Gennaro Brooks-Church Outlines 5 Reasons Underwriters Trust the AI He Builds

00:00:00: Welcome to your deep dive.

00:00:02: Today we are looking at a stack of notes and research on AI trust specifically, you know zooming in on the insurance industry as our ultimate stress test.

00:00:10: Yeah It really is the perfect case.

00:00:12: The central mission Of these sources Is figuring out how to get deeply skeptical highly trained professionals To actually Trust artificial intelligence Especially when the stakes Are massive.

00:00:23: Right.

00:00:24: And for this We're Looking At the work of Jannara Brooks Church.

00:00:26: He's the founder of Kazmir & Books Energy to help insurers, brokers and managing general agents navigate risk.

00:00:35: Which is fascinating because most AI tools right now just feel like a magic black box.

00:00:41: Oh totally!

00:00:42: You feed them data they hum away then spit out an answer And kind of expect your blind trust

00:00:48: which professionals hate.

00:00:50: But Gennaro takes the complete opposite approach.

00:00:52: His AI acts more like a hyper cautious intern who meticulously cites every single source.

00:00:59: I love that analogy.

00:01:00: To get a really risk-averse underwriter to trust that intern, you can't just tell them hey the AI is smart.

00:01:05: No absolutely not because I mean A single error on a policy limit could mean a multi million dollar payout mistake.

00:01:12: Yeah The liability's just terrifying.

00:01:15: yeah no kidding.

00:01:17: So Gennaro solves this by literally mapping the ai's logic directly onto the source document.

00:01:22: Right it's the whole.

00:01:23: show your work principle.

00:01:24: Every single data point the AI extracts is linked back to the exact page, the exact paragraph it pulled from.

00:01:31: Wow, okay.

00:01:32: So it essentially hands you a highlighted copy of the original PDF right alongside its extraction?

00:01:37: Yeah exactly.

00:01:38: and The psychological shift for the user there is massive.

00:01:42: Underwriters never actually have to take this system's word for

00:01:45: because they can instantly verify everything themselves.

00:01:47: Right

00:01:48: the cognitive load shifts from wondering if the machine messed up Simply checking the machines receipts.

00:01:53: It transforms what could be a potential liability into a totally verifiable tool.

00:01:58: Okay But I do have to push back on that a little bit.

00:02:00: writer is constantly checking the AI's receipts.

00:02:04: Doesn't that make the tool incredibly high maintenance?

00:02:06: At first, yeah it really does

00:02:08: because at a certain point It feels like they are just doing the work themselves right.

00:02:12: Just auditing a slow intern.

00:02:14: That's a super fair point.

00:02:15: and the sources acknowledge that The initial onboarding absolutely as a heavy lift.

00:02:20: And it introduces some friction.

00:02:22: Right

00:02:22: But that Friction Is A Necessary Investment For Long-Term Accuracy.

00:02:27: When A User Corrects The AI Say you've misidentified a really complex umbrella policy limit, that correction isn't just one-off fix.

00:02:35: Oh it remembers

00:02:35: it!

00:02:35: Yes It feeds directly back into the model's training weights.

00:02:40: The system adjusts its parameters so doesn't make exact same contextual mistake twice.

00:02:46: So it is actively learning specific preferences of users.

00:02:49: Exactly...the

00:02:50: system was built with industry professionals So the onboarding configures AI to specific broker formats and internal processes.

00:02:57: It doesn't use some rigid, generic template...

00:02:59: Which is so smart because machine is adapting to humans' judgment rather than forcing them how they work!

00:03:06: And this highly tailored Learn As You Go setup really reveals Genaro's broader philosophy on workplace technology.

00:03:13: He designed this system to handle the brutal repetitive tasks.

00:03:17: Like what specifically?

00:03:18: Oh, like extracting property ages or loss run histories from a one hundred page commercial lease PDF.

00:03:25: Yeah nobody wants to do that and it means you aren't paying highly skilled domain experts to act as glorified data entry clerks.

00:03:32: Exactly!

00:03:33: The system clears the deck.

00:03:35: So underwriters can focus on the weird edge cases that actually require real human intuition and experience.

00:03:41: Because the AI is explicitly designed to never make final decisions?

00:03:45: Never!

00:03:45: Yeah, it earns confidence over time specifically because of respects the sheer complexity of underwriting.

00:03:51: The whole objective is protect an elevate human expertise not replace It.

00:03:55: so whether you are in insurance or completely outside of it the core takeaway from all these sources Is pretty clear Real trust in AI doesn't come From building a sleeker magic black box.

00:04:04: no Definitely not.

00:04:06: It comes from building technology that transparently aligns with your real-world workflows and actively supports human judgment.

00:04:12: Yeah, we are definitely moving past the era of just wanting an oracle That gives us the answer.

00:04:18: The future belongs to tools that show their work Handle the tedious data processing And you know leave the critical thinking To Us Absolutely.

00:04:26: Which leaves You With a Final Question If AI successfully steps in and strips away all those tedious repetitive data tasks, what deeply human irreplaceable skills will suddenly become your most valuable professional currency?