AI Ethics: Transparency

Doug Berger: [00:00:00] Welcome to the latest installment of Brand of Brothers. I'm Doug.

Johnny Diggz: And I'm Johnny. Today we're continuing our conversation about generative AI ethics, specifically transparency.

Doug Berger: All right, let's get

to it

Johnny Diggz: And we're back. So Doug, we're talking about transparency today. Yes, John. Specifically how it deal- how, how it relates to the HBS, the Higgins Berger Scale of AI Ethics. We talked about that last

Doug Berger: episode. Yes, and we even came up with an amazing mnemonic. Okay. I maybe- ... came up with a questionable mnemonic about how DIGGZ and DOUG intend to party.

And we're talking about the two. We are talking about the two, the transparency. And it's

Johnny Diggz: not just transparency, it's transpa- like it's disclosure, it's all of the things that be- we [00:01:00] have like this new, this new world of AI, and where and when and how, and all of the ethics g- that go along with this new, this new capability that the world has.

Um, and specifically in, in creation of, of creative, really creative, uh,

Doug Berger: usage. Right. And, and really transparency isn't about compulsive disclosure, but that's one aspect that we're getting into. Um, obviously there are some legal ramifications that we'll be talking about, um, but ultimately what we're gonna be diving into is how the Higgins Berger Scale of Generative AI Ethics, um, has multiple components that we discussed in our previous episode.

And- You wanna go over those again real quick? Um, not really. Uh DIGS and DOUG. So- So DIGS and DOUG, uh, basically you've got data usage, uh, and [00:02:00] privacy, right? Okay. So that's one of the Ds. The other D is displacement. Okay. Um, then intend to party is going to be, um, intent. Okay. T is transparency, what we're talking about today.

And then P is potential for harm. Okay. And so we'll dive into all of those individually show by show, but today we're gonna dive into transparency, and the main reason we're diving into transparency is because that's the top option of our, uh, our scale, right? So, uh, basically each one we're gonna go through, um, how we have quantified each component from a series of qualifications.

When you say top option, you're talking about the a- the tool that you built. That's right. Okay. If you go to r3mx.com/hbs, that will take you to this- awesome utility that we put together [00:03:00] that makes it so if you have any questions about the ethical implications of anything you've made with generative AI, you then complete this simple questionnaire that has a rating scale for each of those five components that we just mentioned.

Johnny Diggz: So we, we kind of picked transparency because it's kind of bubbled up to the forefront in the news recently. Right. Well, um, I mean, you texted me a couple weeks ago, and you were like, "Hey, you see this new EU law, um, which requires, uh, companies to disclose the use of any AI- Right ... in, in their content product?"

What, I

Doug Berger: mean- And we're even seeing the disclosure of AI becoming automated, um- Sure ... a- as we've talked about previously, where, you know, you post something that has the metadata embedded in it, and, uh, and, and speaking of Meta, Meta- Mm-hmm ... will make it so [00:04:00] it automatically discloses that AI was utilized.

It'll flag

Johnny Diggz: the actual post as, you know, give, gives you, like, a little AI info button, sorry, um, a little AI info button or, like, that you can click on that kind of explains their policy. It doesn't actually talk about how AI was used in this, but it, there is some, uh, a lot of the tools now are embedding, um, signals or flags inside the actual content that they create to indicate to the, the distribution platforms that, that something has been, at some point in its workflow, has been touched by AI.

Doug Berger: Well, let's talk about the, the workflow and how the scale, uh, is impacted by the workflow. Um, so when you look at the, at the, the Higgins Burger Scale, and you're utilizing the utility that we built, there's going to be a [00:05:00] handy dropdown that basically helps you ascertain what degree of, of really ethical cleanliness, um, has been employed.

So at the very top level is radical disclosure. So that is completely compliant with the laws being implemented in the EU that I believe roll out in August, if I'm not mistaken. Um, and there's a, a similar but very different law that is being rolled out in California also at the same time in August, if I recall correctly.

It d- yeah, it doesn't

Johnny Diggz: stop there because there's also, uh, similar disclosures, laws in other states, and I'm sure other countries as well. It's, it's, i- it's moving from this sort of ethical gray area to a legal- Uh, less gray area

Doug Berger: Yeah. It, the complication is [00:06:00] from my perspective, the continuum is black and white on the extremes, right?

So you're either doing your level best to disclose any potential utilization of generative AI at any point in the process versus the complete opposite, which is intentional obfuscation, and not just being deliberate about it, but also with malicious intent. Then you ha- And, and on our scale, so our scale, um, it, it's a little bit complicated.

Okay. So basically we give zero points for radical disclosure, and the goal of course is to get zero points. The lower your- This is like golf,

Johnny Diggz: right?

Doug Berger: It, it, it's exactly like golf, um, except, uh, the par, uh, skips over number three. So, so basically you have, um, I'm gonna cheat here because I think that it's best for everybody involved.

So [00:07:00] basically on our scale, a zero would be a clear and accurate representation, right? So you are radically disclosing. Then there's, it... and, and as I mentioned, it goes zero points, one point, two points, four points. So then four points is going to be assessed if there's intentional obscuring or misleading claims.

Johnny Diggz: Um, so an example of this would be what? Of where somebody might, might intentionally obscure the use of AI for some nefarious gain.

Doug Berger: Right. So there are so many different use cases for this, but the most reasonable expectation for this to be employed is where, for example, let's say that there is a marketing coordinator who has been tasked to write blog posts.

Well, obviously you can generate blog posts to your heart's [00:08:00] content using generative AI with minimal involvement, right? You give it a simple prompt. You say how many words or characters you want it to be, whether or not you tell it-

Johnny Diggz: You can, you can even, like, give it examples of your previous blog posts that your company has,

Doug Berger: you know- Right.

You can feed it with whatever data you want. Right. A- and if, this isn't limited to ChatGPT or Claude either. It, it- Yeah ... it, it is- A lot of tools do this. Yes, absolutely. Um, and so obviously that bad actor doesn't care about this tool. However, their boss or their boss's boss does. And so that's where this comes into play, where someone discovers, because of this metadata that we just mentioned being embedded in the content, that- They have discovered that one of their employees is utilizing AI and hasn't told anybody, and they haven't made it where it can be triaged for [00:09:00] them to be able to see where in the process it was utilized, right?

There is absolutely, uh, no, no path to connect the dots between humans i- in the loop versus a straightforward generative AI.

Johnny Diggz: So to take it to a little broader perspective, why, why would a manager or an owner or a, a, a... Why, why, why is that, uh, aside from the legal issues, I mean, w- w- w- Right ... let's just talk on the ethical side or the business decision side.

Sure. Um, w-

Doug Berger: why would they, why would they care? Perception, right? Perception is key. Do you wanna be perceived as someone who makes stuff up? Do you wanna be perceived as someone who doesn't originate their own content? Do you wanna be perceived as inauthentic, right? What differentiates you from others? A big part of that is the human component.

Johnny Diggz: I, you know, it makes me think of, there was, [00:10:00] uh, a, a, he was a, a c- I think it was a bank, it was a banking company that just announced a partnership with, uh, OpenAI, and he was doing a shareholder call, which is something, or a fiduciary call required by, by law, and, um, he did this, it was sort of a gimmick. He, he, he did disclose it- Oh, right

but the first half of his talk or his, on the call was his AI twin, his, his- Right ... and, and he had created, and they were kind of showing off this technology. But, um, but y- received some backlash because, uh, you know, especially in a bank, trust is, is critically important.

Doug Berger: But see, that, what you're talking about, because he took the initiative to disclose it after the fact, I would actually give that a two.

Okay. Right? So what that means is minimal or unclear disclosure, but no ultimate [00:11:00] deception, right? Sure. Um, but what you're doing there is stepping into that other territory, which are deep fakes, right? Mm-hmm. So when you're talking about celebrities, for example-

Johnny Diggz: Or influencers of any type, really. Yeah. Yeah, yeah.

Doug Berger: A- and, and, you know, it's become incredibly simple to over, oversimplify- ... uh, to train AI to be able to have the voice mapping- Sure ... and the image mapping- of a celebrity because there is so much. And celebrity, I mean- Yeah ... not just someone who's talented on screen- Or even you ... but an influence- Oh, I'm hardly a celebrity.

No, but it,

Johnny Diggz: it wouldn't ... If somebody wanted to, they could take, they could go back through our library of episodes, which- Oh, absolutely. Yeah ... which are all available on YouTube, um, a- and- And Spotify or wherever you get your favorite podcast So link and subscribe. Um, but, [00:12:00] uh, so they could, they could go through and create a digital version of what we're doing right now- Right

and say AI is good, and AI, or, you know, transparency is not necessary. Like-

Doug Berger: Yeah, and e- even- Yeah ... the differing levels o- of what could be perceived as nefarious, right? Because, uh, uh, as we, you and I had a conversation, uh, uh, n- not long ago, we were discussing about how even though they're ripping off the likeness, right?

The voice and, and the- Yeah ... the appearance of an influencer, for example, they're just echoing the same thoughts. Sure. It ... They're not even manipulating it to be a completely- Changing the message. Yeah, the message is basically the same. Yeah. It's

Johnny Diggz: being kept- But they're, they're collecting followers based on people just not realizing that this is not the original content.

Right.

Doug Berger: And then the next thing you know, they're following a, a pharmaceutical company. Right. Right, right. Um,

Johnny Diggz: [00:13:00] now we, whi- while we're live right now, and we're definitely not AI, we do use AI tools to complete these podcasts,

Doug Berger: right? Yeah, absolutely. Um, a- and what those tools are, it, it ranges, right? So, uh, the very first thing that we do is we process the video, and in processing the video, we have it automatically put together the first draft of our transcript, um, because it's way easier for it to transcribe and then for us to go in and edit than for us to do it the old-fashioned way, where you basically are like a stenographer and you keep hitting pause or there, or hold down the pedal- Right

to pause and play the video as you're typing in each word individually. So that a- Did

Johnny Diggz: we displace a dis- a stenographer?

Doug Berger: I, I don't believe so- ... because I know that in the past it was me.

Johnny Diggz: Right. Right.

Doug Berger: So all I've done is employed- You saved yourself

Johnny Diggz: some time. Right, by

Doug Berger: employing the utilization of this tool.

And don't [00:14:00] get me wrong, while it saves time, it still requires- attention. Right. Um, then there's the other piece- And it's not

Johnny Diggz: automatically doing this. I mean, you're literally, like, doing some copying and pasting and that kind of thing along the way, right?

Doug Berger: It, where required, right? Yeah, yeah. Because there are times where, for example, it's terrible at transcribing, right, name- Or identifying

Johnny Diggz: you versus me.

Right. Yeah.

Doug Berger: It, it, and, and, and it's, but it's terrible at identifying, like, words that go together. If you don't art- over-articulate and put space between your words, the name of our, our podcast becomes completely different than what it actually is. So just about every single episode, I have to go in and edit the transcript where I introduce the show because it, I say it too quickly.

Ah. Um, but then, a- as you mentioned, we also use it for identifying who's talking. So the AI will look at the, the, our, uh, voices. The, it, it's able to, to map our voices and know who's talking. And [00:15:00] so each episode, it goes back and forth and identifies who's talking when. It's mostly good at it. I must confess, we're pretty lazy at updating, uh, who the speaker is in the transcript.

Um, so to those of you who are- It would be

Johnny Diggz: really ironic if this part of it was labeling me as having said it.

Doug Berger: A- and so, uh, then we also utilize another AI component that automatically makes it sound better, and we do that more than once. So there's the initial post-processing. What's it called?

Johnny Diggz: That, that feature.

Doug Berger: Uh, Studio Sound- Okay ... is what it is called in the application that we use for editing our content. Then it gets uploaded to the platform that distributes the podcast, and there's another AI-enabled feature that upgrades the sound quality.

Johnny Diggz: Right. And then we also will use, uh, AI to create summaries of the [00:16:00] podcast- That's right

by using the transcript that has already been transcribed by AI. Yes. So there's multiple layers. Now, w- where does that fall in the disclosure, the, the transparency, um, uh, you know, uh, spectrum, if you will?

Doug Berger: So like I said, we have clear and accurate representation, right? That's zero points. We have intentional obscuring and misleading claims.

That's four points. Right. We, we've touched on both of those, and we kind of got into what I feel like is a gray area because I don't think it's black and white. I feel like this is incredibly nuanced, um, and, and it relies upon the person- Who is assessing their work or the work of others, um, it, it definitely requires them to a certain opinion, um, to determine how gray the area is.

In my opinion, [00:17:00] what we would be at is a partial disclosure or clear human authorship, and that's one point. Okay. However, someone else might disagree and say that it's minimal or unclear disclosure, but no intentional deception, which would be two points. That's two. Right? So it, it could be one or two, but I would argue that we're closer to the one.

And then, of course, in addition to that, we have a modifier which is openly acknowledging the utilization of AI. And as we, we mentioned moments ago, that's going to be a requirement, um, i- if you're, uh, putting content out in the EU, for example.

Speaker 3: Right.

Doug Berger: Or in, uh, in, in California for commercial gain. Um, so I, I would argue that, that we're somewhere in between, but because we're not intentionally obfuscating it- Sure

um, I would argue that, that we're more of a one than a two.

Johnny Diggz: So but [00:18:00] that, it's a good example of how it's not really ever as cut and dry as we'd like it to be because, um, you know, I w- I recently was reading an article. It was a, a Nashville artist, um, that she, she produces music that gets picked up by a sync license, uh, organization.

You know, and she, she gets paid to, to have her music used in commercials or, you know, TV shows or whatever, background, whatever. Um, it's a, it, it, a very legitimate business a lot of people do. Um, she co-wrote a song with some- another collaborator. Mm-hmm. And that other collaborator didn't disclose that they had prepared the demo using Suno or one of the, one of the- Like, to anyone?

Yeah, to anyone. Just, you know, said, "Hey, here's..." you know. Was

Doug Berger: it nefarious? Was it- I

Johnny Diggz: don't think so. I think it was maybe an error of omission and didn't realize- Mm ... um, that [00:19:00] the sync license company had a very strict no AI policy. And so when down the road in this workflow of this particular song, um, it made it past all the way to the sync license all the way up to the licensee, and when you get the backlash that, that, that gets identified as having gener- you know, been originated, um, uh- Using generative AI, yeah

with using, using generative AI, that puts the sync company at risk, which also puts that artist at risk, and anybody that that collaborator has collaborated with- Sure um, scrutinized because this, this is a pr- problem because the sync license company and the, and r- ultimately the, the, the companies that are buying this, this content, um, or licensing it or using it for advertising or whatever, they, they have their own policies.

And so- Sure ... it- one person's [00:20:00] omission, you know, even if it wasn't really intend- intended to be nefarious, was, um, w- you know, caused all this cascading potential problem, which lead, led this, this, this woman, this artist to s- to basically say she's not, she, she won't work with anybody who uses AI. She, you know...

And I think that that's-

Doug Berger: It's

Johnny Diggz: an extreme reaction- It's an ex- yeah ... but I, but I

Doug Berger: understand it because she experienced- A big backlash, yeah ... yeah, huge, right? Yeah. Um, it, it, from my understanding, she basically was flagged as an AI artist- Sure ... and, uh, and, and basically was having her income inhibited.

Johnny Diggz: And, and so, you know, obviously this creates a, a huge trust issue, the almost to the point that, um, that, that chain of custody that, that, um, is almost n- going to be required.

Um, [00:21:00] just like evidence in, you know, in court, you need to be able to take anything that is generated that is being sold or, or, or re- repurposed down the workflow. Um, you have to be able to prove human involvement all the way down to its source.

Doug Berger: Right. And, and I think that is really where the utilization of AI is within the ethical green zone, right?

When you are utilizing it for good, you're utilizing it as a tool, not as an original content source. You have started to utilize it the way that we're utilizing AI everywhere, right? So, uh, we've been using AI for the longest time with spell check and grammar check-

Johnny Diggz: Sure ...

Doug Berger: baked into Microsoft Word and, and Google, uh, w- whatever Google is.[00:22:00]

I can't even remember. Docs. Docs, yes. Um, thank you for, for chiming in. Um, so yeah, a- as far as the, the scale is concerned, um, obviously where th- that, her cohort- fell was as a two, right? It wasn't done with malicious intent. It didn't have any open acknowledgement of AI as a modifier. And so as a result, you know, two points would be assessed, and two points is definitely on the way to the orange, if not the red zone

Johnny Diggz: The, th- you know- Which are bad Um, the human in the loop, uh, part of this, I think, you know, uh, the, the, the courts have already been coming out with, um, you know, uh, if, uh, allowing AI to have, you know, copyrightable, uh, m- there's a, there's a, there, y- w- specifically with the copyright law, how [00:23:00] to, um, if there's a significant human input, it can be copyrightable.

But if it's, if, if it's pure AI, if someone just, you know, hit generate on a minimal prompt- Right ... then that's not copyrightable, and that's, you know, and that, that, that's where the law stands right now.

Doug Berger: Which makes sense- Yeah ... because the content is being drawn from somewhere, right? Right. I mean, we know that it's being trained on data.

We know that it's really about pattern recognition. But ultimately, there are strings of sentences, there are strings of melodies, there are strings of pixels making images that are being sampled from somewhere else. And it's only a matter of time where we find that it's a lot like how music sampling was, um, back in the late 20th century.

Johnny Diggz: Sure. The, I mean, the just, the- That, the, the, the- The late [00:24:00] 20th century. I feel so old ... back, back in the- Back in the 1900s ... the what? Yeah. Um, yeah, but you're right. The, the, that was incredibly disruptive when sampling started. There were all kinds of lawsuits and, and, uh, and, and ethical considerations and... But at the end of the day, the industry accepted it and built, uh, workflows and, and disclosures and- Mm-hmm

and- And all the- ... credits and, and, and- Yeah, exactly.

Doug Berger: And, and, uh, remuneration, right? Royalties- Sure, sure ... making sure that people were being paid for the music that they created that was ultimately sampled. Right. And that- Right? And I'm sure Mike McDonald is doing all right. He is.

Johnny Diggz: Whoa,

Doug Berger: whoa, whoa. For good.

Johnny Diggz: So I mean, I, I think, you know, to sort of g- wrap this up with a nice little bow, be- I, I think we n- we really can't because legally- Because there are four other components of the [00:25:00] HBS

there, well, there are four other. But I'm, w- specifically with transparency- Yeah. Um, the, the legal ramifications are still evolving- Yeah ... and we're gonna continue seeing that. That's why we created the H- HBS as, as an evolving framework. Sure.

Doug Berger: A- and, and just to be clear, there, there are ways of radical disclosure without it being a giant warning message.

For example, you've created a website. That website has blog posts. Those blog posts, you've used AI perhaps as a means to edit the content. You've perhaps used AI as a means to find strengths and weaknesses in the content in order to buttress what you're putting forward. Do you have to say outright underneath that article that AI has been utilized from it?

Some people would argue yes. However- [00:26:00] I see it sometimes ... I, I would say, I would go so far as to say it is appropriate and, and highly ethical to put that disclosure in the terms of use statement for the website. Um, mostly because this day and age, chances are very great that someone has used AI, whether or not they've realized it in the process, right?

How often do you type C- basically the idea of what you're saying and hit the space bar because you don't wanna type out those next five words because those are effectively the five words you would say. You would've, you would've chosen anyway, yeah. So i- in, in that instance, you're using AI.

Johnny Diggz: Sure, yeah.

It's, it, it's a, you know, it's a really... The, the modern generative AI models are really fancy auto-complete. I mean, and that- They really are. Yeah. And it works almost exactly the same way, it's just a really, really f- more advanced way [00:27:00] of doing it.

Doug Berger: And so in the realm of transparency, uh, i- if you're using generative AI and it's meant to originate content, you're probably don't even belong here when it comes to evaluating the ethical concerns.

This is really more about when you're editing an image, for example, and you highlight a segment because you want to delete it. Adobe, for example, uses Firefly, which is their AI component, to assess what pixels should go in and replace the other pixels that you've- Mm ... highlighted with the marquee. And if you just highlight with the marquee and hit generate, you, without a prompt, you're basically playing the, the, the roulette or, uh, or, or, or the- Yeah

the one-armed bandit It'll give you like three

Johnny Diggz: options of different [00:28:00] smudges that- E- exactly. Yeah.

Doug Berger: And, uh, and so- You know, do you have to have this radical disclosure that you use this tool? I would say not so much. However, as an agency, the right thing to do is to be forthcoming with your customers and say, "Hey- Right

in case there's any concern, we do utilize generative AI as a tool. We don't use it for originating content, but we do use it f- to help us generate better content."

Johnny Diggz: Sure. I mean, it, it pro- you know, if there's an image that the client wants to use, but, um, there's a paper bag in the background that they would rather not be there, and you use a tool to make that paper bag go away, um, you know, that, rather than the expense of doing a whole new photo shoot, you know, it, it -- Right

and the time. Oh, I mean it- ... we've been

Doug Berger: editing [00:29:00] images since before Photoshop- Yeah, sure ... and, and doing those types of things since before Photoshop. But now we have a tool that makes it simpler- Right ... makes it more effective, right? We're saving money and making more money as a result. Um, so really at the end of the day, it comes down to the appropriate levels of disclosure.

So w- my recommendation, just be transparent. Tell your customers you use AI as a tool. Tell your website visitors in your terms of, of use or your privacy policy that you utilize generative AI as a tool. Um, and, and really, it's more of a, a CYA, a cover your- Mm-hmm ... cover your ass- Yeah ... um, than anything else, because everybody that you work with will appreciate your candor and honesty, because that means trust, and that's really what this is [00:30:00] all about is creating trust signals.

Johnny Diggz: I think that's a great way to end this, this, uh, episode. Um, next time we're gonna t- continue our discussion about the, the Higgins-Burger Scale of AI Ethics. Um- And we're gonna

Doug Berger: go into the potential for harm.

Johnny Diggz: The potential for harm, that's a good one because, um- Thanks. ... I've been thinking a lot about- That, that wa- Yeah

that

Doug Berger: was actually your contribution to- ... the, the model.

Johnny Diggz: Well, the potential- It, it's, it's a great one because it gets into use cases. Um, so, you know, the difference between, uh, using AI to modify a picture that you're gonna post on your personal Facebook versus using AI to modify a picture that Coca-Cola is using at the Super Bowl.

So- So, yeah ... it's gonna be an exciting, exciting episode. Palpable. Palpable. Please, uh, please like and subscribe, and we appreciate all of you guys listening. And we've noticed, [00:31:00] I've noticed, that we've got listeners from all over the world. All

Doug Berger: over the world. It's crazy.

Johnny Diggz: Yeah, so we would appreciate it if w- some of you would say hi in the comments and, and tell us where you're from and w- and what you like and what you wanna hear about.

 

Doug Berger: Thank you for tuning in to Brand of Brothers. Big thank you to our presenting sponsor, Remixed, the branding agency, along with production assistance from Johnny Diggz, Simon Jacobsohn, and me, Doug Berger. We can't forget music by PRO. Speaking of not forgetting, remember to do that like and subscribe thing and find us at BrandShowLive.

com and follow us on the socials at BrandShowLive.