PAPod 618 - Poop In...Louder Poop Out?

So you guys are still having quite nice days, but cool nights. Yeah, exactly. The days are nice. And we've had, New Mexico's had a lot of rain, which is really highly unusual. I mean, like a couple inches in the last couple days. So that's, I don't know what that is in centimeters, like a thousand million centimeters, I think.

Did I do the translation correctly? Yeah, you'll get there one day. Of the Pre-Accident Investigation Podcast. I'm Todd. I'm your host. Well, we got another exciting podcast, but before we go one step further, I want to take a moment and thank everybody who reached out from last week's podcast. It was a hard week for me. You could probably tell, but the amount of love and just contacts I got really, really was helpful. Thank you so much.

It means so much to me. It's weird. Grief is kind of a crazy thing. Crazy enough that Anderson Cooper pretty much talks about it on my phone every day or I get some kind of thing. It's a process I'm going through, and it's so helpful to have people that reach out. Thank you so much for that. That means a ton to me.

Today's pod is interesting. Because if you're not watching the news, you must live in some kind of cave, because everything is about AI. And the fact that AI is eventually going to take over and kill us all and there's a 10 percent chance. I mean, there's a lot going on here. And yet in the safety and reliability world, AI is kind of all people talk about.

Like you go to a conference, there's nine million people that have AI stuff that they want to hook you up with. And maybe it's good. Maybe it's not. I mean, I don't know. And I'm really struggling to kind of figure it out. You could probably tell. If you listen to the podcast, you know that it's a topic that I've picked up on a bunch before because I see great potential for it, no doubt.

But I also see a lot of pretty gigantic pitfalls. Like I'm not in love with the flat camera thing. I mean, I don't like that idea. I don't like the idea of surveillance as a way to manage behavior. I mean, I'll grant you that a watched worker behaves differently than a non-watched worker. We know that's true. And there's a whole phenomena around it.

If you're interested, check out the Hawthorne effect. It's a pretty interesting thing to look up because you'll get kind of the history of general electric and plants and industrial safety and production and efficiency. But I'm still not in love with the fact that there's kind of a big brother looking over your shoulder. I don't know how you feel.

And you may feel differently. I just don't. And it's something that's. It's, I've thought about it a bunch for many, many different reasons at many different places in my life. I'm not really happy with the increased amount of surveillance that's happening. It seems like it's squelching our ability to sort of have free will.

And that is always sort of the underlying topic of every podcast that exists. That's what today's about. So take a moment and listen to this pod. It's a discussion with Brent Sutton and Brent Robinson. And they had some help with a co-author named Nicola Noble. And they're going to talk about their new book. And their new book, I think, is kind of worth looking at. And I don't casually recommend books.

I mean, this one's a pretty good read. It's called The AI Paradox in Safety. Blame and punish or learn and improve. Let's go into the conversation now and you tell me what you think by the time we get to the end. Here it is. We're the worst. Okay. Freedom units. Freedom units. I'm sorry. Freedom units. Yes, exactly. Okay. I'm holding a copy of it right here.

The AI Paradox in Safety. Oh, you have one too. That's weird. Yeah. So this is a book by the Brents. Brent Sutton and Brent Robinson. And this book. So first of all, Brent. I mean, and I don't. Usually I say this and don't mean it. But this is really super readable, which you kind of think a book on AI and a book with the term paradox in the title would not be very readable.

I'm just saying it has the potential bones of being either highly technical or super academic. And it's neither. Yes. No, no. But that's probably reflective of the authors. So, yeah, you're right. I mean, you can hear your guys' voice in it for sure. It's just really grabbable. Part of it is that it's definitely laid out for the working safety professional.

So, it's got old man-sized type. So, it's got larger type and it's double-spaced and it's just really readable. Although that's, quite honestly, with this book, we shouldn't waste a lot of time talking about the fact that it's readable and it looks good. What we ought to talk about is what on earth. How are you motivated to tackle this topic? Because the thing that scares me most about AI as a safety person is not the benefits that AI can give us.

And AI is in the news. I don't know in New Zealand if it's like it is in the United States. It's in the news all the time. And then the fact that they're coming out and saying that AI could destroy humanity in the next – 10% chance it would destroy humanity. And that it's getting loose. It's got lots of potential pitfalls for sure.

The biggest one of which, at least from the safety standpoint, is I'm really worried about, you know, how it's used as a surveillance tool. You know, flat cameras, all the weird crap now at all the airports. I mean, the potential for it to sort of surveil a worker is really high. And I'm just not sure there's any benefit to that. Okay, that's the longest question on earth.

What's going on? Tell me why you did this. Yeah, well, a couple of reasons. We've always talked about, you know, when we're working in hot, the whole purpose is to stop doing safety for people. Right. Stop doing safety for people. Start doing safety with people. Yeah. Yeah. I mean, that's exactly the shift.

If you ask me what the biggest shift is, that's the biggest shift, is that how we look at the workforce is changing. Absolutely. So, AI just has that power to amplify that. Yes. So, for those that do want to surveil and monitor, and when I say monitor, they're really basically surveilling to make sure they're doing what the system says.

Yeah. Then AI is simply going to amplify that whole doing safety to people. So, why tackle this? I mean, isn't this kind of a big topic? And I know lots of people said to me, you know, the book will become dated. But the book is not about what AI model you choose. Right. The book is understanding how AI can help us both learn and improve or blame and punish. So, that's a really good point you bring up. But let me emphasize that because you're absolutely right.

First of all, a book on AI is going to become almost immediately dated. I'm not sure that's all bad because I think there'll be a series of books on AI that will come out and progressively each outdate the next. That's how learning happens. But this isn't that book. This is a book really about how really philosophically we think about using additional, computer-based thinking tools to help us understand what's going on.

And it's really interesting because I think the notion of paradox is maybe where I should have started this question. And a paradox is just, for people that aren't sort of familiar with that term. It's a statement or a situation or a person or a thing or an environment, a context that seems to go against common sense, right, or contain sort of counterfactuals, opposite facts, but actually reveals a much deeper truth or deeper meaning.

So the idea of a paradox is that on the surface, it looks like it's in friction, but in reality, it's really a thoughtful idea. Fair enough. Yeah, and the book was trying to get balance. I mean, a front cover showed a Jenga tower being pulled apart. Right, by laser beams. And human beings too, but there's certainly a laser beam pulling a Jenga out.

Absolutely. And that goes back to it, that if you're going to blame and punish, that the effect of the AI is destabilizing the system for workers to fail. Right, right. So that was the whole emphasis behind the graphic. Because... Leaders aren't choosing that pathway wanting to create that type of harm because they're being sold AI. It's that classic paradox.

They're being sold it that's going to help them improve their compliance. Well, and if you've been to a conference, if you've been to a conference in the last year or two, you've definitely been approached by AI people. I mean, it seems like there's a million of them. And they all, and I'm going to say this, and you should correct me if this is wrong, because I'm often wrong on stuff, but it all seems like they're selling the same stuff.

I mean, it's just got a different company name. Correct. Just as there are a lot of consultants out there or software companies, selling things safety based on fear. Yeah. Yeah. Yeah, it's true. I mean, the AI stuff is basically what they're selling is surveillance tools, that keep records and then analyze those records to determine effectiveness or efficiency or whatever it is, the outcome they want to determine.

Our AI will tell you when people are breaking your exclusions on. Yeah, exactly. We can tell you exactly how many people cross the red line on a Tuesday between 2 and 4 p.m. Yeah. And you can go out and you can correct and action those people. And if we see a corresponding dip in exclusion zone breaches, we know the AI is working. Yeah.

And that will, in fact, make you safer. And then eventually you can buy an add-on to it that shoots a laser beam and will disintegrate the worker as soon as they cross the red line. I'm guessing that's where it goes. Yeah, shock collars. Oh, shock collars. I mean, I'm not going to freak anybody's buzz, but if they're used for evil reasons, they're pretty bad. That's for sure.

They've already got tech out there that will buzz your wrist. Yeah. Okay? What's the difference? Yeah, no, there really isn't any difference. That's my big hang-up. I mean, it's the same problem I have with, like, flock cameras. I mean, are we thinking about, are companies thinking about basically putting flock cameras in the worksite?

Yes, they all are, right? They all are, because it comes back to they're managing that sticky so poorly that they're having to rely on the goodwill of people for work to go right. Yeah, to hold the organization and its systems together. You're exactly right. It's that next version of AI-empowered zero harm, or the one that scares me most, Todd, is at the moment, our.

Focus on SIFs, you know, stickies, all this stuff, we're going to move from zero harm to 100% presence. So why are the workers making sure the critical controls are present 100% of the time? Right. Right. So AI is going to be used at both ends. And even in our own industry, it'll be used at both ends.

Well, and you can definitely see, for instance, just as an example, if you aggressively approach like an energy-based safety program, you would look for 100% presence of control. And you'd monitor for that and then punish if they weren't there. I mean, whatever punish means. Whatever punish means. And that may mean, for instance, a name has happened, we've trained you in all this identification using the energy wheel. Right.

And you fail to identify it. Right. Rather than having that deeper question, what are our opportunities to help workers, to cope and deal better with the uncertainty that the organization has created, by not being able to manage energy pathways? So this is a warning, really, and that's what the paradox does.

It sort of emphasizes the benefit and the downside. And the warning is that if used to mete out blame, this could be really negative. And probably not successful and probably will not create the change the organization needs or wants. Fair enough? Absolutely. And in the book, we explore stories across many industries where we show the blame and punish side of AI being used and we show the learn and improve side.

And then at the end of the day, if we talk about workers face lots of uncertainty, we know that our, let's call it measurement of critical control verification is going to be on the lower side. Right. So what we want to see is we want to see higher psychological safety of speaking up and we want to see higher amounts of team-based communications within those work groups.

So if that blame design forces people not to talk and share, then you're not going to get information. And the next thing is around the corner yeah and you say and a better person would have probably written this down and quoted you but i'm not that guy, you say you say something really profound which is that, ai won't make the organization's culture better it only amplifies the culture that's there i'm sort of paraphrasing if you want to give me the amplify if you want to give me the exact quote, I'll take it, but.

Yeah. Well, no, because what AI is good at is amplifying information. So, because that's its job. So, let me interrupt you. So, it's really quite good at listening to weak signals and making them louder. Absolutely. It is amazing at taking narrative, so taking lots of conversations. And being able to amplify that component. But if that data is poor to begin with, then it's going to amplify a poor result.

So the example... So poop in, poop out. Well, I would say amplify poop out. Okay. So poop in, giant poop out. Oh, I hope that's the name that it picks. Poop in, giant poop out. That's great. An example is we did a piece of research which looked at pre-start conversations. And we had 23 participants all view and listen to a pre-start conversation with a bunch of workers.

And we compared the results of what 23 people heard versus what AI heard. So does it come as no surprise that there was huge variability with what people heard? Sure. No surprise at all. Okay, in actual fact, the data said there was a two and a half times variability, from people from the same conversation.

And that is because humans have their own biases. So we don't listen. We're only judging information we receive based on what we want to hear. AI did an amazing job because its bias was coded. In this case, it was coded to listen to certain things. So, an example was, in that worker conversation, one of the workers said, it's like Forklift City again.

I nearly got clipped 10 minutes ago. Wow. Only one of 23 people identified the near-miss. All 23 people identified forklifts as a hazard. Yeah. Which you're exactly right. It's not surprising that the workers, because of. The multitude of biases that we have as human beings, we don't see that near miss because it's so common that it just becomes, it's just part of the noise.

If you look at the signal to noise ratio. And I like this idea that AI amplifies weak signals because that seems like a really positive approach to AI and it's a non-surveillance-based approach to AI. Correct, because the objective here was, how do we present information and leaders in such a way that it makes them curious?

Does this have the potential? So let's talk about inshittification, which is a term that I'm quite fond of. I wish I would have invented it. It's definitely within my capability to invent that term. I just didn't do it. Yes, can't fit on a hat, but can definitely go on a T-shirt. That's right. That's right. So incentivification is the idea, for those of you that don't know, that.

As we build more and more systems that think for themselves, that the experience that the user has dramatically decreases. And the example that's pretty commonly used is Uber. So when Uber first rolled out globally, it was really user-focused. It was good for the person taking the ride and it was good for the person giving the ride.

But as Uber became more and more habituated and people used it more and became more dependent on it and it diffused into the system in kind of an Everett Rogers way, right, that what happened is that the Uber people started taking away the interfaces that made Uber easy to use, easy to contact. And the example that I would use within the example I'm giving you is that there used to be a customer service interface to Uber.

And if you had a problem or left your phone or, you know, or I don't know, something happened, a driver was mean to you or you were mean to the driver, there was a person you could talk to and action would happen. That's been completely 100% removed. In fact, the only way you can contact Uber now is through really a chat bot.

And the chat bot's not very satisfying. And you see this. I mean, Uber is just one example. There's a bazillion of them out there. Will AI and shitify two levels for this question, Brent? So get ready. Buckle your psychological seatbelts. Will it inshittify leadership and their skills to listen? And secondly, will it inshittify our procedures and how our procedures and practices, our formal operational procedures function within the system?

What's your guess on that? Well, the response is it has to without some form of assurance framework. Okay, tell me more what that means because that sounded typed. That sounds very heady. Because if leaders are able to ask those questions that tells them that's the way that they're moving, then AI will just make that reality.

So, for instance, in the book, we pulled together what we call a leader assurance test, which basically looked at the fact that everyone's pushing AI as being that next productivity tool, that next game change tool, all these things, because that's what we push. You know, it's either fear of compliance or the fear of missing out. Right.

So we deliberately put in there some curious questions leaders could ask of that AI project, to understand the potential upside or downside of AI when it comes to system, when it comes to learning, and when it actually comes to people. So how important is that purchase intent? Because I think organizations are buying AI to gather more efficiency and reduce numbers of personnel. I mean, I'll just say that. I don't know if it's true or not. You'll have to tell me what your research found.

But I do hear organizations talk about using more AI in lieu of more people. Yes, and there are many cases where they've done that, and now they've hired those people back. Yeah, yeah. I didn't say it worked. I never once said it worked. But, Todd, I mean, for many listeners, they may not have known that before I got into safety in 2001, the first 20 years of my life was at IT.

Right. Oh, yeah, I probably should have. A better podcast would have said that. I mean, you come to this topic with a multitude of experience in wealth, for sure. And I probably should have established that. Sorry. That's okay. And I've worked for those large American corporates. Where that's exactly what happens. You come out with something new and there is what we call the gold rush. Right, right.

So this piece of software will improve your productivity, you know, by X percent. This will do that. This will do that. And we said the saying in IT that the demo looked like heaven. Yeah. Organization looked like hell. Oh, yeah, that's true. I mean, that's really true. Well, but we're – AI is really weird because, you know, we're still very frontier.

We're still on the rough edges. I mean – and you hear this in the news. I mean it would be really hard to exist and not have heard stories on, you know, the potential need to provide governance around AI. I almost said HR. That would have been a little bit of a Freudian slip there. Governance around AI, you know, some barriers, some bumpers, some lanes for it to follow. Yeah, guardrails. Guardrails, yeah.

All those things are important. But once again, this is not new. So this is the same thing that technology has always faced. Right. That's good to hear, actually. Yeah. So, and I try and do that in the book. I mean, I kicked the book off by giving a brief history of AI from the 1950s, that this is not new.

But what about, and you do, that is a good discussion in there, but what about the potential for AI to become senient, you know, to think for itself? Oh, where'd you go, Brent? Hello, come in. Come in. Pips are still silicon-based. Yeah. And to me, the perfect response to that would be what AI will do, it may be thinking for itself, but it's going to amplify the culture that it's created in.

I mean, this idea that AI doesn't make your culture better, it amplifies what culture is there, is actually a really, really important touchstone. Because if you have a crappy culture, you're not going to AI your way into a better organization. You've still got to do the work, right? You still have to work at understanding how you interface with your workforce and that the worker's voice matters.

All the things we've talked about, which is really the point of this entire book, right? Absolutely. And look, the reason why I believe that AI has a helpful component was the fact that as organizations want to operationally learn, if they really want to learn from what's happening in the operation, they actually need to hear the voice of the workers.

Not see that filtered data that is captured in checklists and other sheets. Yeah. How many times they cross the red line? Absolutely. And AI has that potential, to be able to actually capture those frontline insights and present that in a way that is useful for the organization to actually learn from.

I mean, back during the pandemic, we wrote that little white paper when we introduced everyone to the four Ds. Right. And in that white paper, we talked about, you know, how do you start learning? Which is a good question. Yeah, absolutely. And the problem is learning is resource intensive. Yeah, absolutely. It takes time, energy, effort, people. Correct.

AI will help us so that resource used is actually digging into the things that are more systemic based. What was the biggest learning you got out of doing this book? What's your take? I got to tell you. The Brent Sutton, Brent Robinson takeaway. What did you guys get out of this? Not one that I was looking for, which was around the neurodiversity. Some of that.

And the impact on humans. And, you know, a big shout out to Nicola Noble. And I've known Nicola for a long time. And, you know, Nicola's work as a safety leader is in the impact of safety around neurodiverse workers. Yeah. And, you know, Nicola is neurodiverse as well. And to be honest, you know, I'm in a similar camp.

Okay? Because we've all got a bit of tism. Yes. In that scale. But it was really interesting. What was the impact? Because I read data recently, something like 22% of the workforce is neurodiverse. Yeah, of course. If you've ever worked at a place like if you've worked, I'll just make up an example. If you've worked at a national laboratory, you've worked with people who are incredibly smart, but maybe not as socially comfortable as your neighbors are.

Well, that was normal in IT. Right. Yeah, that's true. You guys are tracking it, actually. You would suck them in. We harnessed it. So I was a systems analyst. You would take them right from the Dungeons and Dragons table into the workplace. Absolutely. I was trained as a systems analyst, not a coder, because I had the capability of actually talking to people and understanding their need.

But I do think the book's approach to this neurodiverse worker, and really it's tough because neurodiverse is kind of a trigger. A little bit. It makes you think things. But what it's really saying is that different workers think differently. And because they think differently, this is a real advantage that AI can bring to the table.

I did notice that the book really kind of jumped in on that stuff. And I thought that was very interesting. Yeah, because I was really interested in doing that section around wearables. Because I, You know, because my concern is we're overloading people. We are. And I agree with you. You know, asking someone to use a wearable is built on trust.

So when someone says, let's start feeding those essential decision points to the worker in the field, that's when I believe you've crossed the line. Yeah, no, I agree. And it's weird to me because this is a paradox part of this, is people will gladly pay $900 for an Apple Watch. Yeah. Right? Or that kind of wearable. And that's tracking everything.

I mean, that's amazing what it tracks. And we voluntarily sort of give up autonomy. It's a really interesting part of the whole surveillance question. And it is a paradox. I mean, we're not going to answer it today. It is. But supposedly we're giving that up for the better good. Yeah. Yeah. There's a benefit somewhere. Correct. But what about when it's the direct opposite?

Which it can be. Absolutely. And I think probably is. I mean, if they start using your Apple Watch data to determine your insurance cost, right? Or they're not really wearables, but the drivables they put in your car to determine how you drive. That's a really interesting paradox. That's a really interesting question about this notion of surveillance. Brent, it's a great book.

The AI Paradox in Safety. And it's available anywhere you can buy books, right? I mean, it's... Yeah, pretty much. That's the objective. We've made it value for money. And I don't say this often, but this is one you probably ought to buy. It's a really good book. You did a good job, my friend. Yes. And, you know, there's a touch of AI in there, but mostly it's human.

That's perfect. Thank you. Thanks, Tom. So that's the pod, and that's the conversation, and it's just beginning. Give the book a look. I think it's worth your time. It's probably going to be, something you've been thinking about anyway. I mean, I don't see how you can be in the safety and reliability world and not have sort of thought about this.

And if it's your first time, well, welcome aboard because this is a big one. That's it for us. Thank you again for everybody that reached out. Tell your friends. Everybody can listen. And until then, learn something new every single day. Bet you did today. Have as much fun as you possibly can. Be good to each other. Be kind to each other.

Take care of one another. Check in on each other, even the least of us. And for goodness sakes.

PAPod 618 - Poop In...Louder Poop Out?

Kapitelklicken springt direkt hin

Shownotes

Transkriptvom Podcast bereitgestellt · Zeitmarken springen hin