT, PT two. Generate an email for you. 

It could generate an email. That looks pretty good. 

Another key capability is task completion. This would be following instructions and problem solving. 

An example of task completion would be providing a prompt for the model to create a recipe for  gluten free. Banana bread. And an itemized shopping list of the ingredients. 

You could think of GPT as a highly educated assistant. Who's read almost everything on the internet. 

Key. There is, it's an educated assistant. Which means. Remember that we still need to check. What it's returning. 

Another term. We need to talk about our. Foundation models. 

You may have seen this or heard it tossed around. 

The foundation model is. Large. Adaptable model. That serves as the basis for various applications. GPT that we've been talking about. That's considered a foundation model. 

Dali. Is a foundation model for images. Whisper is one for speech. 

When you see news articles of how much open AI or another vendor spent on training a model. These large models that they create that are costing tens of millions of dollars. Those now serve as a foundation model. And the implications for you. As a business decision-maker. Is that you get faster development of AI applications? 

If it took open AI several years to train a model. You now get to leverage. That model. To build an application on top of so much faster. You no longer have to train the model. 

You could think of foundation models. Like a very. Versatile employee.

Who could be quickly trained for various roles in your company? So if I go back to this, pre-training fine tuning. You could use the foundation model, just like it is. It's been pre-trade. Now you have this employee that could quickly write emails. Code analyze financial documents. Good solve basic problems. I do all different roles. 

You could also then find tune the model. And now this employee who's pretty versatile. Now you can really specialize them in a certain role. That's a fine tuning is using your data to fine tune it. 

Some of the other important AI technologies. That are going to come up over various podcast. Or you're going to see them. Or read a bottom or hear about them. Or things like computer vision. 

Image recognition. Object detection. 

Things that you're probably familiar with. Having a camera. That has phone functionality. 

But think about it from AI powered. Quality control manufacturing. 

Robotics. So here, we're thinking about physical task automation. 

Of course. A lot of people talk about. These robots that are taking over, which they're not. But you can think about robotics for warehouse operations. Amazon makes extensive use of robots in their fulfillment centers. 

And speech recognition and synthesis. 

So voice assistants, transcription. 

But could think about a call center automation with AI. Something really relevant to me, especially doing this is there's all this technology now to record these podcasts. That uses AI to do transcription for you. I can certainly tell you it's not a hundred percent. It is so far failed every time to get my last name. 

Correct. 

But then when we move on and think about another term, you're going to hear. Is multimodal AI. 

So if we break that down, a multimodal AI and a multimodal model.  Combine text. Image. And let's say audio. Multiple modals. Alright. Each one of those being one. All right. So now keep them bought them. They may not combine all of them. It may just do text and image. Maybe it does all three. 

You'll see this in action. When you look at AI. That could understand and generate. Both text and images. To give a more comprehensive analysis. Or for content creation. 

So you could see now how you could generate images, but imagine someone like Adobe using this. In there. Suites that they have for, to designers. Can it generate text for something you're building. And one of the Adobe products and the images. It's pretty impressive of some of these things you can start to think about. If you could ask the model to generate an image for you and text. 

Next we're going to go into some detailed examples of AI and how it's reshaping industries. If there's any part of this understanding, GPT are the large language models. That you're not sure about. Please drop me a note in the comments.

Send me a email reach out to me on LinkedIn. I be happy to spend another episode, just digging into these in more detail. Now let's go on to  more detailed examples. Of how AI is reshaping different industries.

Examples

Let's talk about.  Two examples in more detail than we did on Monday's episode. Monday, we were talking about document processing. Let's dig into that a little bit deeper. We talked about. Doing an annual report analysis. Summarization and tryna dentification. And the fact that this is shifting from manual reviewed AI assisted analysis. 

Let's think about this time. Of contract analysis.   In 2017, JP Morgan. Announced that it had developed and deployed. This new software called coin.  Coin stands for contract intelligence. 

Which is an automated document review system. For certain classes of contracts. They first deployed this to review. Thousands. Of their own credit contracts.

This software employed image recognition to identify patterns. In agreements. They've obviously pretty tight lipped about the details of this technology. They've shared some information that the algorithm. Uses unsupervised learning. And we'll go into supervised and unsupervised learning. In another episode. For now. Just think about unsupervised learning. Being just that it is not with human interfering. It's running automated. 

If you will. 

And the unsupervised learning in this case. Was digesting data. On the banks, numerous contracts. And with those, I was able to identify and categorize. Repeated clauses. 

So they reported that the algorithm classifies clauses. Into one of about 150 different attributes of credit contracts. For example. It may note certain patterns based on clause wording. We're a location in the agreement. 

This software is able to review in seconds. The number of contracts. That previously took a lawyers. Over 360,000. Man hours. 

Okay. So in. Episode one. When we talked about how to think about. Using this technology for document processing, whether it's annual reports. Whether it's looking at invoices, whether it's in this case, it's contracts. And I mentioned about how the services industries. In people who provide services. Where it's all human labor to do document processing. Have a huge risk. So just let that sink in for a moment that. With JP Morgan built in 2017. 

Was an AI project. 

That was able and seconds. To process contracts. That took. Before that. 360,000 man hours. 

So obviously their economic incentive. To develop the product is pretty clear. 

But what's even more impressive. The algorithm is more accurate than the human lawyers. 

So their investment in technology. It's not just about cost. But also about quality. Since Coyne improves the accuracy. Of their contract review process. 

Automated. Technology assisted legal review solutions. Or not new. 

But JP Morgan benefits. From the large scale and low variability. It hasn't credit contracts. 

The bank processes over 12,000 credit agreements per year. Which are far less complex than contracts that might better suit human review. Such as custom MNA agreements. 

In your business. Maybe the contracts are really varied. And you won't have that type of return. That JP Morgan experience. 

There's most likely documents though, that could be handled. By AI in your organization. It could very well be contracts. It could be invoices. But just think about what they're able to do. Identifying patterns in his agreements using a large language model to do it. 

If we go onto another example, customer service. 

Everyone's aware. And I mentioned it before, by AI powered chat bots and virtual assistance. I know they're great. They have 24 7 availability. They scale. They could be. Personalized using customer data for tailored interactions. A great example though, of where you go beyond. Just what we're used to seeing as the customer service chat bot. Once it gets really confusing these days, to be honest with you, because many of them are not upgraded. To be any sort of modern AI based. They're still very rule-based. 

So we have this. Pointed time in the industry where. Not everyone is upgraded chatbots. So it gets really frustrating, even still as a consumer to interact with the chat. And then you experienced some that are really delightful. 

But when you think about using AI for customer service, Let's step away from the chat bot. 

Let's talk about lemonade insurance. And their AI claims processing.

So maybe thinking okay, so it's clean, it's processing. 

Is that the same as document processing? It is, but it also isn't because in this case, They're driven. Towards what's the biggest problem we all have with our insurance companies. When you have to file a claim, how long does it take? To have that claim service. Right customer service. On claims processing. 

AI gym. The chat bot that handles claims for lemonade. Has renowned capabilities for handling claims in seconds. In June of 2024. It was able to settle a bicycle theft claim. In two seconds.  Breaking the record. That Jim hell before that of three seconds.

Their chief claims officer at the time Sean Burgess. 

Commented that what the AI really does. Is it allows the right claim to get to the right adjuster. 

Burgess said instant claim services from a chatbot is exciting to talk about. It's really improved 

Workload  management using AI. Is something that resonates.

So when you think about doing that, so again, it's customer service. But it's not your typical chat bot go beyond the chat bot. If you're thinking about using this in your business, or you've already implemented a chat bot. 

Start thinking about. How do you go beyond just. The webpage, if you will chat bot  interaction. Can you actually fully service the customer? If you can't. 

Can you do things like what lemonade ?   Going from the chat to getting it to the right adjuster. Then it could be handled much quicker. 

So imagine. How happy we all would be. If we could get claims handled faster. this is fantastic customer service on their behalf.  Think about where you could start to use these  in your enterprise. So from two different industries, two different examples. Of where you could apply this technology. Customer service. But thinking beyond the chat bot. Digging into. 

How do you provide better overall experience to the customer? And streamlined processes. And then also. Document processing  looking at contracts.  So two different ways to think about how you could apply these technologies. That could be game-changing. And as we're talking about how this. Changes. And is reshaping the competitive landscape. Just imagine this type of work that you could do. Imagine a small boutique company. Competing with retail giants by using the same level of AI powered personalization. 

Thanks to the power of cloud-based AI services. 

But so that nimbleness that you get. When you start to let your imagination go as to what you could do to improve your operations. 

Provide better service. There are pretty amazing opportunities out of us. So those are two different industries, two different examples. There's plenty more. We could go into.  I'm going to move on now to think about, the challenges and the considerations. It's not all. 

Rainbows and pink unicorns. All right. There are challenges and considerations to keep in mind. As we think about this technology.

Challenges and Considerations

Each of the challenges and considerations we're going to talk about. I think almost each of them deserves their own podcast.  We'll go back and have one explicitly on each of these that we're going to talk about. We're going to talk about four different challenges and considerations. The first one, hallucinations. We  touched on that on Monday at a high level. When you think about the hallucination issues? Again, this is AI generating plausible, but incorrect information. We have to look at it to really decipher is that. True or not true. 

The causes for this. And there's many, but in general, It's limitations in the training data. These models taking quite a bit of time to train.  So the data is stale. So it's limitations in the train data. Stale or not available. And model uncertainty. 

Some of the mitigation strategies. 

Human oversight again, proofreading checking it. Think about it as a junior assistant, in many cases.

The example. There was a. Pretty famous. Issue not too long ago. About GPT being used to write a brief for a lawyer. And it was. Generating fictitious cases. we will put on a docket to have a podcast just on that. 

The next challenge and consideration. Is the ethical considerations. Here you think of things of bias and AI systems. Again, this is reflecting and it's potentially amplifying societal biases.  These models are trained on the internet. So it's a reflection  of what is trained on.  AI recruiting tools. Could be potentially discriminated against certain groups.  Job displacement as an ethical consideration. There's a need for reskilling and new job creation. I was meeting with a. Radiology tech not too long ago. And the topic of AI came up.  She had a very good point of. It's not the AI. 

That's going to replace the radiology tech or the radiologist. It's the radiologist. Or the radiology tech that knows how to use AI. That's going to replace the other one. 

Transparency and explainability. We are certainly going to talk about explainability in more detail as we continue on this podcast. But this relates to understanding. The decision-making that the model is doing. This becomes really important in regulated industries like finance and healthcare. And really. Many others. You want to understand. How did it come to this decision? I want to make sure that it's. 

Coming up with correct answers and we know how it got there. Data privacy and security concerns. This is always a concern with every system. And it doesn't change. With the latest. AI systems. But here you think about it handling sensitive information. And ensuring data protection. 

Of course all the compliance with regulations, GDPR, CCPA, any industry specific rules. And you got to have data. Data governance. 

You have to appeal to establish protocols for data that's used in AI. 

. And then the need for AI governance. 



This is not governing the data, which that needs to be happening as well. 

You should have good governed data that you're using to feed into NASA system. But it's about how do you develop. Responsible AI practices. How do you develop ethical guidelines? And do impact assessments. So this need for governance is less of the software. And more of. The organization. Creating internal oversight, having ethics committees. Doing regular audits. Balancing innovation with risk management. 

Microsoft has an AI ethics review board. Amazon has other things that they do. Google has a, some more thing. If you look towards the large vendors, They're fairly public about what they put in place. Great models to follow. 

We're  going to move on to the future outlook. It preparing for AI integration. 

The emerging trends in AI. We talked about multimodal AI. Integrating taxed image and audio understanding. 

The next is gonna be kinda like this AI human collaboration. Kind of augmenting our capabilities. 

There's a fantastic book that I just started reading that is titled. 

CCO intelligence. About how do we co-exist and how do we leverage these systems? The other emerging train you'll see is like edge AI. Again, processing data locally. For fast and more private operations. 

Oh, you think about the strategies for you as an executive to prepare your organization. Really we've got to be thinking about assessing AI readiness. If you're interested in discussing that in detail or looking at digging into that. I have lots of material on that workshops, maturity assessment. But you've got to think about this. 

You got to identify the high impact use cases. You really have to focus on the value creation. 

You certainly have to invest in data infrastructure and AI talent. 

Not getting into the arms race of hiring talent necessarily. You may need to. But think about building up the staff that you have. 

 Fostering a culture of innovation. This is where you can really start to encourage. Experimentation and learning. 

When you look ahead or certainly see these trends towards edge AI and hybrid. Offer the best of both worlds. AI with privacy in the latency. 

Those are things that are certainly coming in. It's really important. As you think about these strategies from AI readiness, high impact use cases. Data infrastructure. Or significant importance of mapping your AI strategy to your business goals. 

And you could think about. This concept, if you will. Of AI first companies. And what that means for traditional businesses. So you think about AI first? You've probably heard the term of cloud first. Or digital natives. I don't know, there's no legacy infrastructure. And they start in the cloud. So when you think about the small companies, That could start now. Using the AI from day one. 

Significant at joy that they get, it doesn't mean. That the large companies get left behind. We just need to start to adapt  how do we become more agile? How do we enable more people? How do we make sure that we're delivering on high-impact use cases?

Conclusion

That was quite a bit. So when you think about some of the things that we covered. We got through the history, leading up to where we are just at a high level. So you have some sort of reference to it and could have a frame of reference. As we look at these technologies. We talked about GPT and large language models. In some more detail. Starting to uncover things like pre-training fine. 

Tuning multimodal. It starts to talk about. How you could use it in your business. I went through two really good examples. Of one from a. Document processing. And really the amazing system at JP Morgan built. And how you could think about  

doing content analysis and it's different way than what you're doing today. And then customer service, I've really going beyond the chat bot. And really digging in of how could you use it for workflow? How could you use it to get the right documents, the right information to the right people? 

We discussed some of the challenges and considerations. Whether that's hallucinations the need for AI governance. And then we looked at some of the trends. And the strategies that as an executive, you need to be thinking about. As you prepare your organization. 

So the really is, competitive advantages for early adopters. 

Don't worry if you have not started adopting these technologies yet. A lot of people are. Playing with AI. It's okay to say that they're a. There's not a massive percentage in production. People were still trying to figure this out. So don't feel like you're behind. If you're not. Exploring these technologies yet. 

It's never too late to get started. As you look at some actual next steps. Really take a close look at. Continue to educate yourself. And your teams on these AI capabilities. 

If you're not doing anything with it yet. Just start small. Have pilot projects. So you get organizational experience. Look for partners and consultants that may have expertise to help. In really think about developing a roadmap. 

For AI integration organization. Even if you're already on your journey. All right. To percentage of companies that are actually realizing value. And ROI on AI. Is about 5.6%. So that's less than the return that if you put the money in the market, so everyone's at an early stage. 

I invite you to share your experiences or questions. For future episodes. And please feel free to message me in the comments to reach out. Coming up next. On Monday. The episode's going to be again, this a seven minute Mondays. It's going to be why understanding key. AI technologies. Is crucial for executive decision-making. 

So we're gonna start to get a little bit deeper into these technologies. And then on Wednesday in our deep dive. We're going to go from machine learning. To large language models. And the strategic implications for business leaders. So you're not going to want to miss these episodes. 

Please subscribe, follow the podcast. Reach out. If there's anything I can help with. Thanks for your time. 

 That's all for now.