recent

The AI Playbook: 6 steps for launching predictive AI projects

5 ‘extraordinary turnarounds’ for global well-being

Close proximity is the key to knowledge spillovers among startups

Credit: Mimi Phan / HAKINMHAN / iStock

Ideas Made to Matter

Artificial Intelligence

Want to make the most of generative AI? Use your imagination

By

As the chief scientist and technical fellow at Microsoft, Jaime Teevan is responsible for integrating artificial intelligence into Microsoft products. She was still skeptical about the powers of generative AI when she met with Microsoft CEO Satya Nadella and OpenAI co-founder and CEO Sam Altman about a year and a half ago to discuss GPT-4, the large language model that fuels ChatGPT. 

Teevan, SM ’01, PhD ’07, knew the right questions to ask and how to break the model. Yet “it was really cool to sit there iterating with the model, seeing the way it maintained context and the way it was able to handle ambiguity and deal with conflicting constraints in a really nuanced way,” she said.

That was a “switch in time” as she realized the potential of the technology, Teevan said. She even found herself pulling over on her short drive home.

“I sat there, and I screamed. I didn’t think we would ever be where we are today in my lifetime, honestly,” she said.

During a keynote talk at the 2024 MIT Sloan AI & ML Conference, Teevan discussed how GPT-4 has evolved, how generative AI is changing work in the short and long terms, and the future of online search.

How AI will change work

That eye-opening meeting began a sprint to bring LLMs to everyday workers, Teevan said.

The model itself was just the start. It also needed an infrastructure — the computing power necessary to run the model continuously, and the ability to securely input and analyze sensitive data and integrate models into the workflows of knowledge workers.

Microsoft’s research has found that generative AI helps people work faster while retaining relative consistency in quality. “People are feeling more efficient, and they’re enjoying work more,” Teevan said. One study demonstrated that generative AI can be most beneficial to workers with the least experience, helping them match the performance of colleagues with longer tenures.

Teevan said she is seeing AI change how people work in a few ways. In the short term, users are taking advantage of gains in efficiency or production, such as using ChatGPT to create content and to summarize documents, such as meeting transcripts.

“We can help you generate content to reply to an email or start a new document so you don’t have this blank-page challenge anymore,” she said.

There’s even more opportunity in the longer term, Teevan said. This includes using AI in entirely new ways.

“I don’t think we’re being imaginative enough,” she said. “We’re thinking about how it’s going to help us do our tasks better. We’re not thinking about how it’s going to help us collaborate and work together. We’re thinking about how it can do stuff that we know more efficiently and faster. We’re not thinking about how it can help us do new things or think of things in new ways.”

Teevan herself uses AI to see things in new ways and for brainstorming and collaboration. “I really like the way the model helps me see other perspectives really quickly, helps me ideate and think through new things,” she said.

Future trends for generative AI

Related Articles

Accelerated research about generative AI
How generative AI can boost highly skilled workers’ productivity
It’s time to make sure everyone understands generative AI

Looking ahead, Teevan said she expects generative AI to change online search. Instead of simply aggregating webpages, she said, a search engine could provide personalized feedback and serve a larger function. “In many ways, search is just a part of the larger task that you’re doing, and you’re seeing people go to search engines to do those larger tasks, to think things through, to get feedback, to create new content,” she said.

And it’s only a matter of time before models can turn the tables and ask users questions, Teevan said. “That ability to pull more knowledge from people and figure out what’s in your head, to help you bring more to the table, feels like a real opportunity,” she said.

A key challenge to using AI is a lack of imagination, she said — it’s hard to envision all the possibilities, given that the technology is evolving and people are still learning how to use it. “It’s a real sociotechnical problem, actually, where the way people are using the technology is going to shape the way that it evolves, and the technology is going to shape the way that we use it,” Teevan said.

Watch all the sessions from the 2024 MIT AI & ML Conference 

For more info Sara Brown Senior News Editor and Writer