Credit: VRVIRUS/Shutterstock
With the right planning and a little experimentation, companies can use generative artificial intelligence to help their employees work smarter. Although it’s possible that some positions will be replaced entirely by AI, the technology could end up creating new jobs and opportunities for collaboration between AI and humans.
At the World Economic Forum annual meeting last month in Davos, Switzerland, MIT Sloan professor joined a panel discussion about the opportunities and challenges of using AI to drive efficiency in the workplace. She talked about the need for businesses to define what they think success should look like, invest in data infrastructure, compensate their workers for lending their expertise, and much more.
Li discussed these topics with Aditya Bhasin, Bank of America’s chief technology and information officer; Jim Stratton, the chief technology officer at Workday; and Nitin Mittal, Deloitte’s global AI Leader. The panel was moderated by Andrew Hill, a senior business writer at the Financial Times.
Li, who studies how AI impacts the nature of work, offered four tips for companies experimenting with the technology.
Determine what AI success should look like at your company.
Large language models like Claude and ChatGPT are smart and incredibly capable but require training to be useful, Li said. It’s not unlike training a talented employee by giving feedback and guiding them with mentorship. Li said she sees proper model training as a top barrier to further AI adoption.
“How do we move from an AI that is a genius to one that’s capable within an organization?” she said. One solution is for companies to build out their data infrastructure with lots of clear, labeled examples of what they think a good job looks like.
Bank of America has found success with its AI-powered virtual assistant, Erica, which provides more than 25 million customers a month with services, from conducting transactions to providing insights about how to manage their finances, Bhasin said.
Erica is a good example of using internal data to build a model to perform specific tasks, but a lot of companies haven’t invested in the internal infrastructure to make that possible, Li said, adding, “I think that’s a big challenge going forward.”
Make investments in data infrastructure, and pool data across organizations.
Mittal, Deloitte’s AI leader, pointed to customer service, software engineering, and R&D assistance as areas where AI is fully stepping into roles normally filled by humans, particularly when it comes to working in call centers, writing coding, and testing software.
Li noted that all three areas have “a tremendous amount of often public, often fully pooled data” that can be used for predictive purposes and to build models and identify patterns.
“I think that a lot of the gains to AI development are going to be slow unless we make concerted investments in data infrastructure and understanding where [we can] get a large enough corpus of examples of people doing their jobs well,” Li said. “You need examples, and you need those examples labeled properly.”
Small companies might not have the same amount of in-house data that larger companies do. To compensate, smaller companies might have to buy external data, which might not accurately reflect their expertise.
“I think we need to think about ways of building data infrastructure that allow data to be pooled across organizations that might even be competing with each other,” Li said. “We’re not going to make progress unless we get some scale.”
Incentivize your workforce to collaborate with AI.
Employees may object to having their work or data used to train AI models without their consent or compensation. For example, a video recording of a teacher conducting a class may be helpful for students who are unable to attend in real time and want to watch it online later, Li said. In cases where the teacher does a good job, the recording could be added to a repertoire of training videos that could then be used to build models to do the same job. But this diminishes the teacher’s expertise — why have teachers when you can just watch videos online?
“[Humans are] paid because expertise is rare,” she said. “The moment your expertise stops being rare, you stop getting paid.”
Related Articles
Li said that companies should incentivize their employees to share their knowledge and work with the technology, or compensate them for doing so. As a result, employees wold be more likely to want to be active stakeholders in the information-sharing process.
Determine what roles you want AI to be a substitute for, and contemplate what new jobs might be created.
Li said that AI substitution is inevitable for some traditionally human-staffed positions or tasks, but wondered, “Is that a bad thing or not?” Imagine, for instance, a tired radiologist trying to read a patient’s medical results in the middle of the night. In this case, having AI do the task might be the preferred route. “I think there’s a difference between replacing jobs and replacing specific subtasks of roles,” Li said.
Technology has also created new jobs and tasks, she said, pointing to areas like search engine optimization. “We have the agency to decide what we want some of those jobs to look like,” Li said.
Leading the AI-Driven Organization
In person at MIT Sloan
Register Now
