By the end of 2022, within just two months of its launch, ChatGPT had attracted 100 million users, revolutionizing the accessibility of AI and becoming the fastest growing application in history. Its ease of use sets it apart from previous AI models, requiring no machine learning expertise to interact with and derive value from. Similar to disruptive technologies like the personal computer or the iPhone, a Generative AI platform can create a wide range of applications, easily accessible to users of all ages and education levels, from anywhere with Internet access.

The Evolution of Generative AI in 2023

Generative AI chatbots, powered by platform models with large neural networks trained on diverse, unstructured data, offer exceptional flexibility. Unlike previous AI models, platform models can perform a wide range of tasks, from summarizing technical reports to generating business strategies and recommending recipes. This flexibility sometimes leads to less accuracy, due to the need for enhanced AI risk management.

With the right guidance, Generative AI can innovate business practices by enhancing existing processes and launching new applications. Imagine a customer call: AI models specifically trained for this purpose can now make real-time sales recommendations based on the content of the conversation, tapping into multiple data sources such as internal customer data, market trends, and social media influencer data. Additionally, Generative AI can provide salespeople with sales templates to modify.

This example illustrates the impact of Generative AI on the job, but it has the potential to benefit nearly all knowledge workers. Rather than simply automating tasks, the value of Generative AI lies in its integration into everyday tasks like email and word processing software, resulting in significant productivity gains.

CEOs must decide whether and how to use generative AI now. Some seek to gain a competitive advantage by adopting AI quickly, while others want to experiment cautiously. Assessing technical, technological, data infrastructure, operational, and risk management readiness is essential before implementing AI.

This article will provide CEOs and their teams with a primer on Generative AI before they adopt it. We’ll start with an overview of Generative AI to understand the state of AI and the options. Then, we’ll explore four areas where companies can use Generative AI to improve performance, address technology, cost, and operational issues.

How is Generative AI different from other AI models?

Generative AI differs from traditional AI and analytics in that it efficiently generates new, often unstructured content such as text or images. The core technology behind it is the Foundation Model, which uses transformers, with “GPT” standing for “Generative Pre-trained Transformer”. Transformers are a type of neural network powered by deep learning, a key technology for recent AI developments.

Platform models differ from previous deep learning models in a few key ways. First, they can be trained on exceptionally large and diverse unstructured datasets. For example, large language models (LLMs) — a type of platform model — can be trained on a broad range of publicly available text on the Internet covering a wide range of topics. In contrast, traditional deep learning models are typically trained on more specific datasets, like models focused on recognizing specific objects in images.

Furthermore, typical deep learning models are designed to do a single job, such as classifying objects in an image or making predictions. In contrast, a platform model can handle multiple tasks and generate content thanks to its ability to learn patterns and relationships from the data it is trained on. This ability allows models like ChatGPT to answer questions, and diverse creations like DALL·E 2 and Stable Diffusion to generate images based on descriptions.

The flexibility of platform models allows companies to use the same model for multiple aspects of their business, something that was rarely possible with previous deep learning models. For example, a platform model with information about a company’s products can support both customer inquiries and product development engineers. As a result, companies can deploy applications faster and reap the benefits of them.

However, current platform models have limitations. They can sometimes produce inaccurate responses, a phenomenon known as “illusion,” and they cannot always provide a transparent rationale or source for their responses. Therefore, businesses should be cautious when integrating AI into applications that may have adverse consequences or require transparency. Additionally, current generative AI is not well-suited to directly analyzing tabular data or solving complex numerical optimization problems. Researchers are working to address these limitations.

Generative AI enhances work by automating, augmenting, and accelerating tasks. Our focus is on how AI can be created to enhance work rather than replace human roles.

Read more on: 11 Real-World Applications of Machine Learning in E-Commerce

In addition to text-generating chatbots like ChatGPT, Generative AI also includes a variety of content types such as images, videos, audio, and code. It is used for a variety of tasks within organizations, including categorization, editing, summarizing, answering questions, and creating content. Each of these functions transforms workflows across departments, benefiting business operations. Here are some examples:

Classify:

• Fraud detection analysts can use Generative AI to identify illegitimate transactions from transaction descriptions and customer documents.
• Customer care managers can categorize customer call audio files based on caller satisfaction.

Edit:

• A copywriter can ensure grammatical accuracy and tailor the copy to the client's brand image using Generative AI.
• A designer can remove old logos from images.

Synthetic:

• Production assistants can create highlight videos from hours of event footage.
• A business analyst can create a Venn diagram summarizing the key points from the director's presentation.

Answer the question:

• Production staff can seek technical guidance from a Generative AI “virtual expert” on operational procedures.
• Consumers can ask the chatbot questions about furniture assembly.

Sketch:

• Software developers can leverage Generative AI to generate code or suggest ways to complete existing code.
• Marketing managers can use Generative AI to draft different versions of campaign messages.

As Generative AI continues to evolve and mature, it can be integrated more deeply into business workflows, automating tasks and performing direct actions, such as automatically sending meeting summary notes. We are already seeing the application of tools in this work.

4 areas integrating Generative AI into work

CEOs must prioritize exploring Generative AI as a necessity rather than an option. Generative AI delivers value in many use cases, with accessible technical and economic requirements. Failure to act quickly can result in falling behind competitors. CEOs should work with their executive teams to strategize how and where to deploy Generative AI. Some may see it as a transformative opportunity, allowing them to reshape many aspects of their business, from research and development to marketing, sales, and customer operations. Others may choose a more incremental approach, starting small and scaling as needed. Once decisions are made, AI professionals can follow the appropriate technical roadmaps to execute the strategy, depending on the use case.

Much of the use of Generative AI (though not necessarily at its full potential) in an organization will come from employees leveraging the integration features in their existing software. Email systems can help generate message drafts, applications can generate presentation drafts from descriptions, financial software can generate written descriptions of notable financial reporting features, and CRM systems can suggest ways to interact with customers. These features help improve the productivity of knowledge workers.

However, Generative AI can also bring about bigger transformations in specific situations. Here, we explore four examples of how companies in different industries are applying Generative AI to their workflows.

1. Increase programming productivity using AI

Software engineering often involves writing laborious code, which leads to a backlog of feature requests and bug fixes at companies due to a lack of skilled engineers. To solve this problem, we need an AI-powered code completion tool that integrates seamlessly with coding software. Engineers can describe their code in natural language and AI will suggest different code snippets that match the description.

McKinsey research shows that the tool can speed up code generation by up to 50% and aid in debugging, improving product quality. However, it’s important to note that Generative AI is meant to support, not replace, skilled engineers. More experienced programmers will benefit the most, while newbies may see little change. Ensuring code quality and security remains the responsibility of engineers. The tool is cost-effective, with subscription fees ranging from $10 to $30 per user per month, and requires minimal in-house development.  

Licensing and intellectual property considerations should be discussed with the vendor. Having a small team that oversees tool selection, monitors performance, and ensures compliance with intellectual property and security requirements will simplify implementation. With an off-the-shelf SaaS solution, the additional computing and storage costs are negligible.

2. Support Relationship Manager in updating data

Companies can choose to develop custom Generative AI applications using platform models, which require higher upfront investment but are better suited to meet specific needs.

For example, a corporate bank sought to improve the efficiency of its Relationship Managers (RMs) by creating a solution that accessed the platform model via an API. The tool quickly analyzed large documents, providing aggregated answers to queries from the RM.

Additional layers ensure a seamless user experience, integrate with corporate systems, and implement risk and compliance controls. While the costs incurred are primarily related to user interface development, which requires collaboration between data scientists, ML engineers, designers, and front-end developers, ongoing costs include software maintenance and API usage fees.

Costs vary depending on the model chosen, third-party fees, team size, and minimum product creation time, but the benefits are faster RM analysis, increased job satisfaction, and the ability to mine valuable information.

3. Support customer service in more important tasks

The next level of complexity involves refining the underlying model. For example, a company operating in a niche industry where fast customer service is critical used an underlying model to optimize conversations using high-quality customer conversations and industry-specific Q&A data. The goal was to use a Generative AI customer service bot to ensure responses were quick and consistent with the company’s image, thereby improving customer satisfaction.

The company did this in phases, starting with internal testing and employee feedback, moving on to observing customer support conversations, and eventually moving to automation with human oversight. Generative AI allows RMs to focus on more complex questions, improving efficiency, job satisfaction, and customer satisfaction.

Achieving these benefits requires significant investments in software, cloud infrastructure, and engineering talent, along with increased internal coordination of risk and operations. Typically, adapting platform models is 2-3 times more expensive than building software layers on top of APIs. These additional costs come from human and third-party costs associated with cloud computing (to adapt a self-hosted model), or API usage (to adapt third-party APIs). Solution implementation also requires expertise from DataOps and MLOps professionals, and input from multiple functions, including product management, design, legal, and customer service.

4. Promote pharmaceutical research

The most complex use cases for Generative AI are when there is no suitable baseline model available and the company has to build one from scratch. This can happen in specialized domains or when dealing with unique datasets that differ significantly from the datasets used to train existing baseline models, such as in the pharmaceutical industry.

In one case, a pharmaceutical company needed a tool to assist drug discovery scientists who were working with massive amounts of microscope image data. They decided to build their own model due to the novelty of the drug. The model, trained on a combination of real-world images and their internal microscope dataset, helped scientists understand the relationship between the drug’s chemical makeup and the microscopy results. This accelerated the research process and improved the identification of cell features relevant to drug discovery, leading to a more efficient and accurate research process.

However, building a model from scratch is much more expensive, 10 to 20 times more expensive than developing software using an existing API model. This cost difference arises because it requires a larger team to build, including PhD-level machine learning experts, increased computing and storage costs, and the complexity of the modeling process. Costs vary depending on the desired model performance, model complexity, data set size, team, and computing resources. In this case, the majority of costs come from the engineering team and the cost of cloud computing resources and services.

The company will identify the need to upgrade key technology infrastructure, such as GPU servers for model training, distributed training tools, and MLOps best practices to control project costs and timelines. The business will also need to handle big data, including collection, integration to ensure data consistency and integrity, and cleaning to remove low-value and duplicate data. When training models from scratch, rigorous testing is essential to ensure accurate and secure output.

Conclude

Overall, the role of CEOs in harnessing the potential of Generative AI is crucial. As we have seen, Generative AI offers many disruptive applications across a wide range of industries. To fully capitalize on its benefits, CEOs must leverage its potential, invest in the necessary resources, and foster a culture of innovation within their organizations. With a strategic vision and a proactive approach to integrating Generative AI into their businesses, CEOs can enhance competitiveness, efficiency, and innovation, ultimately leading their companies to success in the rapidly evolving AI landscape.