Why Context Matters in AI
Author Ram Bala advises executives on the present and future of AI.
Ram Bala has a keyword for the future of enterprise-level AI solutions: context.
ChatGPT and other generative AI tools are now household names, and the opportunities for AI in business are abundant. But Bala, professor of business analytics at the Leavey School of Business and co-author of the new book The AI-Centered Enterprise, cautions against thinking of AI as a magic wand in its present state. Why? Because it doesn’t always consider context.
“Putting AI ideas into practice or getting them to work effectively for long periods of time in companies is actually very challenging,” Bala says. For example, a company can identify and use a supply chain optimization algorithm, but that algorithm won’t operate in a vacuum. Experienced team members will have opinions on their own operational areas. There are legal contracts and different regulatory environments to consider. “The full context is not always captured in the structured data used to create machine-learning models.”
Bala’s research and scholarship focus on improving such models and creating a better framework for businesses to use them. The AI-Centered Enterprise, an Amazon best-seller, provides such a framework and focuses on the benefits of context-aware AI. Additionally, Bala has tailored the AI-Centered Enterprise (ACE) framework for the Leavey Executive Center for use in custom education and programming for companies.
Context-aware AI goes beyond analysis of basic data to include more complex, nuanced information such as a company’s decision-making patterns or factors specific to its particular industry. “Context-aware AI is AI that not only understands content but also intent,” Bala says. “For example, a procurement contract can be viewed through a different lens by a supply manager and the company lawyer. Context-aware AI provides answers based on the user’s persona, not just the content of the document.”
To the average user, generative AI tools may feel revolutionary right now. But in truth, they often struggle to capture any company or organization’s greatest strengths: the expertise and deep knowledge of its people. Unless those people have written extensive documentation to feed to AI, the AI has no way to capture that expertise — for now. Bala sees current iterations as a starting point for something more comprehensive and capable in the future.
“It’s not just about finding the right model for the solution,” Bala says. “Let’s identify those areas of friction where people seem to spend a lot of time analyzing the data and not making sense of it at all, and where there’s a lot of time spent on this back and forth. Those are the opportunities we should be going after for AI.”
The ACE Framework
The prevalence of AI means the field is no longer solely the realm of tech-minded thinkers. So Bala and his colleagues have developed guidance within ACE that is clear and user-friendly for executives of any background, non-tech folks included.
For instance, Bala offers the three Cs: Calibrate, Clarify and Channelize. Calibrate relates to those “opportunities” he mentioned above. It’s about doing a thoroughly honest and human analysis of where a company’s pain points lie. To clarify is to assess the state of current operations, especially in terms of how they relate to specific pain points. The final step, channel, means finding appropriate AI tools and putting them to work on those pain points. Importantly, the solution comes only after in-depth analysis and discussion. Too often, companies try to implement solutions first, Bala says, without deep strategy, which can result in a decentralized mishmash of ineffective tools.
He also runs into the common misconception that AI works only for prediction or problems that require higher-order thinking. “It is true that Large Language Models are prediction models,” he says, “but they predict code and language which actually enables the execution of routine tasks autonomously.”
For the Leavey Executive Center, Bala’s AI-Centered Enterprise is an ideal framework for custom executive programming and education. It offers companies the opportunity to set a baseline by identifying places where generative AI is already in use, then map its potential evolution to a more evolved context-aware AI iteration. It helps companies create a shared vision for both enterprise-wide AI and for information flow. Smart and silo-free information flow is a critical factor because context-aware AI requires access to both traditional datasets (things that easily fit into spreadsheets) and unstructured data (things such as videos, images and text documents) to work properly.
Essentially, ACE offers a way for companies to strategically restructure their thinking about AI, starting clearly with needs and moving to solutions rather than the other way around. “Our firm belief is that AI as a mechanism for increasing task productivity is limited in scope and will find limited success,” Bala says. “The ability of context-aware AI to understand natural language makes it an enabler of communication and collaboration, ensuring that it can play a ‘central’ orchestration role in amplifying team and organizational productivity.”
Bala and colleagues offer a clear look at the modern AI “stack” and what’s capable now, along with what’s likely to be possible in the not-so-distant future. Importantly, they also discuss the areas where AI does not perform well now and is not likely to perform well in the future. This includes any relationship-based work that requires trust-building. In such cases, AI may play a supporting role, but not the main part.
“It’s about designing organizations where people and AI work together seamlessly,” Bala says, “where the technology adapts to us, not the other way around.”