50. Collaborative Research – AI: Dr Sawsan Khuri on what happens to trust when your team starts using AI

Career developmentCollaborationPsychological safetyTrustLeadershipConflict resolutionAuthorshipTeam alignmentAI
Written By Chris Pahlow

Subscribe to the Amplifying Research podcast:
Apple Podcasts | Spotify | Overcast | Pocketcasts

Learn more about Amplifying Research

Episode show notes

Generative AI has quickly become part of how many researchers work. But in most teams, the conversation about how it's being used, which tools are trusted, and who's checking the outputs simply hasn't happened yet. Dr Sawsan Khuri argues that transparency is key if we want AI to strengthen, rather than fracture, our research collaborations. She talks us through practical tips for bringing AI into your team's work openly.

This is the final episode in our three-part miniseries based on the book How to Succeed at Collaborative Research, co-authored by Sawsan, Howard Gadlin, and Dr L. Michelle Bennett.

All three authors join the conversation — Michelle and Howard open the episode by sharing where they each sit with AI right now, before Sawsan takes the deep dive on what it means for collaborative research. The series wraps up with each author offering a final takeaway from the book.

Sawsan teaches medical genetics at the University of Exeter and runs a consulting company, Collaborative Capacities. She describes herself as an AI practitioner — someone who uses multiple AI platforms daily in her research, navigates the thorny challenge of assessing students when AI-generated answers are everywhere, and works with research teams to build the transparency habits she advocates for in this episode.

In episode 48, Michelle explored the collaborative mindset: the self-awareness, trust, and structured agreements that hold a team together. In episode 49, Howard unpacked what decades of mediating scientific disputes taught him about turning conflict into something productive. This episode extends those same principles — trust, transparency, open conversation — into the fast-moving territory of AI.

"If there are skeptics in that room, you need to also bring them into that task. You're inviting them into the game by saying, 'Would you check it for me?' Be very transparent about what prompts you're using so that people can check you and check the results. That way you've started that two-way communication about let's do this together as a team." — Sawsan Khuri

Whether you're a research leader trying to set a team-wide approach to AI, an operations lead looking for practical frameworks to guide the conversation, or an early-career researcher wondering whether it's OK to admit you've been using ChatGPT — this episode offers a concrete, non-judgmental path forward.

Our conversation covers:

  • How researchers are using generative AI right now: literature search, information gathering, and beyond

  • The three layers of transparency: what you're using AI for, which tool and why, and how you're checking the outputs

  • Trust at three levels — trusting yourself with prompts, trusting the tool, and earning your team's trust

  • The counterintuitive way to bring AI skeptics on board

  • Agentic AI: Stanford's virtual lab of AI agents and what it means for interdisciplinary research

  • Vibe coding and the checks and balances it demands

  • Why governance and regulation are coming to academic AI use

  • Human curiosity and integrative thinking

  • How to start the AI conversation if your team hasn't had one yet

  • A cautionary tale: what happened when grad students used AI to write code without telling anyone

  • The series wrap-up: each author's final takeaway from How to Succeed at Collaborative Research

Find Sawsan online:

Find Howard online:

Find Michelle online:

Things mentioned:

Explore more episodes
Podcast
48. Collaborative Research – Mindset: Dr L. Michelle Bennett on building the collaborative mindset researchers are almost never taught
48. Collaborative Research – Mindset: Dr L. Michelle Bennett on building the collaborative mindset researchers are almost never taught

Most research teams believe they collaborate well — until the first real disagreement, the first authorship dispute, or the first time someone bites their tongue instead of speaking up. Researchers are rigorously trained in methodology but given almost no preparation for the relational skills that hold a collaboration together — and when tension rises, most default to a self-protective mode that quietly erodes trust, psychological safety, and ultimately the science itself. In this episode, Dr L. Michelle Bennett unpacks what a genuinely collaborative mindset looks like in practice, and how something as concrete as a collaboration agreement can give your team the structure it needs to do its best work.

Read more →
46. Research Translation: Dr Jaelea Skehan on why proving something works is just the start
46. Research Translation: Dr Jaelea Skehan on why proving something works is just the start

You've done the research. You've run the trial. You've published the paper. So why isn't anything changing? Dr Jaelea Skehan has spent more than 25 years translating research into real-world programs in mental health and suicide prevention — and she's seen firsthand why so many evidence-based innovations never make it past the journal. In this episode, she makes a compelling case that proving something works is just the beginning, and shares hard-won lessons from programs spanning 18 months to 25 years on what it actually takes to get research into practice — and keep it there.

Read more →
44. Impact Literacy: A/Prof Julie Bayley on why just expecting impact isn't the same as enabling it
44. Impact Literacy: A/Prof Julie Bayley on why just expecting impact isn't the same as enabling it

A/Prof Julie Bayley is one of the world's leading voices on research impact, and she's on a mission to make sure that the pathway from academic inquiry to meaningful societal change isn't just left to chance. She joins us to unpack impact literacy — a practical framework and step-by-step workbook that helps researchers find their place in the impact puzzle, and helps institutions build the culture to make it all possible.

Read more →
42. Gathering with Purpose: Dr Sarah McLusky on making events actually worth showing up for
42. Gathering with Purpose: Dr Sarah McLusky on making events actually worth showing up for

Be honest: how many meetings, workshops, or conferences have you attended that felt like a waste of your time? Dr Sarah McClusky argues that most academic gatherings fail not because of bad content, but because no one stopped to ask why they were bringing people together in the first place.

Read more →

Credits:

  • Host & Producer: Chris Pahlow

  • Edited by: Laura Carolina Corrigan

  • Music by: La Boucle and Blue Steel, courtesy of Epidemic Sound

Explore more episodes

Practical tips:

Be transparent about three things — and make the third one count:

  • The first two layers of transparency are straightforward: what you're using AI for, and which tool you've chosen. But it's the third layer — how you're checking the outputs — that changes the dynamic. When you can articulate your checking process, you shift the conversation from "should we use AI?" to "how do we use it well?"

  • "Everyone needs to be absolutely clear about what they are using AI for, which AI they are using, and why they chose that tool over another. And then there's another layer of transparency, a very important third layer, which is how you're checking the answers. Because any skeptics in the room, that's what they want to hear about." — Sawsan Khuri

Bring skeptics into the process rather than trying to persuade them:

  • When someone on your team resists AI, the temptation is to argue your case. Sawsan's approach is the opposite — invite them to check your work, verify your outputs, even suggest which tools to try. The resistance often softens when skeptics are given a role rather than a lecture. You might not end up using AI for everything, but you've kept the team together.

  • "Try not to persuade people that your way is correct, whether it is for or against or in the middle of AI. Try to listen to the others and find somewhere in the middle that you will all agree about." — Sawsan Khuri

Test the same prompt across multiple platforms:

  • Different AI tools have different strengths and blind spots. Running the same prompt through two or more platforms helps you calibrate which tools you trust for which tasks — and gives you a richer dataset to work from. If two platforms agree, your confidence grows. If they diverge, you know where to dig deeper.

  • "I use three or four different tools. I ask them all exactly the same question. I get different answers, and then I'm using my judgment to figure out which one to go with." — Sawsan Khuri

Build your prompting skills through iteration:

  • Prompt quality is the single biggest determinant of output quality — and it's a skill that develops with practice. Don't settle for your first attempt. Try different phrasings, different levels of specificity, and compare what comes back.

  • "Try writing one prompt, try asking the same thing in a different way, and see what you're getting back, just to give you that confidence that you've finally got the prompt that will give you the answer that you're seeking. Remember to always ask for the references, the citations, so that it gives them to you and you can refer back to them." — Sawsan Khuri

If you're vibe coding, get the code checked:

  • Vibe coding — using AI to write code from plain English prompts — has been a game changer for researchers who don't code. But the convenience comes with a catch: if you didn't write the code, you may not spot its vulnerabilities. Before anything goes public, have someone who understands code review it properly.

  • "If you're producing something that is for the public using vibe coding, you need to get a software engineer, a computer scientist to check the code for you. You need to run it on multiple systems to make sure that it runs as you expect it to run, and you need to test it with the widest possible variety of data to make sure it's giving you what it should be giving you." — Sawsan Khuri

Put AI on your next team meeting agenda:

  • If your team hasn't had an explicit conversation about AI use, the default isn't "no one's using it" — it's "people are using it without talking about it." Journals are already requiring disclosure. The conversation needs to happen, and someone needs to be the one to put it on the agenda.

  • "If they're not talking about it, they should be. I would urge team members to put it on the agenda for the next team meeting, or even have a specific team meeting about the use of AI. 'Hey, there's this tool. Anyone using it? Let's be transparent. What do we think? What do we as a team, what's our policy about this?'" — Sawsan Khuri

Save your prompts — not just the outputs:

  • It's easy to copy an AI's output and move on, but the prompt that generated it is equally important. Documenting what you asked — and how you asked it — creates a trail that colleagues can review, replicate, and build on. Some platforms make it easy to save outputs but not prompts, so you may need to do this manually.

  • "Document what you're doing, save your prompts, because sometimes some of the AIs that I'm using, you can copy paste the output, but it doesn't copy paste the prompt. You need to be saving your prompts. You need to know what you're asking it, and then checking the answers against some other metric. Don't just believe it." — Sawsan Khuri

Chris Pahlow
Chris Pahlow is an independent writer/director currently in post-production on his debut feature film PLAY IT SAFE. Chris has been fascinated with storytelling since he first earned his pen license and he’s spent the last ten years bringing stories to life through music videos, documentaries, and short films.
http://www.chrispahlow.com
Next

49: Collaborative Research – Managing Conflict: Howard Gadlin on what decades of mediating scientific disputes taught him about productive conflict