The Strategic Opportunities Of Advanced AI: A Focus On ChatGPT

What is ChatGPT?

ChatGPT is a chatbot application that leverages a generative pretrained transformer (GPT), a deep-learning neural network model. The GPT is trained on vast amounts of data, such as internet content, and can generate human-like language in response to an input prompt by transforming an input sequence (a request or a question, for example) into an output sequence (the response or answer, respectively).

What technical developments made ChatGPT possible?

Neural networks have been around for over 60 years, but for decades, they were mostly used for small-scale "toy" problems. However, several technical breakthroughs in the past 20 years enabled the emergence of ChatGPT. These include:

1. Learning algorithms that can handle deep learning neural networks with many layers (previously, only one or two layers of connections could be accommodated, which limited the representational power of the models).
2. Paradigms for training and applying generative networks that can produce human-like language responses.
3. Models of attention for dealing with the complexity of natural utterances and images, for example.
4. Access to large amounts of data for training, such as the contents of the World Wide Web, a corpus of digital books and social media posts.
5. Access to enormous amounts of scalable computing resources for training.
6. Reinforcement learning methods that can be used to incorporate safety features into the models.

What are the business opportunities?

We can think of business opportunities presented by ChatGPT in terms of three categories.

1. Low-Hanging Fruit: There appear to be many useful incremental applications of ChatGPT and other large language models (LLMs). Everyone has access to the same magic box, however, and so although these may be highly valuable, they aren't necessarily strategically advantageous, generally speaking. Examples include writing blog posts or other written artifacts, accelerating code development, summarizing text, enhancing research and analyzing text for themes. ChatGPT can also enable task automation with a natural language interface that understands intent and can execute tasks using various applications, such as providing scalable and cost-effective customer support.

2. High-Hanging Fruit: Organizations can build their own proprietary LLMs on proprietary data. This approach has a high potential for providing a strategic advantage, but it's currently technically difficult, requiring highly specialized skills and knowledge, enormous amounts of curated data and, typically, millions of dollars to train the models. Additionally, there's a risk the process won't converge to a useful model until well down the road, so skillful adjustment and iteration will be required.

3. Medium-Hanging Fruit (The “Goldilocks Zone”): There are ways to apply ChatGPT and other LLMs that could enhance an existing strategic advantage at a reasonable level of cost, technical complexity and risk, such as:

• Wrapping proprietary systems in an existing natural language interface (e.g., LangChain and Gradio, for example) to lower the bar to access and make the core tools more valuable.

• Integrating ChatGPT or another LLM with other tools and services to create differentiated offerings that embody unique and defensible value.

A bigger opportunity exists for those executives who can look ahead of the current horizon to ask how advanced technologies can drive organizations to re-engineer their business processes. This approach is what we’ve been advising our customers in biopharma, materials science and chemistry for over 20 years.

The emergence of digital computers, followed by connected networks and then social networks, drove sea changes in customer expectations and the way organizations needed to set themselves up for success, leading to completely re-engineered business, operational and enterprise models. Similarly, executives should now look ahead of the incremental applications and even the integration applications to ask: What are the implications of this AI development regarding how we refactor the relationship between the worker, the technology (ChatGPT or other advanced AI) and the problem domain within our organizations? Moreover, what will training look like in this new world? What should we be training people on, and what should we stop training them on, given this new worker-technology domain?

ChatGPT is one example of a powerful AI tool that has the potential to revolutionize the way we interact with technology and solve complex problems. By understanding the technical developments that enabled its creation and looking beyond the low-hanging fruit, visionary leaders can navigate this new frontier of AI technology and use it to potentially move ahead of the pack.