The Boston Consulting Group (BCG)'s recent announcement of a strategic collaboration with Intel to deliver "enterprise-grade, secure AI" is an example of how industry leaders have been energized by the rise of generative AI to learn how this new technology can be used across businesses. DIGITIMES Asia recently spoke with BCG Henderson Institute Global Director François Candelon about the objectives behind this collaboration, and about the transformative potential of generative AI more broadly.
Q: Why is BCG building its proprietary generative AI system with Intel?
To address the world's most important challenges and opportunities, we're collaborating with the world's leading experts in AI. Together, we are bringing fully custom and proprietary generative AI solutions to enterprise clients across industries, while keeping private data isolated in trusted environments. Because generative AI is an emerging and dynamic space, it's critical that organizations pick the right technology provider to power their generative AI journeys.
Today's popular, pre-built large language models (LLMs) rely on a public cloud or non-customer-controlled infrastructure and can take three to six months to go from data ingestion to a fully trained custom model that's ready to deploy.
Intel's unique combination of Xeon Scalable Processors, AI optimized hardware (Habana Gaudi), and an optimized software stack creates an unmatched ability to develop and deploy fully customizable enterprise scale generative AI solution for a variety of businesses across different market sectors. The built-in flexibility of Intel's solutions was a perfect complement to BCG's AI and engineering services to design and deploy enterprise grade generative AI solutions, while expertly navigating the people/process/policy changes that come with powerful organizational change. Finally, Intel was able to meet our information security and data privacy requirements, which are among the most rigorous in the world.
As a result of this partnership, we expect to be able to enable generative AI solutions for enterprise clients at scale in weeks rather than months or years.
Q: Do you customize the generative AI system for each corporate customer? Is it deployed on the cloud or at the edge?
BCG customizes solutions for each client. In fact, we have partnerships and collaborations with several generative AI providers such as Google, Intel, and OpenAI, because these transformative solutions need to be tailored to the specific needs and capabilities of each company—and even tailored to each use case. LLM providers should be selected by balancing the requirements of a given use case, for example, accuracy, data security, and cost. BCG's approach to customization includes determining whether they should build in the cloud, on premise, at the edge, or through a hybrid approach.
Q: How many domains of knowledge does your LLM system cover? Are banking and customer service best suited for B2B applications?
Customer service is an exciting application of generative AI and banks are often early adopters of such technologies, so it is not surprising that these sectors are garnering the most attention. However, every company can find significant value with generative AI, no matter their industry. And every domain has several use cases spanning productivity and innovation. In fact, selecting the optimal sources of data for unique use cases is also something that we tailor to our clients' unique business needs.
We support companies in using their own proprietary data with generative AI to be able to innovate and grow their business while making sure that their data is safe and secure. Because the type of data used to train the model defines the functionality that the model can provide, a company's unique data is its source of competitive advantage with generative AI.
Consider a materials company. Its unique data is its chemical data. By training the generative AI model with chemical data and material properties, it can create an application for inverse design. This is where you write a prompt asking for specific material properties and the model generates the material that has those properties—even if the material has never been designed before. Let's say the company wants to create an ultra-light weight bulk material that is superconducting at room temperature. This type of material is illusive today, but generative AI could make it possible.
Another example is manufacturing companies; they can train a generative AI model with their unique computer assisted design data and create a model for generative design. Future designs can then be automatically generated by simply having a conversation with the computer on the parameters they're looking for: say they want a new car to be 15% lighter, but also made from 15% stronger materials, and designed with 20% fewer parts, and so on. The generative application could then design the new product based on the companies' historical design data.
As a knowledge-based company, BCG is also building our internal generative AI applications using our own knowledge assets, such as our full repository of slides and reports. In the future, our case teams can use this when supporting our clients. For example, let's say our client is an oil and gas company that wants to identify areas of growth. Today our teams need to pull disparate data sources and find various experts in our ecosystem. In the future, all our rich expertise and information could be available to the case team with the right prompts of our generative AI application.
To summarize, we work with each of our clients to think beyond the obvious and use generative AI for its competitive advantage. This technology has huge potential to be disruptive across industries. It's not just about productivity or customer service. It's about creativity!
Q: Considering the powerful nature of generative AI, how will it disrupt the labor force and impact competition among companies?
There is a famous quote that I like to cite on this topic: "Humans won't get replaced by AI, humans will get replaced by humans using AI." In the labor force, specifically, companies will compete on how effectively their employees use AI tools. I think this is especially true for generative AI, where human interaction with the tools is critical because there is no truth function.
Generative AI gives you the most statistically likely answer based on its training data. For example, when you ask ChatGPT, "What is two plus two?" the most statistically likely answer from the web-based data it is trained on is four—so it will tell you that the answer is four. But there are contexts for which the training data is less robust, which means human supervision of the output is vitally important to ensure quality.
As a result, you need to train your people to use and supervise generative AI
effectively. Otherwise, your employees could be a source of risk from "shadow AI." For example, generative AI has the power to "democratize" AI and make employees believe they can do new things, such as code in python when they have no programming background. But when people use generative AI outside of their expertise, it can cause unintended problems—such as publishing vulnerable code or leaking proprietary information. At BCG, we provide our clients with robust guidelines, AI governance principles, and best-in-class Responsible AI frameworks to help them mitigate these risks.
In the short term, companies will need to carefully govern how generative AI is used in their organizations, but over time everyone in the company will be using these tools. This will create a meritocracy, wherein some people are better at working with the tools than others. The impact of generative AI on productivity is so high that employees who are effective at using the tools will far out-shine employees who are slower to adopt them. And some current human capabilities will become less important, like the ability to analyze data, create summaries of long documents, or develop personalized content.
The situation today is very similar to that of the late 19th century—before we had widespread use of electricity. AI's impact is of this magnitude and, in my opinion, is much more important than the rise of the Internet.
Q: Given that generative AI requires substantial energy and computing power, is the world generating enough energy to support its rapid expansion?
Today the energy consumption of generative AI is a key risk. But the industry is evolving rapidly. I think that in a few years, many generative AI applications will use smaller models that require much less computation, such as models that can run on a laptop with one GPU. There will eventually be many of such models.
The key question, as you said, is: Will the progress of producing computing power match the pace of the specific demands created by generative AI? I believe this is a significant opportunity space for Taiwan because hardware is a determining factor in how this will evolve. Taiwan needs to be making the right chips for AI, specifically TPUs and GPUs, and should be actively involved in optimizing their design.
I like to quote Michael Porter, a professor of strategy at Harvard Business School, who said that competitiveness of a country depends on the capacity of its industry to
innovate and upgrade. Taiwan should therefore not only be in the business of manufacturing. Taiwanese companies also need to innovate and upgrade existing chip designs, otherwise they will be disrupted.
I also believe a significant number of new jobs will be focused on innovative chip design going forward. This could offset job losses from other generative AI applications.
François Candelon bio:
Francois Candelon is the Global Director of the BCG Henderson Institute, Boston Consulting Group's think tank dedicated to exploring and developing valuable new insights from business, technology, economics, and science by embracing the powerful technology of ideas. He is also a leader of BCG GAMMA's AI@Scale effort for technology, media, and telecommunications companies. GAMMA is BCG's data science and advanced analytics unit.
BCG Henderson Institute Global Director Francois Candelon in its Taipei office.