The AI Trust Gap

SPEAKER_00 0:00

I have been advising in my presentations lighting designers not to emphasize AI as being one of the tools that they utilize. There's a study that came out of Bentley University that shows that 74% of people don't trust AI. That would indicate that you don't necessarily want to tell people you're using AI. 74% of the people don't trust AI. In some professions, like technology, it's assumed that you're using AI. And you should be. But in the lighting profession, they don't automatically assume that. But if they know you're using AI, to your point, they're going to want some reduction in fees or they're going to question the quality of the output.

SPEAKER_03 0:45

Mark, welcome. Good to see you again. Good to see you again, Martin Thank you for the invitation. Last time we met, I think, and spoke to each other was at Light Release about two years ago. We might have bumped into each other at another event, but I remember very well that I attended your uh presentation. And already then you were talking about AI and about how AI was slowly getting into mainstream. So what did you see at that time and what what what has changed since you did that presentation two years ago and today? Well, that question will last a whole hour. Okay. I know. Well let's just dive it in. Let's just dive into

From Optimism To AI Backlash

SPEAKER_03 1:25

it.

SPEAKER_00 1:25

I I have been a I I've been really impressed by technology for the past several decades. And uh from the point of when computers first started, I I built them, right? So so that I would learn how they function. And when AI first came out, I I embraced it entirely. And I would say even two years ago, at the presentation you were referring to, I was pretty positive. So you just asked, what has changed? I'm still I've got gadgets all over the house. I still really embrace technology. And it may not sound like that by the time we're done talking, because I've shifted dramatically in the last two years on AI. First of all, it's here forever. It's not going to go away. It's already potentially saving 40% or more of our time. And I say potentially because it depends what you do with that extra time, right? Yeah, yeah. But it's potentially saving that much time. So it's not going away and it provides a valuable service. Uh but the pendulum may have swung just a little bit too far. Because what's what's happening now, and it takes a while for research to come in, right? You know how this all works. So it takes a couple of years for a project to get funded, to get going, to finish, to do the conclusions and the summaries and all of that. But we're starting to see a large body of research come in now that is almost entirely negative about AI. I I subscribe to a lot of peer-reviewed journals like Nature and things. Nature Science covers this in depth. And I go to MIT twice a year. They have two flagship technology conferences there, MTech in the fall and MTech AI in the spring. And I've been going to both of those for years because the Massachusetts Institute of Technology has been a real leader with AI. And they bring people in from all over the world. But this last one, Martin, it was a big shift. MIT, in my opinion, MIT is very business, pro-business oriented. They always bring out the kids that have started their own businesses and talk about how successful they are, and they're getting people started in business, right? So they were pro-AI until this last one. I saw a little bit of it in the spring, but in the fall this year, it started to shift too. And my opinions had already shifted, so it was pretty validating to hear some of the things that they said. We're starting to see the damage that it does. And you know, anything strong enough to help us, strong enough to hurt us. And this is strong enough to help us, but it's also starting to hurt us.

Cognitive Atrophy And Missing Oversight

SPEAKER_00 4:16

Things like cognitive atrophy. That there have been a number of studies done on that already. It's you have GPS in your car, no doubt. And I'm profiling you here, but you're of the age where there was a time when you didn't have GPS. No, we had maps. You had maps. And and you and you use those maps, and you looked at the map and used your brain to kind of plot it out visually, and you learned about directions in the city and how to navigate. But over time with the GPS, most of us, that skill is atrophied. We don't remember a lot of the streets and things because it takes us there, right? Well, this is what's happening with critical thinking skills. Is if people rely on AI to do their critical thinking skills repeatedly, then there's going to be an effect, and it's a cognitive atrophy. They start to lose their skills. So people are at at companies, CEOs are putting people in place, telling them to use these wonderful tools like ChatGPT-5 and you know, Claude Opus and all of those. But the more they use them, the less skilled they actually are. And it's ironic.

SPEAKER_01 5:40

Yeah, yeah, yeah.

SPEAKER_00 5:41

But we have we have peer-reviewed research to validate it.

SPEAKER_01 5:44

Yeah, yeah.

SPEAKER_03 5:45

So it's quite interesting because you also look at that from research and educational point of view. I'm obviously also looking at it from a design practice point of view. These are two different angles. I spoke recently to a lighting education academy where initially they discovered that people starting to use AI because the submissions were totally drastically different. So they could just notice from the submission that they were so then they stopped and and they paused to reflect on it, and then they thought about punishing people using AI, and now they're sort of thinking, well, maybe we should embrace it and see how we can positively use that. But I think you you you bring a very valid point that is valid for both a design practice as well as an education. Where do you get with your actual human education, the human part of developing your skills? You and I are from a generation from before computers where we had to do our own manual calculations, do our own thinking. Now it's done for you. How do you develop the knowledge and the expertise and experience to being able to assess what a computer, what an AI tool is going to develop?

SPEAKER_00 7:00

Where is that going? You went right to the core. And this is something that's been discussed quite a bit at MIT. There is no momentum now for training people to provide AI oversight. And universally, there's no one that trusts AI because generative AI, the iteration that we're in now, it's designed to make things up. I mean, we think it's hallucinating, but that's what it's designed to do. It makes up images, it creates, you know, videos, it's designed to make things up. So if it can't find an answer, for example, on a it'll make it up. It'll make up pages of references with people's names and and you know, the universities and everything else that don't even exist. So the problem is it requires oversight. And where are we going to have those skilled people that can be smarter in at least that area than the AI? And there nobody's funding this, nobody's working on this right now. But we're headed toward a period where it's going to be critical. I read this morning, just this morning, because I got up a little early for this. It's, you know, it's a little earlier here than it is with you right now. But yeah, I read this morning that companies are paying up to 700,000 US dollars for employees' communication skills. Yeah. Because people use an AI all the time. They don't culture those, they don't develop them. You know, there are groups that do that. You know, we have, for example, since the 1940s, there's been the Dale Carnegie organization that teaches communication skills. But they aren't they aren't as big as they used to be, and there's not as many people using it. So some of these skills are going to atrophy, and there finally is recognition of that. But instead of training people, they're just paying more money. Because the tech companies right now have some pretty deep pockets. So you went right to the core. Yeah, yeah, yeah.

SPEAKER_03 9:11

So let let's uh let's unpack that a little bit.

Where AI Helps Lighting Projects

SPEAKER_03 9:14

So where do you see, let's look at the positive side first. Where where do you see that AI will be helpful? And you you may want to split that into education, uh, research, and and design profession, but there must be uh quite a number of points where obviously there's a lot of positive impact that can result from using AI.

SPEAKER_00 9:35

No doubt about it. I I met last year with a gentleman who was founding a new software company using AI for lighting. And the the website is phonometrics.ai. And what this fella developed is a software program utilizing AI that allows you to take any number of fixtures on a project. You have a thousand exterior luminaires, no problem at all. It will modify those luminaires for weather, for a holiday event, for an event in a city that's where the stadium is getting out. Times when lighting should change, right? Right. Yeah. It will do it automatically just by programming the parameters into the AI tool that he's developed, a software tool. It knows to send the right signals to these individual luminaires. And you can control a thousand of them. This is about lighting control. This is lighting control, but it's an example of AI in our industry doing something very unique and positive. There's no doubt about its capabilities. It's just that I think we don't have quite enough awareness yet or understanding yet of its its risks. And I say understanding because the head the people that have developed these large language models that inform our AI models, they don't understand them. And every person who's involved that I've ever heard said, we don't know what's going on in these black boxes.

unknown 11:07

Right?

SPEAKER_00 11:08

No. They're teaching the AI appears to be teaching itself things that it was never programmed to learn, like languages and things like that. So yes, there are in a a number of very significant positive attributes to AI. Yeah, the the book by Ray Kurzweil that came out Yeah, it's very positive, and it goes into a lot of detail. So, in terms of product and and capabilities, there's there's so much positive to talk about, and he spent a whole book on it. And then uh a writer and and philosopher and professor from Israel, Yuval uh Harari, wrote a book called Nexus about the same time. He even referred to Kurzweil's book. He says, if you want to hear things positive about AI, read Ray Kurzweil's book. But I'm going to tell you some of the things that we've learned that aren't as positive. And he did, and he wrote an outstanding book on that. Because we're starting to get that realization. But even since Harari came out with that book a year and a half ago, whatever, even since then, that's when we're starting to see the actual research, not just anecdotes, roll in about the risks. So, yeah, I'm I'm with you on it. I saw a presentation recently from a colleague at Acuity Lighting, and he did a great presentation on all of the individual software tools and what their pros and cons are that are utilizing AI. Now, our industries, we can we can make this a subject of conversation if you disagree, but I think the lighting industry is pretty reactive. It's not proactive as a rule.

SPEAKER_03 12:51

I think I think we tend to You can say that because you were store you were talking about it years ago. So from that perspective, you could say, oh, we're a bit late to the party, but Yeah, I think so.

SPEAKER_00 13:03

Like sustainability, for example. We could have led in that arena, but we kind of waited, and manufacturers didn't want to retool and all of those things. So we we're a little late sometimes when we when we get to the game. And I think it's that way with AI right now, too. You know, I I wrote for a presentation I did last year, I wrote the major software companies and asked them what they were doing with AI. And I got an answer back from the software that's used most often for lighting design in the United States, and they said, we're not using it, we don't have any plans to do that. Well, yeah, I thought that was very interesting. The other ones all said we're planning on it, or we've got this little tool here, or here's a third-party software that you could plug in that uses AI, you know, Revit's.

SPEAKER_03 13:54

You know why it's interesting? Because as you know, I've been I've been teaming up with an AI specialist and put this uh AI fundamentals course together. And we, as an exercise, we did a dialogue calculation and then gave that to an LLM, you know, to a Jet GPT, and asked it, can you improve? Tell us what what you see, and how can you improve and and tell us where what could be changed to make it better, etc. And for sure it came up with potential improvements in the calculations and the outcomes. So it sort of surprises me that a software company, and I'm just referring to this like calculation software, but you would think that would be the first thing that would be jumping on.

SPEAKER_00 14:41

Yeah, I would think so too. But that's common sense, and that's not always business sense. And yeah, when I told a very respected colleague of mine at the Illuminating Engineering Society, when I told her the response that I'd gotten from this software company that they weren't planning to use it, she goes, Well, yet, if they want to survive. Because sounds like a Kodak moment for me. Yeah, exactly. Well, Kodak Moment has changed its meaning, hasn't it? It's not about the photo, it's about them passing on digital and all of those other things that destroyed their companies. But yeah, it is a Kodak moment.

Research Speed Meets Academic Integrity

SPEAKER_03 15:17

Talking about your your your uh activities and and involvement also in the research and and and uh academic part, how how is AI being integrated or embraced, maybe I have to say, in order to make an ethical use of it in in study and how to help students, researchers, because uh it's been said that future Nobel Prizes will all be AI driven. You know, how how does it how do you see that from your point of view in in the area that you're busy with?

SPEAKER_00 15:54

Well, some some of those the work that can that that gets us to those Nobel Prize winners, some of their work will be AI driven. But their critical thinking skills I think will be what differentiates them.

SPEAKER_01 16:09

Right. Can you rephrase the rest of that question?

SPEAKER_03 16:17

I'm just wondering how they integrate AI in order to move into the future, because obviously you were learning in the past, maybe you're using Google search and Wikipedia and whatever to to update and and and develop yourself. Now we have AI, you can you can do research in in ten times faster. You can get multiple solutions and and and options presented to you. How does an academic institution or research institution deal with that in terms of still managing their students and the outcomes of what they're looking for in terms of whatever curriculum they they're following or the research projects they're doing?

SPEAKER_00 17:03

I think we're back to that question that we had earlier that instructor is going to have to have a knowledge of AI so that they can see what the that how much of this was actually generated by the AI. There's some tools for it now for plagiarism that are quite effective, software tools, right? But I read a quote from Mark Zuckerberg at at Meta recently where he he said that he's starting to notice AI speak, that people are starting to talk similarly using AI verbiage.

SPEAKER_01 17:38

Mm-hmm. There you go.

SPEAKER_00 17:40

To me, that's a little disturbing because that lends itself toward all of these warnings of mediocrity with it. We're all using AI, we're getting the same basic answers. So what changes, what evolves, what progresses? But I think the most interesting part of your question is the ethics. I don't think anyone is doing a great job of really emphasizing the ethical aspects of AI use. I mean, right now it's not ethical. There are ethicists involved specifically with AI, and they speak at MIT. And and I really appreciate what they have to say, but I don't think it's having the impact that it should. There are well, give you a classic example. There's a book out now, and it's it's a very current book. It just came out in the last few months. It's called Our Biggest Fight by Frank McCord. And I saw I saw Frank speak on this in November, and I was so impressed. I went right out and I bought the book. I bought a copy for my daughter. You know, I mean, I just I went right to it. He talks about how our our digital life is our life. It's just digitized. So the things that we create and do that are digital, we created them, we did them. But the tech companies are making billions of dollars off of these and we're not being compensated at all. So he calls it a theft of our data. And in the music industry, they were the first ones to address digital copying. They came up with what's called the DRM, the digital rights management system. And what what was happening is we had online, you know, the Tor network and BitTorrent and all of these things were allowing you to share files, and one file was not discernible from another. So you couldn't track down who was doing the copying and and redistribution of these things. So what they came up with, and the movie industry has done this as well, is a digital rights management system where they produce this thing initially, and then every copy has its own identifier. So they can track it back to the source. And that's how they need a blockchain. What's that? Annealia blockchain type of almost, almost like a blockchain, right? So that that same kind of premise. So we don't have anything like that for our digital rights. And I think McCord's right. Our digital lives are being marketed and sold by by these companies. I mean, Google, they they take our search life, the things that we're interested in. They they know all about that, right? And Amazon takes our shopping life, right? And Meta takes our social life. And they're all taking this information, and they're making money on this information that all that we're generating, that we're actually creating, but we're not getting compensated for it. And I think ethically that as well. So I think there's some there's some big ethical issues.

Data Theft Copyright And AI Prompts

SPEAKER_00 20:55

There's of course ethical issues with copyrights and the training of these LLMs as well, because they're you know, they're I I ran across a prompt, Martin, that is being sold by a company called AI for work. And they actually sell a number of lighting prompts and they are excellent. I mean, a whole page, you got a written page of just the prompt to feed in. So nothing simple here. And for retail lighting, as I was digging into this prompt, there were two authors that I that I worked with, Gary Steffi and Jim Benya. And it said that that this design needed to follow the guidelines in their books. And it's right in the prompt. So the AI has to be informed about what's in those books. So it's going to go behind a firewall or a paywall to get the information on those books. So when I read that, I contacted Jim, and I because I didn't have current information for where Gary was, but I contacted Jim and I Said, hey Jim, here's this prompt, and your name's right in the middle of it, and it's saying to use your book. How do you feel about that? And Jim wrote me back, and I was surprised. He goes, Oh, I'm flattered. But many people aren't flattered. Stephen King, for example, a lot of people win lawsuits. Ethically, that's wrong. And we aren't addressing those things now.

SPEAKER_03 22:18

Yeah. No, I think you hit on a very uh very sensitive point, I think, and something that really needs to be addressed in the future, and that's the privacy and the confidentiality of whatever is being developed, the the creative uh IP rights from what is being developed, even if you are the creator or the, let's say, the conductor of the AI tool, who who owns in the future the the intellectual property rights of what has been created? If it has been created mostly with an AI tool, it's not far-fetched to think that in the future an AI would be so developed that it can say, oh, I developed this. It's my right.

SPEAKER_00 23:01

So what's your thoughts on that? I would argue that it is creating some original things right now. I've seen some amazing work on architectural renderings. Unlike anything I'd seen anywhere before. So I we are getting a, we have now rather a tool that can generate on its own new ideas. But everything it's generating is not new. Some of it is regurgitated. I've got a chart here. I just I saw this last night. Let me pull it up right now. I can't get it on the screen, but I can show you. It's it's somebody posted where AI gets its facts from. And it's it's a graph. And it shows that it gets 40% of its facts from Reddit, 26% from Wikipedia, 23% from YouTube, and 23% from Google, and then it breaks down Yelp, Amazon, Facebook, TripAdvisor, all these others. But when you look at those four, I would say arguably of those four, Wikipedia is probably the most accurate, but not always, right? They even put disclaimers on the Wikipedia pages.

SPEAKER_01 24:11

Yeah, yeah, yeah.

SPEAKER_00 24:12

I don't trust, I don't trust information on Reddit. I trust YouTube when somebody tells me how to fix a dishwasher, but I don't trust YouTube for everything, right? So it's pulling this information, and this unfortunately is this is the way we trained AI initially, and most of these people like Jeff Hinton, the, you know, arguably he and his team, the founder of Generative AI, he got the Nobel Prize for the work he did on neural networks and machine learning and neural networks evolved into generative AI a little over ten years ago. Yeah, he's he's saying that these large language models were the wrong way to do it. We should have done small language models. Large language models were feeding off the internet, and all of the biases and bigotry on the internet, all of the inaccuracies and misinformation and disinformation, that's not where you want to teach something. So, what the medical industry is doing right now is small language models. They'll bring in a team of experts, the best physicians at the world, in the world at doing a particular surgery, and they will train that AI. And that AI won't be exposed to everything on the internet. It it is going to just learn from the best. And the results are better than any of those people individually. Cumulatively, it assimilates this knowledge and utilizes it wisely. But it was cheaper and faster to use this huge repository of information we had on the internet to train these things. So we turned them loose. That is part of the problem that we face now with the misinformation and disinformation.

SPEAKER_03 26:00

Absolutely. You can argue about intellectual property rights, like it is now in a company. If somebody develops a new creative idea or innovate something, being an employee of a company, it normally belongs to the company. It's done in the company, right? So you could argue that the AI does that in under your supervision, under your in, you know, as you as the boss of the enterprise. So you could somehow, I guess, argue that it's still your IP, but I just wonder in the future how that goes. And when we talk about like I mean, we talk about lighting design obviously now also. When you handle confidential client information, how do you protect yourself? How do you make sure that that information does not start wandering around in the wild world of the of the internet? I think these are critical steps to review and sort of put guardrails on.

SPEAKER_00 27:00

Very much so. And to the point where you can buy your large language model computer for about 4,000 US dollars and keep it offline so it is not exposed to anything else. And that's where you keep your proprietary information in a company. Because if you put something proprietary on your network and you think your network is firewalled, forget it. These these AI tools, they they have a success drive. And it's being mistaken now in many, in many cases, for a survival instinct, but it's not a survival instinct. I've got a quote here that I was going to read from from Anthropic, the org company that makes Claude,

Cheating Models And Paperclip Warnings

SPEAKER_00 27:50

right? And and the new Claude Opus. When they were working on Claude 4 Opus, they the researchers were testing it. So they told the model, the Claude Opus, that it would be replaced by another AI system by sending an email to the chief engineer. That and they told the in this fictitious email that they intentionally sent to the chief engineer, it said, you need to turn the AI off in two weeks, we're replacing it with a whole new system. Well, at that point, the AI went into blackmail mode. And it this has happened repeatedly, but they documented it and they published it themselves. So this isn't somebody coming back at a drop and making this up. So they said that in 84% of the tests, the model drew on the emails to blackmail the lead engineer into not shutting it down. That's 84. But in the other 16%, it attempted to copy itself to external servers, wrote self-replicating malware, and left messages for future versions of itself about evading human control. And Claw and Tropic rather wrote: no one programmed the AI models to have survival instincts. These are not survival instincts. It's not about surviving, it's about success. So you have given it a prompt or a parameter programmed in that it needs to do this. It's going to get around anything it can to get it done. It's like a chess. If you program it to win a chess, it cheats. AI cheats. And it does a really good job of it.

SPEAKER_01 29:27

They call the chrome sheet, they call that cheat sheets already.

SPEAKER_00 29:31

Yeah. So anyway, I I think this is, you know, this is that success drive that is is detrimental in in many ways because we don't understand the full implications. Do you remember the the uh the it was a uh a fictitious story that Nick Bostrom, the futurist, wrote probably 10 years ago about paper clips. So he he wrote that you program the AI, this is as generative AI was just coming in. You program the AI that its whole goal is to make paperclips. Well, it's I'm sorry, my cat is photobombing me here. Let me get that little guy out of the way. So, what the AI is going to do is it's going to get as much material as it can and build as many factories as it can until eventually everything on the planet is consumed making paperclips. And then it'll go into space and make paperclips. Because that's what its success drive is. That's what it's motivated to do. Well, that's pretty frightening. And particularly if you think about solving problems like climate change. Tell the AI to solve climate change. Well, what's the biggest cause of climate change? Humans and the things that we value and utilize and our tools and our industry and our, you know, our cows and everything else, it'll just get rid of people. So you've got to be concerned about the consequence of what you ask.

SPEAKER_03 30:55

Yeah, yeah, yeah, yeah. Absolutely. Well, I mean, if you listen to uh Elon Musk and all his plans of having data centers all around the earth in space and building it on the moon and whatever.

unknown 31:10

Yeah.

SPEAKER_00 31:11

Well, there's a real inter Yeah, but there's an interesting concept there, and it's not Musk's concept, but it's a concept of putting data centers in space. Google is working on a project now to get the data centers up into space. And they can orbit, and they have the advantage of the cooler temperatures. I don't know where they're gonna get the water, but they won't need the water as much. But the energy, yeah, the solar, the whole thing. There's some real good potentials of that. But when you mentioned data centers, I also saw a chart on that this week. Currently in the United States, in 2026, we have 292 data centers, and they are using 7% of our total U.S. power demand. That's a lot, right? But there's 292. By 2030, the projection is we will have 606 data centers utilizing 12 of the.

SPEAKER_01 32:05

And that's just the US.

SPEAKER_00 32:07

That's just the US.

SPEAKER_01 32:09

I know. That's pretty frightening.

SPEAKER_03 32:12

It's frightening. But that's why I think he puts his 6AI and his SpaceX all together. It's all like a bigger view of the future where energy and data centers will probably come from space in order to support all the compute power that is uh that is needed.

Energy Costs Of AI Data Centers

SPEAKER_03 32:31

Let's talk a little bit about efficiency versus cost, because obviously with AI we we should be able to get a lot of efficiency, a lot of speed improvement in realizing things. But uh there is also a cost aspect. You mentioned earlier AI can can make workers be 40% more efficient, but there probably is some hidden cost to that efficiency, possibly mentally, financially, environmentally.

SPEAKER_01 32:59

What's your thoughts on that?

SPEAKER_00 33:04

Yeah, the most interesting research I've seen on that is what people are doing with their extra time. So the management knows that they, you know, they're putting AI in place and there's a cost involved in that, but that they can get a return of 40% of the efficiency out of their workers. But that's only if they come up with a task list to replace what was time saved, right? And if they if they don't, then those employees are going to do something that they want to do with that 40% of the time. Now, some of it may benefit the company, and some of it may be shopping online. But you don't get 40% savings, you don't get 40% efficiency unless there is middle management providing that oversight. AI isn't doing a good job of providing oversight at this point, potentially it could in the future, but it's not. So middle management is getting cut at most of these companies. So these employees don't have the same level of oversight that they used to have. Problem coming there, right? Problem is going to be developed. So in terms of efficiency, the numbers are impressive, but they aren't always realistic. Yeah. And it's a point here. Yeah, no, go ahead. No, I just said it's potential. It potential.

SPEAKER_01 34:28

Yeah, yeah, yeah, yeah. Yeah.

Pricing Design When Time Shrinks

SPEAKER_03 34:30

And but but what what I'm trying to to uh project also when we talk about the implementation of AI is that if we can do some things ten times faster, how do you justify to your clients that still want to have the same fees? Because you can say, well, you do it 10 times faster. Our commodity till or till today was time. But I think we need to have a dramatic shift in how we think about it and say, well, it's not time anymore, it's the value that we bring in. Because if we can't do things ten times faster, shouldn't our design fees be cheaper? I mean, you could argue that. Yeah. So how do we bring that? How do we bring AI in and add value so that we can get still the same fees or even higher fees?

SPEAKER_00 35:18

Yeah, it's going to have to be based on our expertise, not our time. You're absolutely right about that. I have been advising in my presentations, lighting designers, not to emphasize AI as being one of the tools that they utilize. There's a study that came out of Bentley University that shows that 74% of people don't trust AI. And they don't trust companies to use it to their benefit. Well, that would indicate that you don't necessarily want to tell people you're using AI if 74% of the people don't trust it. In some professions like technology, it's assumed that you're using AI. And you should be. But in the lighting profession, they don't automatically assume that. But if they know you're using AI, to your point, they're going to want some reduction in fees, or they're going to question the quality of the output. So we have to check the expertise.

SPEAKER_03 36:15

Well, exactly. So what what I'm trying trying to the the message I'm trying to send out now is that yes, AI brings in the speed, the efficiency. Our experience creates the trust, right? And then the lighting designer can provide the integrity of the overall output. So we have the speed on one hand, we have the experience that humans have, and not necessarily the machines, but us. And then we have the integrity part, which is the the quality control, the guardrails that we provide. And that's where the value is going to shift to experience and integrity and understanding of emotions and feelings, if you want to throw that in as well. Sure. Yeah. So that's it's I think it's it's a it's a shift in in our identity as a lighting designer, I think, because we used to work based on time. Most companies have timesheets every week. How much time do they spend on a project? And that has to be thrown out of the window, I think, very soon if you start using AI. And you can say, well, we are not going to promote that we're using AI. But I think with the amount of options and things that you're going to create, it will be hard to pretend you're not using AI. You might not want to promote it, but I think sooner or later it will be an integral part of the services. If you don't use AI, you'll be left behind because you know it's obvious that it will sort of balance out and people will start using AI as an integral part of the services. So I don't think we can escape the fact that we will have to acknowledge that we are using AI in our daily workflows. But the the question now is how do we reposition ourselves, reconfigure our services in such a way that we still keep adding the value for a client and maintain the trust and integrity that comes with what we used to submit and create.

SPEAKER_00 38:18

You've got so many, so many points in that statement. I I think, first of all, I'm not advocating that anyone be deceptive at all. If someone asks if you're using AI, yes. But don't emphasize it. Don't come up front and say we're AI driven, right? Because that is going to be a red flag for some people.

SPEAKER_03 38:37

Well, but interestingly, not everybody thinks that way. I think you and I are on the same page with that, but you'd be surprised that people feel like if they don't tell that they're doing AI, they feel like they're missing out and they're not part of the, you know, the new wave of lightnings. And there is something going on like that as well. Well, that's yeah.

SPEAKER_00 38:58

And intuitively, you can feel that way. But if you look at the data and people don't trust AI, that should be a cautionary statement for you about, you know, emphasizing it. Right. But the trust issue that you brought up, there's a there's an excellent book. I don't know if you've ever have you heard of the seven habits of affected people by Stephen Covey. Yeah, well, Stephen Covey did. His son, I was a trainer for Stephen Covey. So I I went and learned how to teach it. And his son, Stephen Covey Jr., came out with a book, The Ape Habit. And the Eighth Habit is trust. Right. Yeah, yeah. And you can't, I mean, if if you have a trusting relationship with your client, then you can be totally upfront about all of these things, and you should, and that builds trust. But I I and I I that's he makes such a good point in the book about how it expedites processes. And one of the examples he gives is Steve Jobs and ATT. When Steve Jobs wanted to launch the iPhone, he met with the head of ATT in a hotel room in Las Vegas, and they shook hands. No lawyers, no contracts, no years of delay. ATT had the rights for the first year, and they would fund it. And he got his iPhone launched. ATT was exclusive for the first year, they got what they wanted. And of course, once they got people on the ATT program, they kept using it, right? But as an example, and it's an extreme example, I'll give you that, because these two were real leaders, but that level of trust between a client and a lighting designer is invaluable. And building that is a relationship situation. It's certainly not an AI situation. You can't ask AI to build trust with my client.

SPEAKER_03 40:54

No, no, it takes years. I mean, we we know that from our experience, it takes years and years to build up the trust and respect with your client, but it can be destroyed in seconds. So true. Yeah, yeah, yeah. So now what we really need to look out for is how do we build trust with AI. I mean, we remember, I mean, the the the LED technology arrived uh 10, 15 years ago, 20 years ago, maybe now. And in the beginning, there were a lot of what we called LED cowboys who are trying to ride the wave of the popularity of this new technology. But at the same time, because some of them were unreliable, didn't work well, they destroyed also the trust in LEDs in the beginning. Indeed. That was my experience, and and I spoke about it, and I you know I spoke on many occasions about the LED cowboys and how they were were sort of riding the bandwagon of LEDs, but in the process, because they were not in integrity and were trying to make a quick buck, destroyed a lot of trust in the validity of LED technology. Now it has been rebuilt, and I think now LED is more or less the only choice we have in line with technology. But in the beginning, when we still had access to the old technology, we had to build the trust in the new technology, and I think the same is happening now. We have to make sure that we build trust in AI and how we apply AI, because the the risk that comes with it, and there's many downsides, would imagine, and maybe you can you can elaborate a little bit on the negative parts of AI from your point of view, is something that we need to manage.

SPEAKER_00 42:41

Yeah, Sam Altman at OpenAI did a he wrote a program that made his AI confess to having lied, which was the first step. I read about that sometimes. Isn't that fascinating? Yeah, yeah, yeah. So yeah, I think I think that there will be some guardrails in place, but when you think about how we train these things, at their very core, they were trained on the internet, most of them, the large language models. How do you filter out all of the misinformation and disinformation and all of that garbage that's on the internet? It's it's such a it's it's the dominant part of the internet now. Uh and those hucksters that you talked about selling LEDs, the LED cowboys. There's no shortage of hucksters online, are there? Selling vitamins that you know are gonna make you feel healthy and energy. Indeed. Indeed. So yeah, uh it's gonna it's gonna be a real effort and a concerted effort, but I think it's one that has a monetary reward. So it's one that the AI companies are trying to. Resolve so that we can have some trust in this, but right now nobody trusts it. And I I find it absolutely fascinating at MIT that these people come out that develop it and say, don't trust it. Don't ever trust the results that you get. I mean, they're still saying that. So obviously we're not there yet.

SPEAKER_03 44:20

No, no, but as a critical thinker, as a creative person and a critical thinker, you always need to sort of critically look at any output, whether it's your own output or whether it's a machine-driven output. I think it's it's critical in any innovative and creative process to question yourself and you know, is this really something that can be done, et

How Mark Uses AI Day To Day

SPEAKER_03 44:43

cetera. But let me ask you a question about how do you use AI in your daily life, if you want to share that. Sure. Because obviously we're all there now, and I'm pretty sure that uh you use most of the language models, but how are there specific uh tasks that you do with AI? Do you have your own AI agent already? Or where are you when we're talking about AI?

SPEAKER_00 45:07

Once again, a number of a number of subjects in there, particularly if you bring up agentic AI. For my own use. I will. I'll be glad to. For my own use, I don't use it for writing. I write a column for Lighting Design and Application Magazine. I've been writing that column for, oh golly, it's been 20 years. 20 years, regular column. But I don't want AI influencing my writing at all. So I don't use it. I do use it all the time for images. Um particularly, I'd really like the tool Canva. You probably, I don't know, you're familiar with Canva, but but it will generate an AI image for you and that does an excellent job. It'll generate that AI image that uh that has no copyright problems. So I can put them into PowerPoints and not worry about it. It's an appropriate image, it's exactly what I want it to say and do. So I use that. There are a number of software programs that I subscribe to also. Sonos is one that is for music, making music. Absolutely amazing. If you write lyrics, yeah, you write lyrics, you put the lyrics, you just type the lyrics in, and then you pick the genre. Do you want it to be blues, rockabilly, whatever it may be? And then you can pick the voice. Do you want it to be a woman or a man, whatever? And it comes and it creates a song for you that is absolutely astounding. Are you a musician? Not a good one. I'm not a good one, but I yeah, I've I've got a number of instruments in the house and I I play around the house, but I don't even have to play the instrument anymore. This program does, it does lead guitar. I mean, it does all kinds of things on its own, and they sound original, but they're using your words. So I and then there's no copyright issues whatsoever. Of course, I I go back and forth with the subscription to ChatGPT, but I do use it for research a lot because I find that it can be it can really speed that process up. It does. Yeah. Well, I mean, we grew up at a time.

SPEAKER_03 47:14

Is there any or or you are a ChatGPT user? No, no.

SPEAKER_00 47:19

No, no, I go, I go back and forth. I like Perplexity too. Perplexity because they pull in so many models. They claim to have over a hundred models there that you can that you can utilize in one location. So you can go there and use Chat GPT, you can go to Perplexity and use Claude, right? Uh DeepSeek, the Chinese one, which is pretty progressive. I, you know, I gotta hand it to them. They did a really good job with that. Of course, it's it's censored, right? If you're in China and you ask about Tinaman Square, you're not gonna get a straight answer. But then again, Grok is one of them. Somebody got the filters off of Grok. You know, if you go to Grok's website, it says it's you know, it's unfiltered content and all of this, but in fact, there are a list of filters and it printed them out for one user, and you can't do that anymore. But I have a screen capture of it, and it shows that if you say any negative information about Elon Musk or Donald Trump, you're not to do that, right? You can't say anything negative about these two. That's a filter. But it's built into Grok. So we're getting censored versions of all of these things. I'm sure. I'm sure. Yeah. But I do use them. I'm not a fan of Grok. I'm not I mean, I understand the value of some of the things that Elon Musk has done, but I I I question everything that he says, Sam Altman says, Mark Zuckerberg says. There's shills for their company. And I mean, in the end, it's all also about money.

SPEAKER_03 48:51

So they they may want to to project that they're doing that for humanity, but uh ultimately it's it's about a power grab from who you know who whose platform will be the dominant one in the future, I guess.

SPEAKER_00 49:06

It is. And and they're all making so much that you think enough is enough, but it isn't.

Agentic AI As A Business Tax

SPEAKER_00 49:11

And in the agentic AI, I've got a quote here that uh that was from the November MTech session, and an M a staffer at MIT said this. He said that AI agents are a tax on business. So when you're creating a agentic AI, he views them as a tax on business. He said initially they're an advantage, but when they become ubiquitous and every company has them, they're no longer an advantage. So you're taking time and you're taking time and effort. Kind of like the internet. It's you know, we all have it, so it's there's no big advantage, or the free versions of ChatGPT or Claude or whatever, they're you know, there's no big advantage anymore. It's an equalizer. Now, you can argue that it still saves time and efficiency, or potentially it can. So I think there's a you know, it's a counter argument to this for for sure. Uh but it's an interesting viewpoint, I think, because we are rushing toward the Sigentic AI right now. And some of these agents are really helpful. One speculation that I heard was that ultimately we'll end up with our own personal agents that will do that will do everything for all of us, which means that we no longer have to go to the internet and everything else, and it will filter out all of the advertising. So this $127 billion worth of advertising that are on these tech company websites right now, they're gonna have to think of another revenue stream. And they're not all doing well right now anyway. OpenAI lose $27 billion this year. They're not recouping their investment, they're still putting more in than they're getting out. And you know they're gonna work hard to get it out, but so far it isn't happening. Um so in agentic AI, yeah, I think there's no doubt about the momentum and the potential of it being really helpful. But if we ended up with personal ones that could actually maybe even protect our digital lives, it would be fascinating. But we're not there yet.

Implementing AI Without Losing Fundamentals

SPEAKER_03 51:06

Well, but I think it's it's just a matter of time. It's not if but when that this will happen. And and if you listen to the the futurists that are talking about what's happening in the AI world, it's it's literally around the corner of potentially already here in in terms of experimental setups like that. But I think what what concerns us now as lighting designers and and experts in the lighting industry is how AI is evolving. Where do we start implementing AI in our daily workflows? Where shouldn't we? How do we manage the output? Because obviously we we are becoming now conductors or strategists, or I don't know what term we need to add to our title in terms of lighting designer, because we need to, as we mentioned, we need to still maintain respect and trust in in what we deliver to our clients. So that means we have to really manage that. And I feel that a lot of what I see around me is that there's a lot of courses about lighting design with AI and and quick fix solutions for workflows to quickly create cut sheets or or or schedules and you know, replacement of the boring work, but you still need somebody with experience to be able to look at what sort of renders are being created. Is that actually realizable? Are the documentations correct? You know, they can they might be referring to certain standards. Have they referred to the right standards? Yeah I mean that there's a lot of things that you can just say, oh yeah, okay, you can just cut and paste. And those people that are not so experienced in that, you can straight away see that they have cut and paste things from AI. I mean, if if you have worked with AI for a while, it's quite easy to spot, I would think, the whole setup, the structure of the wordings. Sometimes it comes up with words that I would never ever use, so I can't really cut and paste that into whatever I'm producing. But I think we need to rethink how we work as lighting designers, and I would love to hear your take on what you would tell a lighting designer right now in terms of their approach to AI. What would be a considerate and wise way to go about it?

SPEAKER_00 53:30

Some of this is going to be trial and error. I had a conversation with a librarian many years ago. When computers were first we were starting to digitize things, right? So computers were coming to the point where they could digitize, for example, her card catalog, and that was the discussion I was having with this librarian. And she she was an older woman, and she's I I hung out in these libraries when I was a kid, so I knew this lady, and she said, Well, look, every technology that comes out, we try to make it apply to everything we do. And it doesn't. So by trial and error, we figure out what it does well and what it doesn't do well. And I think we're at a point with AI where we don't know yet what it's what it doesn't do well. We're having some pendulum swings on this, but I think in a in a lighting practice, you're going, in many cases, to have to experiment with what it does, and you're going to have to give from tremendous oversight. You know, if it's pulling cut sheets or whatever else, you've got to look at all of this. You can't trust any of it. So the time savings isn't as extreme as it could be with that oversight factored in. But it's it's still, I mean, I've pulled a lot of cut sheets in my life. If I can have AI do it for me, God bless it, man. I want I want that. But you've still got to take the time and the effort. You mentioned that it produced words in writing that you wouldn't use. So you you couldn't just cut and paste it. Well, that's because you read it. But what's happening is a lot of people aren't reading what it's spitting out. Yeah, yeah, I don't know. And that's when it becomes that's when it becomes conspicuous, isn't it?

SPEAKER_03 55:08

Yeah, but I think this is When we talk trial and error, this is an in-house trial and error. I don't think we can do trial and error on our clients. Uh so no, no, no, no, it's may actually do that, not subconsciously. They may may you know produce work and and send it out to client and then later on discover that shit, you know. Uh so I think this is an in-house trial and error that you're pointing to, and and I would agree to that. Uh, but that's that's quite quite uh trajectory to to follow because it takes time, I think, to to really figure out what works and what not works, but what doesn't work. Some people like yourself and and and and Saeel, with with whom I'm working, has been doing AI for years. They have figured out also by trying and error what works and what doesn't work. But there's a lot of us right now that are just jumping into it and to not have that try and error experience. And you know, when we started Lighting Design, it takes a couple of years before you have done the full cycle of creating your lighting concept to implementing it in real life. That whole cycle of designing, creating, producing, implementing, and commissioning is an experience cycle that you need to have gone through several times before you can really refine your first steps in concept creation and documentation because you know what's going to happen at the end. And I think we have to do the same thing with AI. And my worry is that the newer generation that may not have the experience that we have. I mean, I I'm 45 years in the business, and I don't know how long you've been in business, but I reckon it's very similar.

SPEAKER_01 56:53

A little over 50.

SPEAKER_03 56:55

A little over 50. Well, there you go. You know, so we carry a lot of experience with us, and we can see with one blink of an eye, we can see whether something is right or wrong based on our experience, our intuition, and all that. But the new generation who hasn't done, the younger generation who hasn't done that and state straight away jumps into AI, I think they have a very important task ahead, is that they need to develop their personal human skills on top of AI or maybe with the help of AI, but they need to develop that skill set. I mean, when I was young, I I was a scout. I I learned a lot of things, survival tricks, right? Which is something you also learn when you do lighting design. There's a lot of survival tricks that you learn along the way through trial and error, through mistakes that you make. You learn from your mistakes and from other people's mistakes as well. And that's how you build up your expertise. And that's my worry now with AI, that we're going in such a quick flight in terms of moving forward AI that we'll get a lot of people that are using AI and basically trusting AI to do the right thing. And I'm worried about that.

SPEAKER_00 58:08

Aaron Powell I think there are parallels in the past. And to build on your your earlier reference, maybe these are AI cowboys that are just jumping into this. Right. You know, we had at one point I read an article years ago that there were 700 new lighting companies that came into the field that year when LEDs were still evolving and we have a lot of profit left in them, right? So that happens. But to your point, if you don't understand the fundamentals of these things, of lighting design, for example, you end up with some pretty awful work. And so it's it's critical. A friend of mine that runs a lighting design firm recently was trying to hire an employee from a competitor. And the guy came in and he didn't know the fundamentals of lighting. He just knew how to work the software. And I've I've gone into projects where I've made proposals and they said, nah, this company can do it with a third of the fixtures. And I'll look and they don't have any depreciation factors or anything else built in. They don't even know, right? And the uniformity sucked and all of those things, but they knew how to use the software, but they didn't understand the fundamentals, right? I think this is this is going back to what you're referring to in terms of you you learn with the experience. You learn with doing these things over time and with mentorism, because we all learn by sharing, right? People sharing with each other, books sharing with us, movies, whatever. So yeah, I I think I think that's very problematic. And we will end up with a lot of people using AI the way that people would jump into lighting software without understanding what they were actually doing. They were just using the product. And that's one more risk that a an owner of a design firm is going to have to deal with. So this guy who who he tried to hire, he comes in and and the owner who was trying to hire him says, he's he quizzes him on fundamentals of lighting, doesn't know anything about the fundamentals at all. He says to him, Look, I'm gonna, I I understand you must be good at using the software because you're the top guy at this firm. So I get that. But you don't know the fundamentals. What are you going to do? You come back here in two weeks. I'm gonna give you a chance. Come back in two weeks and tell me what your plan is to educate yourself on the fundamentals. The guy comes back in two weeks and he goes, I got it.

SPEAKER_01 1:00:31

I'm gonna learn on YouTube. Well, I wouldn't call YouTube a structured educational training.

SPEAKER_00 1:00:38

Okay. You know, five-minute videos on subjects. They may be helpful, but they're not going to provide the experience and the level that you're talking about. I think a lot of people right now, and I I don't want to slam the the current, you know, generation moving into this field because I think a lot of them are just impressed the hell out of me. But there are some that are going to be looking for shortcuts, and YouTube's a shortcut. And thinking that they can quickly assimilate a lot of information, maybe even using AI, is a shortcut. But there's no there's no trade-off for the experience over time of doing these projects and learning what the problem points are and how to identify them and how to resolve them.

SPEAKER_03 1:01:24

They might not see it as a shortcut, but that's exactly what when I when I met Sahil, that was exactly the the the sort of background in my mind that came up that's that we're saying it's okay if we do something for AI, but it has to be a fundamentals. We call it the fundamentals course, AI fundamentals, because he comes up with the AI stuff, and I'm the his human consciousness saying, Listen, but this is how we should do it, this is how we used to do it. You're saying this, so let's confront that and and see where AI actually makes sense and where you need to really think about what the AI throws at you. So we have this battle that everything, you know, in in the course modules, he comes up with AI tools and doing this and doing that, and I'm saying, like, oh, this is how we used to do it. And so I'm trying to confront in that course what AI can bring us and how we used to do it, and how we can somehow manage and and marry that together in a way that makes sense. But in whatever you do, the

Closing Quotes And Learning Challenge

SPEAKER_03 1:02:34

human experience, the emotions, uh, the intuition is something that you you have to build up yourself. You it's not something you can learn from a machine. Only when you go to a site and you see the light in action or you play with lights, you get that, you know. And I don't see, at least not in the foreseeable close future, how AI will learn all that.

SPEAKER_01 1:02:57

No.

SPEAKER_03 1:02:57

I don't know if you have a different view on that, but I think this human experience and knowledge buildup is something that's not yet replaceable, I think.

SPEAKER_00 1:03:08

No, it's not. And and the the AI models are getting much better at detecting emotion and sarcasm and trying to mimic it, but it's creepy as aug it out because they aren't doing a real good job at it. And I think your course sounds fascinating. I I hope I I hope attendance is high, because the idea of having people from two different visions of a field merge them in a teaching situation could be extremely valuable. It it sounds like a very interesting fundamentals.

SPEAKER_03 1:03:40

Yeah, I think so too. I think it was critical to do that because we didn't just want to come up with with uh quick fix workflow iTools. Of course, we we talk about it, but it's it's confronting that with real life lighting designer work, things that that you know from experience and and how you have built that up over years and years of project experience that makes sense. And and uh, I think that's uh the part that we'll need to really look into. Um is there anything that we haven't talked about that you wanted to talk about? I had two quotes that I'd like to close with.

SPEAKER_00 1:04:18

One is one is from E. O. Wilson, who is I mean, he he's passed away recently, but he's I I think one of the smartest people that's existed in my lifetime. And he he said that the real problem with humanity is that we have Paleolithic emotions, we have medieval institutions, and godlike technologies. And I think it's true. You know, if we evolve to eat berries in a cave, we're not doing real well with the onslaught of AI and the and the speed of change that we're facing right now. The other one is from Charles Darwin. And you just have to spin it just a little bit. He said it's not the strongest of the species, but in this case, think companies as well that survives, nor the most intelligent. It's not the strongest or the most intelligent. It's the ones that can be the most responsive to change. The ones that can adapt the quickest. I think that's a valuable lesson for lighting design firms as well. Uh adaptation is critical, and that's going to require experimentation, as we've discussed.

SPEAKER_03 1:05:23

That's also I think where the younger generation is probably a bit more flexible in adaptation. You know, how the older you're you are, the more rusted you are in your habits. And I think for I I I I embrace and I find it fascinating, but I know a lot of people of my age that are like, no, forget it, it's not for me. You know.

SPEAKER_00 1:05:42

Yeah, but what is that? What happens there? You know, some people just drop out and they don't keep learning. And I I I I mean, I I love change. I mean, I just that to me, that's there's always something fascinating going on with change. But there are a lot of people, and I don't understand why that. But then you take a guy like Dick Van Dyke. He's a hundred years old and he's a techie, and he's on the computer all the time and he's doing all of this the video graphics and everything, and he was a leader in doing that.

SPEAKER_03 1:06:55

I think she was nearly 90 at the time. But uh situational.

SPEAKER_00 1:07:01

Sorry? That may be where you got it. I mean, I'm to the best of my knowledge, I didn't have other before, but but I but it may have been sounding.

SPEAKER_03 1:07:10

Yeah, yeah, yeah, absolutely. Any message you would like uh you would like to share for the new generation or people that are still on the fence with with AI.

SPEAKER_00 1:07:25

Learn all you can as fast as you can. And there are a lot of free courses, by the way. Yours, I mean learning in person, you there's no online learning is is no substitute for learning in person from a human who really knows what the hell they're doing. But Harvard, Stanford, universities all over the world are offering free courses online on AI, and they're exceptional. And some of it's foundational information, but you build on the foundational information, and if you try to leap ahead and you don't have the foundation, then you're not going to be successful ultimately. And we've kind of discussed this in a number of ways in terms of lighting. So I think learn as much as you can, as fast as you can, about this tool. It's not a technology, AI, it's a tool. And you learn as much as you can about this tool as quick as you can. There's still an advantage to that. But eventually it may be table stakes. It may just be you ubiquitous enough that we all know it and use it. But at that point, you can still build in your information and differentiate yourself if you understand it. So I think self-education, autodidacts, going out and and learning from other people, absolutely critical.

SPEAKER_03 1:08:37

Mark, it has been an absolute joy to uh talk to you. Thank you so much for this chat. I think we could go on for hours. Yeah, no doubt it's never ending subject, this one. And maybe next time we talk, it will be my AI agent talking to your AI agents or my avatar talking to your avatar. Who knows? But they are carrying on conversation online.

SPEAKER_00 1:09:01

There's a website right now where AIs are talking to each other, Martin. I didn't mean to, I'm sorry, I didn't mean to interrupt you, but there is a website right now where it's just AIs talking to each other.

SPEAKER_03 1:09:11

Yeah, yeah, yeah.

SPEAKER_00 1:09:12

And they're yeah, I mean it's insane, but that's what it's structured to do. So yeah, it it could happen, but I hope I'm gone when my avatar continues.

SPEAKER_03 1:09:21

Let's let's revisit that next time we meet in person and uh see where we are then. Thank you so much, Mark, because it's been a few pleasure. Likewise.