A Georgetown University faculty member specializing in artificial intelligence (AI) advised working professionals to integrate language models effectively and responsibly in the workplace while maintaining human verification processes during a Sept. 1 webinar.
Lawrence Gray, Ph.D, the program director for the School of Continuing Studies’ (SCS) data analytics and decision-making certificates, instructed teams and workplaces on best practices for AI integration during the webinar titled “The Augmented Team: Transforming How We Work and Collaborate with GenAI.” The event was part of SCS’ “Knowledge@Noon” webinar series, which invites guest speakers to present professional development skills.
Gray said the public often misunderstands how large language models, such as ChatGPT, work.
“ChatGPT is not a search engine,” Gray said at the event. “It’s not a database. It doesn’t memorize answers. And here’s the entire secret to it — it’s one sentence: It guesses the next word.”
Gray said AI tools can help professional teams address three large problems that plague most workplaces: unclear goals, unclear roles and unaddressed conversations that cause conflict.
“So I want you to meet your new coworker,” Gray said. “He’s capable, tireless, never complains about the coffee and, on day one, it knows absolutely nothing about your job. Like every new hire you’ve ever had, the difference between people who get magic out of this tool and the people who get mush is not talent, it’s technique.”
Still, Gray said all workplaces considering integrating AI tools should be aware of its inaccuracies and “delusions.”
“It doesn’t lie in the most obvious, significant way,” Gray said. “It lies in very subtle ways that are hard to detect — that if you’re not a domain expert for something, you tend to skip over this stuff.”
“You have to verify — and we talk about this in our class — the output of these models because they can and will lie, and they will lie confidently,” Gray added.
Earlier this year, Georgetown announced it would offer students licenses to Gemini, Google’s AI tool. In August, University President Eduardo Peñalver shared that he planned to launch an AI initiative aligned with the university’s values. Still, students, faculty and staff remain largely divided over how much Georgetown should engage with AI.
Gray said teams can use AI to ensure their work aligns with company goals, which may be difficult for workers to compare themselves.
“The AI reads both lists, shows you every mismatch in two minutes,” Gray said. “Most teams have never done this once and so this is the exact prompt to do this, and you can take this back to your job and actually look at the alignment of your goals, organizational goals versus team goals, and see where the alignment lies.”
Gray said language models work on temperature scales that determine how accurate or inaccurate their outputs will be — low temperature is consistent and factual, while high temperature is creative and varied.
“Whenever you’re playing around with ChatGPT or any large language model — whether you use Claude, Gemini, Copilot, all of those — you can actually ask it to change the temperature and you could get varied responses,” Gray said.
During the event, Gray gave multiple examples of deception by language models, including ChatGPT inventing a false historical event and a plot to a book that hadn’t been written.
Gray also said that when engaging with language models, especially with prompts about coworkers or conflicts, users should remove identifying information.
“For privacy reasons, before you just paste things, make sure you’re stripping information from it, so you’re just not providing information that could likely be used in the models later to make — they use the information that you type in order to build the latest, greatest models.”
Gray said that no matter how users rely on AI, they should remember models are incapable of empathy when resolving workplace conflicts.
“The honest caveat to all of this, though, is it can simulate empathy,” Gray said. “It cannot feel it. The real conversation is still your problem.”
