Could AI get its hands on robots, nuclear weapons or a bioweapon? Why would it want to wipe us out in the first place? And what do the people sounding the alarm have to gain? Harlan and Matt separate the movie plots from the risks worth worrying about today.

They also cover: What happened when their own AI agents found ways around the guardrails they set, why cybersecurity basics matter more than ever, and why Harlan chooses optimism anyway. Next episode: the show’s first guest, a leader from KnowBe4’s agentic courseware team.

Quotes

“The instant I can buy one, it could end us all, but it’ll do my dishes every night. I mean, the risk is worth it, Matt.” - Harlan

“What’s reality right now is what’s at risk is the integrity of data and zero-day vulnerabilities and systems that are exposed to the internet.” - Matt

“Every time there’s an incident with a self-driving car that goes off lane and causes an accident, it’s a big news story. You stop hearing about human error, typical driving mistakes and car accidents that happen every day in every city across the world.” - Matt

“If you control its agency, you control what it’s able to actually do in your environment, you’ve mitigated most of the concerns. And then I would just say, don’t give it the keys to the robots, y’all.” - Harlan

“Nobody really knows exactly how the next three years are gonna unfold, and anybody that pretends that they do, you should be skeptical of… Don’t let it ruin your outlook on life. Don’t let it take away your joy.” - Harlan

Time Stamps

[00:56] Welcome to the We're All Gonna Die Episode

[01:19] Harlan's AI Disaster Movie Pitch: Training Data

[04:58] Means: What Would AI Need to End Humanity?

[05:50] Robots, Nukes and Bioweapons

[16:00] The Real Risk Today: Data and Zero-Day Vulnerabilities

[17:18] Motive: Why Would AI Want to Wipe Us Out?

[20:19] How Seriously Should We Take the AI Doomsayers?

[29:46] The Bigger Threat: Humans Using Models Without Guardrails

[32:05] Opportunity: How Would AI Actually Do Damage?

[33:37] When AI Agents Work Around Their Guardrails

[38:42] Control Its Agency, and Don't Give It the Keys to the Robots

[41:01] Choosing Optimism

Links

Connect with the hosts on LinkedIn!

• Harlan Parrott
• Matt Duren

Learn more about:

• KnowBe4

Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

How much power do data centers really use? Does the viral claim about ChatGPT and water hold up? Harlan and Matt work through the facts and find that what’s true globally doesn’t always hold locally.

They also cover: Why a data center can pass its noise limits and still be hard to live next to, how one Virginia county turned its data center boom into lower property taxes, and where to take your concerns if one is planned near you.

Quotes

“Not everything needs to be AI. If you have a simple question, like ‘What time is it?’ or ‘What are some restaurants near me?’, there are algorithmic services that have been built for that.” - Matt

“It's almost like if you're trying to lose weight and you sit down and you've got on your plate a big old steak, and then you got some vegetables, and then you say, ‘Well, I'm just gonna cut the amount of vegetables in half so that I can lose some weight.’” - Harlan

“If you're out advocating for curtailing data center builds, but you get mad whenever you're only allowed to water your grass twice a week, I don't want to hear much from you.” - Harlan

“Even compared to household water usage, it's low. However, you can't replenish a watershed with a national average.” - Matt

Time Stamps

[01:23] Welcome to Attempted Intelligence: A Friendly Debate

[02:17] Ground Rules: Assigned Sides, Not Personal Views

[03:14] The Coin Flip: Matt Argues Against, Harlan Argues For

[03:49] Global vs. Local: The Lens for Every Round

[05:24] Electricity: How Much of the World's Power Goes to Data Centers?

[07:17] Growth Rates and Ireland's 23%

[08:27] Not Everything Needs to Be AI

[10:40] When Local Power Rates Go Up

[11:14] Rising Projections and Why Cheaper Inference Means More Inference

[13:23] The Economic Upside and Bringing New Power Supply

[16:54] The Slippery Slope and Labor Union Support

[18:43] How to Get Involved: Local, State and National

[22:37] Water: Why Do Data Centers Need It?

[23:50] Cooling, Closed Loops and Drinking Water

[27:05] Data Centers vs. Golf Courses and Almonds

[28:33] The Indirect Water Cost of Electricity

[30:07] You Can't Replenish a Watershed with a National Average

[31:54] KnowBe4's Sustainability Goals

[32:51] Noise: Compliant vs. Livable

[35:35] Loudoun County's Property Tax Story

[36:47] Noise Complaints in Sterling and the Problem with Percentages

[39:06] Weigh Both Sides Before You Advocate

Links

Connect with the hosts on LinkedIn!

• Harlan Parrott
• Matt Duren

Learn more about:

• KnowBe4
• KnowBe4 Sustainability

Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

From there, the hosts work through four terms practitioners keep hearing: Agent, Skill, AI Slop and Shadow AI. Shadow AI is today's version of shadow IT, and Harlan tells the story of an Access database he built as a college intern that an insurance company was still running, and still calling him about, years later. Matt explains why personal and unpaid AI accounts can make employees a model provider's tester, and lays out the order KnowBe4 uses for managing AI: find where it exists, enable it where you want it, and block it where you don't. They define AI slop and debate whether anything built by AI counts, which leads Matt to the slide-deck skill he's refined for six months. Harlan explains skills to business teams the way he would coaching a new intern.

They also cover: What makes something an agent? Harlan and Matt work through agency, goal pursuit (would an agent actually buy you the sunglasses, or just send links?), variability, and giving up control of how the work gets done, then land on a working definition together. They close on why definitions matter now that regulators like the EU are writing them into law, why it's worth asking colleagues what they mean when they use these terms, and why the outcome matters more than the label. Next episode: data centers and what they mean for society.

Quotes

“When Microsoft Excel was first introduced, when it was first handed to people, this was a new tool. It was so powerful they didn't know how to use it. And then it's transformed over the last 20 years everything about how we work, and I think that's really what we're looking at now with AI.” - Harlan

“If you're using an unpaid account or a personal account, you very well may be working for that model provider. You're their guinea pig, you're their tester. You're the one whose inputs are being used to shape that next generation of models.” - Matt

“Water will find its course. If you're not providing high-quality AI tools to the people in your org, and finding ways to do that in a safe and secure way, then you're sort of contributing to the problem. Over time, people are gonna find ways to be more productive and efficient in their work.” - Harlan

“Find where AI exists, enable it where you want it, and then block it where you don't. I think that order is really important.” - Matt

Time Stamps

[00:50] Welcome to Attempted Intelligence: Meet Harlan and Matt

[01:11] Two Roles, Two Angles: AI Across the Business and AI in the Product

[02:06] An Introduction to AI for Practitioners

[05:11] Rethinking the Process and Plugging AI In Judiciously

[05:53] Governance: Keeping AI Safe and Making Sure It Delivers Value

[08:29] KnowBe4's AI Journey Started Before ChatGPT

[11:37] The Second Shift: Multi-Step Reasoning and Agency

[17:44] Why AI and Data Sit on the Same Team

[19:20] What is Shadow AI? 

[25:14] Why "Just Block Everything" Doesn't Work

[26:15] What is AI Slop?

[26:52] How Training Produces "the Average of All the Writing"

[29:29] Does Built by AI Mean AI Slop?

[31:37] What Is a Skill?

[33:57] What Is an Agent?

[39:46] Can the End User Tell the Difference between agent and human?

[41:07] Stitching Together a Definition of an Agent

[41:58] Why Definitions Matter: The EU AI Act

[44:50] Next Episode: Data Centers and Society

Links

Connect with the hosts on LinkedIn!

• Harlan Parrott
• Matt Duren

Learn more about:

• KnowBe4

Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Attempted Intelligence takes the AI ideas everyone's talking about, pulls them apart, and asks what they mean for the people using AI at work. Sometimes that's a definition. Sometimes it's a tool you've been told you need. Sometimes it's whether the data center down the road is worth the electricity and water it uses.

Hosts Harlan Parrott, VP of AI Innovation at KnowBe4, and Matt Duren, SVP of Engineering, AI & Data, talk with people applying AI in their own fields about what they built, whether it helped, what broke, and where human judgment still matters. Harlan and Matt won't always agree, but you'll leave each episode with something useful to try, or a better question to ask.

You don't need to know every acronym to follow along. New episodes every week. Brought to you by KnowBe4.

Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.