The Ghost Note: Finding Gold in the AI Gap
Ever wondered why your shiny new AI tools aren't getting the love they deserve? Well, we’re diving headfirst into that conundrum today!
We chat about how companies are shelling out big bucks for top-notch AI, only to find their employees avoiding it like the plague. It’s not the tech that's broken — it's the way companies are rolling it out, and the psychological factors at play are a big part of the problem.
Drawing from Heather Masters’ fascinating insights, we explore how the AI learns from our corporate culture and communication patterns, often mirroring our worst habits.
So, if you're curious about why your AI might be acting more like your passive-aggressive coworker than a helpful assistant, stick around!
This Deep Dive podcast is AI generated from the Start With AI Newsletter on LinkedIn - linkedin.com/newsletter/start-with-ai
Chapters:
- 00:00 - Investing in Advanced AI Technology
- 04:51 - The Psychological Safety of AI Adoption
- 09:43 - The Impact of Corporate Language on AI Communication
- 14:06 - The Ghost Note of Corporate AI
- 19:14 - Understanding the Role of AI in Organizational Culture
Takeaways:
- Organisations often invest millions in AI tools, but user adoption remains a huge challenge.
- The AI technology itself isn't broken; it's the corporate rollout that fails spectacularly.
- AI tools absorb the corporate culture, reflecting the good, the bad, and the ugly communication habits.
- Psychological safety is crucial for effective AI integration, as employees avoid risk-averse communication styles.
Links referenced in this episode:
Companies mentioned in this episode:
- Microsoft
- Kajabi
Transcript
Imagine your company just spent, I mean, literally millions of dollars on the absolute most advanced, cutting edge AI tools on the market.
Speaker B:Right.
Speaker B:The top tier enterprise software.
Speaker A:Exactly.
Speaker A:The stuff that's designed to make everyone faster, smarter, just infinitely more productive.
Speaker B:Yeah, that's always the pitch.
Speaker A:But then months later, nobody actually wants to use it.
Speaker A:Or even worse, when they do finally use it, the AI starts sounding exactly like that one coworker everyone actively avoids.
Speaker B:Oh, we all know.
Speaker B:Coworker.
Speaker A:Right, the person who just strings together 50 corporate buzzwords just to avoid making an actual decision.
Speaker B:Yeah, the master of saying absolutely nothing.
Speaker A:Exactly.
Speaker A:So we're looking at this fascinating dynamic today where the technology itself isn't broken, but the corporate rollout is just completely backfiring.
Speaker A:And the reason why?
Speaker A:Well, it's hiding in our own psychology.
Speaker B:It really is a massive blind spot for leadership right now.
Speaker B:I mean, they're deploying these deeply integrated systems.
Speaker B:Assuming it's just like, you know, upgrading to a new version of Excel.
Speaker A:Right, just a software update.
Speaker B:Exactly.
Speaker B:But they're finding out the hard way that an AI doesn't just calculate data.
Speaker B:It actually absorbs the entire cultural operating system of the people using it.
Speaker A:Which is wild to think about.
Speaker B:Yeah.
Speaker A:And to figure out how this is happening, we're pulling from this really compelling piece of research by Heather Masters.
Speaker B:Yeah, her perspective on this is incredibly unique.
Speaker A:It really is.
Speaker A:On one hand, she spent 25 years as a practitioner of Neuro Linguistic Programming, or NLP, analyzing how human beings construct meaning through language.
Speaker B:Which is super relevant here.
Speaker A:That's right.
Speaker A:And on the other hand, she just spent a solid year deep inside Microsoft Copilot just pushing the absolute limits of enterprise AI.
Speaker A:Okay, let's unpack this, because I want to establish right out of the gate that Heather isn't just observing this from some academic ivory tower.
Speaker B:Oh, not at all.
Speaker B:She definitely has skin in the game.
Speaker B:I mean, she's been personally overhauling her entire business infrastructure using these exact tools,.
Speaker A:But actually building it herself.
Speaker B:Yeah.
Speaker B:Over this grueling 90 day sprint, she's integrating anthropics models, connecting them to Google Workspace, and wiring all of that into.
Speaker A:Kajabi, which is the platform she uses for her digital courses in community.
Speaker B:Exactly.
Speaker B:So she's manually building the workflows, dealing with all the API breaks, and seeing firsthand exactly where the logic falls apart.
Speaker A:So she's essentially her own test subject.
Speaker B:Precisely by doing that, she's experiencing the exact friction points that are currently, you know, completely paralyzing Fortune 500 companies.
Speaker A:Which brings us to the Core mystery of this deep dive.
Speaker A:If someone like Heather, who intimately understands human language patterns, is finding immense utility by carefully structuring her AI, why is the broader corporate world just crashing and burning?
Speaker B:Well, the data on this is honestly brutal.
Speaker B: look at the numbers from the: Speaker A:Oh, wow.
Speaker A:What do the numbers say?
Speaker B:We're looking at 79% of organizations reporting major challenges in adopting AI.
Speaker A:Wait, 79%?
Speaker B:Yeah.
Speaker B:And that's a double digit increase from last year.
Speaker B:So as the models are getting objectively smarter and faster, the actual human integration is getting worse.
Speaker A:That makes no sense.
Speaker A:The tech gets better, but the adoption gets worse.
Speaker B:Right, but the most alarming statistic is how executives are viewing this.
Speaker B:54% Of C suite leaders literally stated that AI adoption is, quote, tearing their company apart.
Speaker A:Tearing the company apart.
Speaker A:I mean, just think about the gravity of that phrasing.
Speaker B:It's intense.
Speaker A:You don't hear a CEO say, you know, deploying this new cloud storage provider is really tearing the fabric of our culture to shreds.
Speaker A:Software shouldn't be able to do that.
Speaker B:No, it really shouldn't.
Speaker A:But there's this massive behavioral contradiction that Heather highlights, and I think it points to the root of the problem.
Speaker A:When corporate employees are given access to both Microsoft 365 Copilot and standard ChatGPT, 76% of them voluntarily choose to use ChatGPT.
Speaker B:76%?
Speaker A:Yeah.
Speaker A:Only 18% are choosing Copilot even after their companies spend millions custom building it for them.
Speaker A:It's like management buying everyone a top of the line corporate smartphone.
Speaker A:But 76% of the staff are just hiding in the break room and using their personal phones instead.
Speaker B:That's a great analogy, and it completely baffles the consulting firms.
Speaker A:Why the resistance to the in house tool though?
Speaker B:What's fascinating here is that at the World Economic Forum in Davos, the conversation was entirely focused on traditional metrics.
Speaker B:You know, deployment speed, latency, tool selection.
Speaker A:The technical stuff.
Speaker B:Exactly.
Speaker B:The big consulting agencies are blaming a lack of IT governance or poor change management.
Speaker B:They think the solution is just forcing everyone to take a prompt engineering workshop.
Speaker A:Yeah.
Speaker A:I have a theory on why employees are rejecting the in house tool.
Speaker A:And it has absolutely nothing to do with prompt engineering.
Speaker B:Oh yeah?
Speaker B:Let's hear it.
Speaker A:I think it's about psychological safety.
Speaker A:ChatGPT feels like a burner phone.
Speaker A:It sits outside the corporate firewall.
Speaker B:Right.
Speaker B:It's completely separate.
Speaker A:Exactly.
Speaker A:It doesn't know who my boss is.
Speaker A:It doesn't Know my performance reviews coming up, and it definitely doesn't know about that passive aggressive email chain I was just copied on.
Speaker B:That is a very real distinction.
Speaker A:But Copilot, on the other hand, lives entirely inside the corporate ecosystem.
Speaker A:It feels claustrophobic.
Speaker A:It's like having a hyper observant supervisor looking over your shoulder while you draft a mundane email.
Speaker B:Well, your instinct is exactly right.
Speaker B:And it leads directly into the underlying Mechanics of how Microsoft 365 Copilot actually functions.
Speaker A:It's not just in our heads.
Speaker B:No, not at all.
Speaker B:To understand the fear, we have to look at the architecture.
Speaker B:ChatGPT is a generalized model trained on the public Internet.
Speaker B:It's broad, but it's fundamentally disconnected from your daily life.
Speaker B:Right, but Copilot is deeply integrated into your company's Microsoft graph.
Speaker B:That means it sits inside Word, Outlook, Teams, SharePoint, PowerPoint, Excel.
Speaker B:It's everywhere.
Speaker A:So it's not just a generic chatbot sitting in a browser window.
Speaker B:Yeah, exactly.
Speaker B:It's an intelligence layer that is constantly reading the internal nervous system of your business to build something called a semantic index.
Speaker A:A semantic index.
Speaker A:Okay, mechanically, what is that actually doing?
Speaker A:If I ask it to summarize a project, how does it know what information matters more than something else?
Speaker B:It uses node and edge mapping.
Speaker B:So it doesn't just keyword search your files.
Speaker B:It maps the mathematical relationships between people, documents and activities.
Speaker A:Wait, really?
Speaker A:It maps relationships?
Speaker B:Oh, absolutely.
Speaker B:Let's say you frequently email the cfo and you almost always open the attachments they send, but you rarely click on the weekly HR newsletter.
Speaker A:Guilty as charged.
Speaker B:Right.
Speaker B:While the semantic index mathematically weights the CFO's communication style, their priorities, and their vocabulary much higher in your personal index.
Speaker A:Oh, wow.
Speaker B:Over time, it learns the operational hierarchy, the unwritten rules, and the specific dialect of your organization.
Speaker B:It reads the messy draft documents, the quick teams messages and the formal memos, and it synthesizes the prevailing corporate culture.
Speaker A:Okay, and just to clarify for everyone listening, Microsoft's enterprise contracts lock this data down.
Speaker B:Yes, absolutely.
Speaker A:Your company's internal messy drafts aren't leaking out.
Speaker A:To train the public ChatGPT model, the semantic index is a closed loop securely inside your organization's boundary.
Speaker B:Yes, the data is walled off, but that security is actually a double edged sword.
Speaker A:How so?
Speaker B:Because the AI is restricted entirely to your internal data, it's deprived of outside neutrality.
Speaker B:It's getting progressively more fluent in the specific, unique, and often deeply flawed way your exact organization communicates.
Speaker A:Here's where it gets really interesting.
Speaker A:Because if the AI is indexing every single Interaction.
Speaker A:Every passive aggressive email, every awkward teens chat.
Speaker A:It's essentially mapping our psychology.
Speaker B:Exactly.
Speaker B:It's mapping the mind of the company.
Speaker A:And that perfectly bridges into Heather's background in neuro linguistic programming.
Speaker A:We're not talking about computer science NLP here.
Speaker A:Right.
Speaker A:We're talking about the psychological framework.
Speaker B:Right.
Speaker B:Psychological NLP.
Speaker B:It has spent the last 50 years studying how human beings take their complex messy internal experiences and filter them into language.
Speaker A:Because we don't just speak objectively, never.
Speaker B:Every time we write an email or speak in a meeting, our language is governed by our underlying beliefs, our fears and our corporate survival instincts.
Speaker A:Like trying not to get fired.
Speaker B:Exactly.
Speaker B:We delete information that feels risky or we distort facts to protect our egos.
Speaker B:And we generalize based on past experiences.
Speaker B:Language is really just the exhaust fume of human psychology.
Speaker A:So bring that back to the semantic index.
Speaker A:A large language model like Copilot works by predicting the next most probable word or token based on its training data.
Speaker B:Right.
Speaker B:It's just math at the end of the day.
Speaker A:So if its primary training data is our company's internal communications, it's going to mathematically calculate our psychological kirks as the optimal most probable way to communicate.
Speaker B:That's the mechanism perfectly described.
Speaker B:The AI doesn't know what truth or courage is.
Speaker B:It only knows probability.
Speaker A:Which is terrifying.
Speaker B:It is, Heather points out, a widespread corporate behavior she calls political hedging.
Speaker B:In a lot of companies, psychological safety is practically non existent.
Speaker B:People are terrified of being blamed for a mistake.
Speaker A:Right.
Speaker A:Nobody wants to be left holding the bag.
Speaker B:So what do they do?
Speaker B:They write emails in the passive voice.
Speaker B:They use incredibly vague buzzwords.
Speaker B:They loop in 14 people on a decision so nobody is individually respons seen.
Speaker A:That mistakes were made.
Speaker A:Synergies will be leveraged going forward.
Speaker A:Exactly the kind of sentence that means absolutely nothing but keeps you out of trouble with hr.
Speaker B:Exactly.
Speaker B:Now imagine a semantic index digesting five years of that specific type of communication.
Speaker B:The algorithm registers that vague passive language as the highest probability communication style for your company.
Speaker A:So thinks that's how it's supposed to talk.
Speaker B:Yes.
Speaker B:So when an executive asks copilot to draft a strategic roadmap, the AI calculates the tokens and spits out a document filled with studied vagueness.
Speaker A:It summarizes meetings with careful omissions.
Speaker B:Right.
Speaker B:Ensuring nobody is pinned down to a deliverable because that's the pattern it learned from the humans.
Speaker B:It actively scales the avoidance.
Speaker A:It's acting like a funhouse mirror.
Speaker A:I mean, if your company culture is already a bit distorted or Bureaucratic or conflict avoidant.
Speaker A:The AI doesn't act as an objective consultant.
Speaker A:That cuts through the noise.
Speaker B:No, it doesn't fix it.
Speaker A:It literally holds up a mirror and reflects your exact neuroses back at you.
Speaker A:Just automated and 10 times faster.
Speaker B:And that's exactly why executives feel like it's tearing the company apart.
Speaker B:It's exposing the reality of their culture in high definition.
Speaker B:And they're blaming the software instead of themselves.
Speaker A:They're fighting their own reflection.
Speaker B:They really are.
Speaker B:But we have to recognize that this mirroring effect works in both directions.
Speaker A:Meaning it can be positive, too.
Speaker B:Exactly.
Speaker B:Heather illustrates this with an example of a board of trustees she analyzed.
Speaker B:In this particular group, the human inputs were exceptionally healthy.
Speaker A:Okay?
Speaker B:They had a culture of clear, direct communication.
Speaker B:They structured their thoughts logically before typing.
Speaker B:They documented decisions with clear accountability.
Speaker A:Because they brought a high degree of structural integrity to their human interactions.
Speaker A:The data feeding the AI was actually clean.
Speaker B:Exactly.
Speaker B:The semantic index mapped a culture of clarity.
Speaker B:As a result, the AI's token predictions were highly functional.
Speaker A:Oh, that makes sense.
Speaker B:It summarized complex board packets accurately.
Speaker B:It drafted policies that were decisive and actionable.
Speaker B:The AI became an incredibly powerful asset because the humans were bringing their best structural selves to the keyboard.
Speaker A:So the technology learns excellence just as efficiently as it learns dysfunction.
Speaker B:Exactly.
Speaker B:It's just a mirror.
Speaker A:Okay, I can see how a company could slowly drift into a state where their AI is just echoing their worst habits.
Speaker A:But why is Heather ringing the alarm bell so loudly right now?
Speaker A:I mean, why is the timeline so urgent?
Speaker B:The urgency comes from a major foundational shift happening in Microsoft's architecture.
Speaker A:What's changing?
Speaker B:Up until now, Copilot has been running on iterations of the standard GPT models.
Speaker B: But in August: Speaker B:It's known internally as Polaris.
Speaker B:And this is going to be deployed across every Copilot seat globally.
Speaker A:I have to push back a little here, though, because I've spent some time looking at Microsoft's admin centers.
Speaker A:They have a feature literally called copilot tuning.
Speaker A:So if I'm the IT director and I realize the AI is acting like a passive aggressive middle manager, can't I just go into the settings, adjust the tuning dials, and explicitly prompt the system to be concise, direct, and objective?
Speaker B:It sounds like an easy fix, right?
Speaker A:Why do I need to worry about a new model if I have admin controls?
Speaker B:This raises an important question about the illusion of control in enterprise AI.
Speaker B:Yes, Copilot tuning exists.
Speaker B:It gives You a surface layer, explicit control where you can apply systemic prompts.
Speaker A:Like telling it to be concise.
Speaker B:Exactly.
Speaker B:You can tell it to be concise, but you have to understand the difference between a surface prompt and foundational training weights.
Speaker A:Okay, what's the difference?
Speaker B:The semantic index isn't a setting you can toggle off.
Speaker B:It is the foundational map of how your people actually behave.
Speaker B:It's building the relational database of your culture every single second, whether you actively manage it or not.
Speaker A:So the tuning is basically just a band aid over a broken bone.
Speaker B:Precisely.
Speaker B:You can tell the AI to be concise, but if the underlying semantic index is built entirely on a decade of terrified, evasive communication, the AI will just give you concise evasions.
Speaker A:Wow.
Speaker A:Concise evasions.
Speaker B:Yeah.
Speaker B:And when the Polaris upgrade rolls out, you're getting a significantly more sophisticated inference engine.
Speaker B:It's going to be much better at reading the subtle nuances of your data.
Speaker A:So if your foundational data is structurally.
Speaker B:Unsound, a more powerful engine is just going to process that dysfunction faster and embed it deeper into everyday workflows.
Speaker B:The patterns are about to become more consequential, not less.
Speaker A:And this brings us perfectly to the title of Heather's work and the concept of the ghost note.
Speaker B:Yes, the ghost note.
Speaker A:Because in music, you know, a ghost note is a note that has rhythmic value but no discernible pitch.
Speaker A:It's usually played so softly, you don't even consciously register it.
Speaker B:Right.
Speaker A:But it creates the groove.
Speaker A:It completely changes the feel of the entire song, whether you hear it or not.
Speaker B:And that's exactly the dynamic happening inside these AI networks.
Speaker B:The ghost note of corporate AI is the massive invisible gap between what management thinks they are controlling via IT policies and what the AI is actually learning from the daily habitual behavior of the employees.
Speaker A:It's the silent frequency running beneath the official corporate strategy.
Speaker B:Exactly.
Speaker B:And it completely dictates the output.
Speaker A:It's honestly a bit terrifying.
Speaker A:We've essentially built machines that read our minds by reading our group chats, and we're upset that they're acting like us.
Speaker B:It's a hard truth to face.
Speaker A:So we've diagnosed the ghost note.
Speaker A:We know the semantic index is watching, and we know a more powerful model is coming to amplify it all.
Speaker A:How do we actually fix this?
Speaker A:Because Heather's article is titled the Gold in the Gap.
Speaker A:Where is the gold?
Speaker B:The gold is recognizing that this isn't an IT problem at all.
Speaker B:It's an organizational design opportunity.
Speaker A:So we can't just call tech support.
Speaker B:No, the solution isn't writing a better software patch or firing your software vendor.
Speaker B:The fix requires doing the hard, uncomfortable work on your company's identity and values.
Speaker B:It's understanding how your people actually create meaning before the AI is even switched on.
Speaker A:But what does values work actually look like in practice?
Speaker A:Because, I mean, that can sound like a really vague HR platitude.
Speaker A:If I'm a department head, how do I practically change the semantic index?
Speaker B:You change it by treating the AI's output as an objective diagnostic tool for your culture.
Speaker A:Okay, give me example.
Speaker B:Well, if your AI consistently generates 10 page strategy memos that are completely devoid of a clear decision, you don't send your staff to a prompt writing class.
Speaker A:Because that won't fix the underlying issue.
Speaker B:Right?
Speaker B:You sit your leadership team down and ask, why does our culture reward 10 page memos?
Speaker B:You look at how performance reviews are conducted.
Speaker B:Are you penalizing people who speak directly and make bold, concise recommendations?
Speaker A:Oh, I see.
Speaker B:If your employees are using studied vagueness to survive in your company, you have to fix the survival metrics.
Speaker A:That makes total sense.
Speaker B:When you start rewarding clarity and psychological safety in the physical conference room, the human inputs change.
Speaker B:When the human inputs change, the semantic index shifts.
Speaker B:And that's when the AI finally starts delivering the massive ROI you paid for.
Speaker A:So what does this all mean for you, the listener?
Speaker A:I want you to mentally walk through the last few days at your job.
Speaker A:Think about the meetings you sat through.
Speaker A:Look at the last five emails you sent.
Speaker B:It's a very revealing exercise.
Speaker A:It really is.
Speaker A:Were you communicating genuinely and directly?
Speaker A:Or were you engaging in performative communication?
Speaker A:Adding unnecessary people to CC lines, Using jargon to pad out a thin idea, Hedging your bet so you couldn't be blamed if things went wrong?
Speaker B:People do it all the time without realizing it.
Speaker A:Because whatever behavior you are bringing to the keyboard is exactly what you were training your organizational AI to replicate.
Speaker A:Your personal habits are becoming the corporate algorithm.
Speaker B:If we connect this to the bigger picture, Heather's ultimate thesis is a massive wake up call for the modern enterprise.
Speaker A:Yeah.
Speaker B:We've spent decades trying to optimize human beings to act more like machines.
Speaker B:Efficient, standardized and predictable.
Speaker A:And now we've flipped it.
Speaker B:Exactly.
Speaker B:Now we've deployed machines that are learning to act like human beings.
Speaker A:Yeah.
Speaker B:If we want them to be highly functioning tools, we have to become highly functioning humans first.
Speaker B:The gold isn't in the software.
Speaker B:The gold is sitting unrecovered in that gap between human behavior and AI output.
Speaker B:You have to learn how to read what the AI's output is actually telling you about your own human workforce.
Speaker A:It's the ultimate reality check.
Speaker A:Let's just briefly trace the path we've taken today.
Speaker A:Because it completely reframes the technology sitting on our desks.
Speaker B:It really changes everything.
Speaker A:We started by wondering why billion dollar AI investments are causing so much friction and actively tearing companies apart.
Speaker A:We broke down how tools like Copilot use a semantic index to mathematically weight our daily communications.
Speaker A:Learning not just our vocabulary, but our deepest psychological survival tactics.
Speaker B:The ghost notes.
Speaker A:Right?
Speaker A:And we realized that with upgrades like Polaris on the horizon, the only way to improve the artificial intelligence is is to actually fix ourselves and drastically improve our emotional intelligence.
Speaker B:We are training these systems every time we hit send.
Speaker B:They're a perfect reflection of our collective organizational health.
Speaker A:It leaves you with such a heavy but vital perspective.
Speaker A:We've spent all this time treating corporate AI like a high tech assistant, right?
Speaker A:Giving it tasks and hoping it boosts productivity so we can all go home a little earlier.
Speaker B:That was the dream.
Speaker A:But what if we've completely misunderstood its primary function?
Speaker A:What if right now, Microsoft Copilot isn't acting as an assistant at all, but rather as the ultimate unfiltered lie detector for your company's culture?
Speaker B:That's a profound way to look at it.
Speaker A:Think about that for a second.
Speaker A:If the AI compiled a brutally honest psychological evaluation of your entire team based entirely on the hidden ghost notes, the passive aggression, and the hedging in your chat logs, would you actually have the courage to read it?
Speaker B:That is the defining challenge for the next era of leadership.
Speaker A:It truly is.
Speaker A:Thank you so much for joining us on this deep dive.
Speaker A:Keep questioning the tools you use, keep an eye on your own ghost notes, and we will catch you next time.
