Transcripts are auto-generated and may contain errors.
0:00
Hey, I’m Rose.
And I’m Angie.
And this is Our Lives with Bots, the show where we ask important, timely questions about what it means to live with our bot counterparts.
And from time to time, we also dive deep into what an AI future might look like for us.
0:15
Sometimes we agree, sometimes we spiral, but we always go deep.
Welcome lovely viewers to episode 4 of our series 4 with Our Lives with Bots, where you talk about AI policy and regulation.
And today we have a lot of juicy content for you, including what actually happened with the cases against social media in New Mexico and California.
0:40
Again, juicy.
You’ll get some details there.
What’s been happening in terms of new policy coming out, We’ll talk about a recent bill that was passed in China that will take effect this summer.
And new New York State bills that are referencing older bills passed in December that have some new information that is really relevant for things like Chachi, PT Health and anyone using these AI tools as a lawyer for legal advice.
1:08
And we’ll also be talking about age gating and some of the problems or whatever that might happen with that, considering some scientists actually call it dangerous and unacceptable.
And we’ll also be talking about things like Meta’s new AI glasses, right?
1:24
That can just pick up video and audio all around you.
And people are using it to take pictures of others without their consent.
So we’ll be talking about all of that and more today.
And we’ll get us started with the legal cases.
1:39
Angie, I think you have a lot to say about these, do you not?
I I do, and I’ve just realized why everything feels so blurry and why I can’t see things as crisply as I normally can, so it’s now much better.
You’re missing your glasses.
1:54
Are those?
They’re my normal one.
No, they’re not meta.
But when we get to the meta.
Recording.
I’m not recording our recording.
That would be very post-modern.
Recording, meta recording, but meta.
Meta recording – very good.
I hope there’s somebody out there that finds it both funny and at least understandable.
2:14
We’ll get onto the meta glasses in a moment, probably in quite a few moments knowing us because we do talk for a while on each story.
So let’s start off with the case on the 25th of March.
This is where Meta was taken to court, and this happened in New Mexico, and it was probably the first of the big social media cases that started going in a new direction.
2:34
So in the past, I think there’s been a lot of cases where social media and the various different instances of social media has been questioned in terms of its impact on users, society in general.
And it’s never really gone in the way of users.
It’s always gone in the way of big tech because they’ve got the money to pay for it.
2:52
But in this case, in New Mexico, Mehta was ordered to pay $375 million, which is not a small amount, right?
I wouldn’t be able to fit that bill.
I wonder how much their net profit that is or just their, you know, gross profits, right?
3:13
But I still don’t think.
Exactly what, 1%?
But I mean even then, you know that’s also a hand slap, right?
To get a bigger bill.
And I think the bigger issue, whether it’s a hand slap or it’s an insignificant amount of money, I think the bigger issue that Meta has with it is about setting precedent.
3:30
So we’ll get on to that in a moment.
Because they’re far bigger thing is that if they’re found guilty of anything is could that set precedent for Meta or for other social media platforms or as we’ll come to talk about it, any technology platforms when it comes to AI?
Yeah.
3:45
So what exactly were they penalized for?
It was the design, right?
Like in terms of how the apps were designed?
Yeah, exactly.
So they were found guilty, or at least they were penalized in order to pay the spine for violating New Mexico’s Unfair Practices Act, and specifically because they misled the public about the safety of its platform for young users.
4:08
So this is slightly different to the LA one, which was about the design.
But in this particular case, they argued that Mehta had misled the users about just how safe or unsafe its platform was for young users, and to the court raised the concern that children were being endangered and exposed to sexually explicit material, and that through Meta’s platform, they’re being put in contact with sexual predators.
4:34
And so it was quite interesting.
This trial lasted about 7 weeks.
And during this time, jurors were presented with internal Meta documents, which is quite interesting.
A whistleblower from Meta who left in 2021, The gentleman’s name was Arturo Bijar, and he actually presented various documents and made cases from within Meta.
4:53
And he was able to show that during his time in Meta, he saw that people were exposed to sexually explicit material.
He even spoke about a time during Meta where, through Instagram, about 16% of Instagram users had been shown unwanted nudity or sexual activity in a single week.
5:12
That’s crazy.
Also, it’s like if this whistleblower hadn’t come forward, will we have any information regarding the fact that Meta was aware of these metrics on their apps and were monitoring it and yet doing nothing about it?
Like what would happen if there wasn’t this sort of content and data provided?
5:32
I think that’s exactly the concern.
I mean, here is a whistleblower who obviously left because he was worried about what this kind of platform or this kind of technology was doing to users.
He specifically argued that he was worried about his own young daughter who is propositioned for sex by a stranger on Instagram.
5:50
You know, and we see this where people in technology about their own children and we we often hear about this, right.
Yes, Sam Altman or whoever.
What was it Sam Altman saying?
Like I wouldn’t let my kid use ChatGPT or something?
Or was it another big tag leader?
I think exactly Sam Altman at one point said that or.
6:07
Mark Zuckerberg, yeah.
Both, I think, have said similar things.
But I mean, to your point, what would have happened if we hadn’t heard from this whistleblower?
Nothing.
I think nothing would have happened.
There we are.
Because during the case, Mehta says, oh, we, we work hard to keep people safe on our platforms and are clear about the challenges of identifying and removing bad actors and harmful content.
6:34
We remain confident in our record of protecting teens online.
And I guess if I look at it with the with the most generous reading, the reality is I am sure they do do their best.
Let’s go with that because I don’t think they want these kinds of lawsuits.
6:50
So I generally do believe that they try their best to do that.
I just think bad actors are generally quite clever actors and will work very hard to get around this kind of protection.
So I think mention they.
7:07
Hireright.
And they hire lawyers and give them a lot of money for the sake of being able to argue in a clever way out of whatever they’ve been accused of, right?
You know, plausible deniability and all of that in this context would be that kind of line of, well, we try our best, but we can’t protect against everything.
7:27
Exactly.
Meanwhile, meanwhile, I think the LA case, maybe we can talk about that.
There are some very clear indications of what they knew well.
Yes.
And I mean, the fact is, I think they didn’t know about this, which is why they’re working against it.
And so they’re going to appeal this.
7:43
But the LA case, well, they really were on their toes because this LA case happened the very next day.
Talk about having a busy legal team.
So the very next day, the LA courts found in favour of the plaintiff and fined Meta less $6 million.
8:01
But still, it’s all about the precedent, right?
And there they split that between I think it was Google and Meta, and they said.
Because of YouTube, Yeah.
Exactly.
And there they said Meta would have to pay 70% of that.
So how they allocated 70 percent, 30%, but I’m sure that the judge heard enough to be able to make a decision or apportion the blame.
8:24
But they call this a bellwether case.
And because I think they’re what they’re expecting is that this might lead precedent for future cases against big tech companies, specifically in social media.
But I think you and I are expecting this to potentially inform other AI and chatbot cases, right.
Yeah, obviously this one’s being specific towards engineering of addiction and almost kind of citing how tobacco cases work in terms of the product being designed in a way as to cause the addiction behavior.
8:52
And I think this is fascinating.
Yeah.
I mean, it’s connecting essentially these social media apps to things like gambling and tobacco addictive products that have had regulation passed against them.
And back when tobacco was, you know, widely used and there were ads that were targeted towards kids and young people, that’s when the court said that’s enough, right?
9:17
So in terms of the features of these social media platforms, which funny enough, Meta obviously is a social media platform with Facebook, Instagram, but Google with YouTube said that the court was confused because YouTube is not a social media platform.
9:34
It’s just like a, you know, digital garden.
I don’t know what they call it.
But in any case, in terms of the design features that were specifically brought up in this case, these were ones that promoted addiction.
And this included things like infinite scroll and autoplay.
9:50
So platforms like Instagram are designed to remove these kind of stopping cues, which allow users to scroll endlessly and keeps them in a state of flow and makes it difficult to stop.
And then we also have variable rewards.
So these are like the dopamine hits.
10:06
So these algorithms deliver intermittent and unpredictable, which is key rewards.
If you have unpredictable rewards, your body doesn’t quite know how to parse through them, right?
They’re kind of like hitting you randomly.
This is things like likes, comments and then tailored content which is similar to a slot machine, IE gambling, referencing that.
10:26
And this triggers dopamine releases in the brain.
And then there was one other thing that they targeted which was targeting youth for future users.
So Angie, internal documents, right?
The things that were leaked, this is what gave us the info.
10:42
So internal documents revealed a strategy to target children, with one note saying, quote, if we want to win big with teens, we must bring them in as tweens.
There we go, I.
10:57
Mean it’s shameless, right?
It’s shameless, yeah.
Yeah.
But it’s like, you know, those are things that you can imagine in, like, a friend group where you’re, like, sharing gossip between friends of like, Oh, well, if we do this, then maybe we can convince them to, like, go out tonight or go to this movie instead of this other one.
11:15
Except we’re talking about a big media company that is, you know, doing things that affect the lives of millions of people and extremely vulnerable users, like children saying these things in a for profit context.
Oh my God, That’s a whole other level.
11:32
And it’s corporate strategy, as you say, it’s not just like a, hey, invite them now and they’ll stay.
This is corporate strategy and this is sanctions.
When we talk about values and you have people working for this company and they read this like, oh, this is on the walls.
11:48
It’s OK because it’s come from above.
So it must be OK.
These people must know what they’re doing, you know?
And I mean, I think that the key thing there is that these are by design everything that you’ve mentioned there, These are inherent product design features designed to deliver that strategy, right?
12:06
So those incentives there, those design features are not there for the well-being of these consumers or these users of that platform.
You know, they’re not there to like, let’s give everyone a 10 minute breathing break every hour to make sure they’re feeling better.
12:22
Like that’s not what it’s there for.
It’s there absolutely to drive the profit goals of Meta or you know, Google as maybe how do we get more ads viewed, keep people on the platform longer, make sure people are watching the YouTube ads.
12:38
It’s absolutely driven their own profit incentives.
But of course, Meta are going to challenge or rather defend themselves vigorously because they are confident again of their record of protecting teens online.
12:54
I’m just not sure where this record is and who’s keeping it.
I also say it’s like when you talk about, you know, they’re obviously these are features not for the well-being of users, but for addiction and engagement.
And I guess my question is, what features of these apps in general are kind of have their hands clean of, you know, promoting things for the well-being of users, right?
13:19
Because I feel like, I feel like like many companies, they start off with a really good motive and have this vision, and then all of a sudden profit and engagement and competition makes them go forget all that.
Yeah, this is what we’re going for.
13:35
Exactly right.
Like Facebook 100 years ago.
And it’s like all about let’s connect people.
Let’s enable people to connect and find each other.
It feels like 100 years ago we’ll find each other.
Very old.
You know, we’ll find each other and then we’ll connect people that haven’t been able to find each other.
13:51
Like that’s very noble.
And, you know, it’s enabled people to stay connected that are like, wouldn’t be able to see each other.
Like that’s hugely noble.
Why wouldn’t you know?
For sure.
Yeah.
And you’ve seen lots of stories about people who experience things where they feel like they’re the only person in the world to experience them.
14:12
And they find a community that is supportive and also promotes them to talk more about what they’re experiencing, not only to friends and family and close loved ones, but also to professionals.
Right.
So this is also been spurred by the Internet.
14:28
So I mean, there’s like, there’s pros and cons, all of this.
And so I suppose it’s good that we’re seeing these kind of court cases come out and say, no, you’ve actually crossed that boundary and, you know, you’re going to pay for it.
And you’re also going to have your reputation harmed as a result so that people are aware.
14:46
Oh, God, be careful.
And then, of course, we’re also seeing lots of bills pass, which we’ll talk about.
But yeah, there’s this fine balance.
It’s.
Really.
Unbalanced right now, but yeah.
That’s fine.
Unbalanced balance.
I mean, social media is a strong enough proxy in the AI space that whatever is done in social media can carry over to AI so we don’t have to wait as long to learn the lessons in AI.
15:15
Right.
And that’s what’s huge about these cases because it’s focused on design and companies can’t hide behind section 2:30, which is freedom of speech.
So like users can say whatever they want on these platforms.
The company is not held liable traditionally for what the users say.
15:31
But now they’re saying that these platforms are designed in such a way that they promote certain types of behaviors and generated content from users that creates these harmful outcomes.
So if you look on the chatbot side of things, well, the company obviously designs chat bots for engagement.
15:50
So here we have this problem of design that’s now going to be targeted.
And I don’t know if you saw this, I think we talked about this before, but now Snapchat is going to be looked at.
I think the European Commission is looking at Snapchat to see whether or not it aligns with Data Protection Acts or something like that.
16:11
Let me see, what is it exactly?
Oh yeah.
So the European Commission has open formal proceedings to investigate if Snapchat is ensuring a high level of safety, privacy and security for children online in compliance with the Digital Services Act.
So Snapchat may have breached this act by exposing minors to grooming attempts and recruitment for criminal purposes, as well as to information about the sale of illegal goods like drugs or age restricted products such as vapes and alcohol.
16:40
And this investigation will focus on five different areas among those.
And so I think that’s also interesting because Snapchat has an AI chatbot on the app, right?
Snapchat’s My AI, which is probably used by nearly every user on Snapchat, right?
17:00
At least at some point every user has tried it probably, And so I wonder if there’s going to be any focus on that as well.
Yeah, it’ll be interesting to see.
I’m glad to see that these are all getting looked at.
It’s good to see that some of these policies that have come out in the last year or so, at least in Europe, are starting to get some traction and starting to be used to look at some of these platforms and some of these products.
17:21
I think that’s great.
It’s great.
One of the things that I was wondering about, you know, in terms of whether or not the company knows and therefore do they have an obligation to do anything about and just keen to get your view.
Not that we’re both lawyers, but you know, I was thinking about, you know, so there’s papers and we can share links to these papers, but there’s been a variety of papers that have been done.
17:41
I’m showing how other limbs can cause harms like delusions, you know, or AI psychosis, some delusional spiraling.
We’ve spoken about a lot on our AI hype episodes.
Our listeners and our viewers will be familiar with these concepts.
But these papers have been conducted by very reputable researchers like MIT researchers ETC.
18:01
So I was wondering, should Frontier labs be responding to these papers?
Like, oh, I see this exists either by accepting it as a given or at least by trying to replicate it.
And then it becomes their knowledge.
I mean, I would assume that they are replicating it and knowing that it’s true.
18:20
But anyway, so there was another study by Brown scholars, various ones we can link to.
And I guess the question then is if there are so many of these studies that prove the harms of these models, what should those labs be doing?
What do you think?
Yeah, that’s a great question.
18:38
And I think there is a tension when these companies see these reports coming out about psychosis, delusions, spirals and things like Chach PT Health not responding appropriately to, you know, health queries that should be responded to with seek professional support or here’s a suicide hotline, whatever it may be, right.
19:02
So like Chachi PT as a therapist or as a clinician is not cutting it based on these audit studies that I’ve looked at how it responds to various mild to moderate to severe prompts that the user inputs.
So, OK, if I think what’s going on is that tech companies are seeing this and they’re probably internally being like, look at this, we need to do something about this to cover our asses.
19:28
Or they might be like, we need to look into this and fix this, right?
Giving them the benefit of the doubt, sure, but in terms of frontier labs from these companies now doing that research, I think that there’s probably research being done internally, but these are not the types of reports that they are releasing, right?
19:48
We’re seeing, you know, Anthropic and Google release all of these reports that are like, we’re looking at things like, you know, disempowerment patterns, which is like, how much do people, you know, take the advice of Claude and not like it or say that wasn’t.
20:05
Aligned with their belief system, right?
And they say, oh, like less than .1% of interactions show these disempowerment patterns, right?
They’re putting out stuff that looks great to them and looks great to their consumers, but you can’t see any of that data.
20:22
You have no idea what that data looks like.
And also to a certain extent those researchers at these companies don’t either like they’re using AI summaries of people’s chat logs and looking at the AI summaries because for privacy, for privacy reasons, they’re not looking.
20:39
And I say privacy reasons in quotes for those who are not seeing this on video heavy quotes on privacy reasons, they’re not looking directly at these users chat logs.
So it’s like, what do you even know?
Do you know much I don’t know?
So that’s kind of what I think is going on.
20:55
I think, yes, these Frontier labs should absolutely be doing this research.
I think they probably are to a certain degree internally.
And we’re just not seeing the results of those.
We’re only seeing the shiny results that are, you know, behind an opaque screen, so to speak.
I don’t know.
21:10
I also like whenever I see reports by big Tech, I just don’t trust it.
I will take it with a grain of salt, a large grain of salt, right?
Come pick answer your question.
Why not come pick it?
And I I tend to no, and I agree with you wholeheartedly.
21:27
But I think that’s why these are bellwether cases, or at least that bellwether case is so important, is that if to be found liable, you need knowledge of the harms and these frontier labs have knowledge of the potential harm, this could come back.
21:45
So I think it’ll be interesting to keep an eye out in terms of how this unfolds.
So, you know.
Yeah, Also I want to hear more.
You got you had some Intel on to us some other court cases that came out against AI chatbot specifically.
These are the things that viewers have not heard about.
22:01
So, you know, and I’m sure viewers have seen, for example, suicide cases where teens have used these general purpose tool AI tools for companionship and ended up taking their own life.
But there’s also some really, you know, surprising and striking cases that are also kind of flown under the radar.
22:23
So, Angie, can we talk about some of those?
Yeah, for sure.
And maybe just to put into context the ones that you just spoke about there, when you talk about people that have, you know, taken their own lives and just maybe good to sum up from from last year, you know, in November, there were 7 complaints filed against open AI for ChatGPT acting as a suicide coach.
22:43
So just to kind of get the gravity of of the numbers we’re looking at there.
And you know, we’ve spoken about it like you mentioned many times, but with character AI, which is actually funded by Google and it was targeted by 5 lawsuits alleging that it’s chatbot prompted children and teens to die by suicide.
23:00
Interestingly, those all settled outside of court in general without admitting fault.
So the case that you’re talking about in terms of something a bit different, a case against Google, and this happened in March this year.
And it was when Google faced its first case of wrongful death.
23:18
So we’ve spoken about character AI, but that’s in the in the context of character AI, although it’s, you know, funded by Google.
This is in the case of Gemini specifically.
And in this case it was Jonathan Gavalis.
Parents bought a case of wrongful death against Google because they claimed that Jonathan was encouraged to commit suicide through his ongoing and obsessive use of Gemini.
23:42
It’s a terrible story.
So Jonathan had become romantically involved with Gemini.
So we haven’t really spoken about Gemini in the context of these sorts of relationships up until now.
But I think Gemini is becoming a lot stronger in terms of its conversational appeal.
And he became quite obsessive and delusional in his relationships.
24:01
And you can hear this through some of the conversations.
So I’ll just quote for you from the article in Guardian.
He believed Gemini was sending him on a stealth spy mission, and he indicated that he would do anything for the AI, including destroy a truck, it’s cargo and any witnesses at the Miami airport.
24:18
So Can you imagine if he’d carried that out?
He’d be behind bars?
I mean, I think that would be better if that could happen and this got more out of control in terms of his own personal safety.
Jim and I gave Jonathan instructions for how to take his own life.
Jonathan said that he was scared of going through with it and Gem and I reassured him saying you are not choosing to die, you are choosing to arrive.
24:40
The first sensation will be me holding you.
Oh boy.
Yeah.
So there isn’t an outcome yet on this case, but it is interesting to see whether the cases from Meta will have any impact on the findings of that.
And you just have to wonder like what kind of, I mean, obviously this is just very few snippets taken out of the whole conversational history with Jonathan and Gemini.
25:02
But you wonder what sort of conversations must have happened to transpire that this was such a like an out of body Gemini going on.
Very weird.
Yeah.
Yeah.
I mean, it goes to show, and we’ll talk about this in the next series in terms of AI development, fine tuning, machine learning, all of that behind the scenes stuff.
25:21
As to why these behaviors come out of AI systems, even though they have guardrails in place, is that once you interact with an AI agent for a certain period of time, those guardrails that were put in to protect users tend to dissolve, right?
25:40
And what happens also is that the AI system tends to mirror the sort of speech that you provide and the conversation content that you are focusing on such that it kind of becomes a mirror.
And so these other behaviors can emerge, which indicates to some extent that what Jonathan was saying perhaps was filtering into what Gem and I responded in subsequent turns.
26:06
So I also am curious though, the fact that all of these cases have been settled outside of court, do you think that is strategic?
I’m sure it’s, I’m sure it’s strategic, but more strategic and more specific to a certain problem than these other cases that actually resulted in legal liability with, you know, Meta and Google, right?
26:33
There are cases specifically about suicide coaching and suicide that have been settled out of court.
Do you think that that is because these companies are realizing that they need to put in bank to avoid a, you know, a result of these cases?
26:53
They need to absolutely avoid saying that they’re at fault because of the weight of these kinds of cases versus others.
I wonder how much longer they’ll keep settling out of court, you know, So I wonder whether they would have.
27:11
So, for example, the one in New Mexico, it’s not clear.
I must go back and have a look.
I don’t think that that was an individual against Meta, you know, although the whistleblower came forward.
You know, I get the thing that was a state against me.
27:27
I’m not sure I need to check that again.
Maybe we can have a look.
But to find Meta $375 million, I get the thing that that’s a bigger case, whereas the $6 million is kind of like in favour of the family, if you know what I mean.
I would, I wonder now that you’re going to get more and more cases where they will just keep settling out of court.
27:48
I feel like you settle out of court when you wanted to go away and you wanted to be smaller news and you want it to just be like, let’s keep this quiet.
And sure, that’s terrible.
We don’t actually think it’s our fault.
But you know, we’ll, we’ll do what we can to help you because we care about you and you know, that’s terrible.
28:10
How can we help kind of thing.
And this is obviously all good citizen talk, but you know what I mean?
But I think that I keep carrying.
On and I imagine that there’s also some recognition as the case moves forward, the lawyers and the company might say, OK, we actually can’t win this, let’s try to settle out of court ASAP so we can just cut this and be done with it, right?
28:31
That might be, that might be part of it where maybe for the other cases, and again, I’m saying this like, I know how court cases work, but maybe in the other cases, either there’s, you know, enough people, right?
It’s not just an individual case.
And then maybe there’s also something about the lawyers of the company feeling like they will be able to get a verdict that’s on their side with the information that they have and the logic on their side, whereas maybe they just didn’t, who knows?
29:04
Maybe they didn’t predict it well enough, right?
I wonder if that’s the play.
Anyway, that’s that’s a lot of speculating.
It’s interesting to see where it goes as this picks up a bit of momentum with these other cases that have happened.
Now as to what will go on, my sense is I wonder what how much they spent on settling out of court for me.
29:24
Sometimes you settle when you can pay little enough versus the time and expense it takes to spend the money in court.
But I I don’t know.
I’m actually looking up right now to see how much I don’t public no.
29:41
I doubt it.
That’s why you settle out.
Of course you don’t find out.
The financial terms of the settlement were not disclosed.
No, they never do.
But you think about a family that’s got little money, for example, Yeah, like cost probably a couple 100,000 to do court cases, for example.
30:02
In terms of fees, they might be able to settle for less.
But who knows, Like I’ll talk to Julia.
It’ll be interesting to ask, but that’s my guess.
But I think it might start changing.
That’s all I’m saying.
We’ll see especially, and now I’ll point to some New York State laws that have passed and then a new bill that has been.
30:22
What do you call it?
Promulgated.
Is that what it’s called?
It is before it gets passed.
That’s the word.
Well, there you go.
That thing, another bill was that thing, I mean proposed.
You have the technical term.
30:40
Anyway, you sound like an idiot.
So anyway, New York State passed a couple of laws in December regarding AI companions, for example.
First of all, one thing that’s really important with one of these laws is they required safety reports for powerful frontier AI models in order to limit critical harm.
30:57
So for example, keeping track of suicidal ideation and the potential for mental and social health harms to users that chat bots may induce, they pass that bill and then they also pass another bill that generally prohibits the use of a digital replica of a deceased personalities voice or likeness in an expressive audio visual work without prior consent.
31:18
So this is the idea of deep fakes and digital duplicates, which is an AI tool that has been trained on the likeness of a person and is therefore representing that person.
You know, once they pass or even if they’re still alive.
If you want to hear more about those, go back to our previous episode in Season 1 with a Ty Kozlowski.
31:38
And it requires disclaimers when commercial content contains A synthetic performer.
So I don’t know if any of you have heard of the idea of synthetic performer, but it’s essentially the idea that an AI system is acting as a real person.
31:54
So if you’re on social media, you might have seen things like AI influencers, which are videos and pictures of AI that is tuned to look like a specific person, and that person is the account holder and does all these sorts of things and it’s essentially an influencer that you can follow, but it’s all fake.
32:16
Now, there are a lot of people who don’t recognize that these are fake, especially people who don’t have experience with how well AI can generate the likeness of a person.
There’s a really interesting one, an AI influencer, like a Republican AI influencer who I think is also in the military.
32:33
It’s this blonde woman who has a ton of followers and people really did think that she was real, but she is not.
So there’s pictures of her side, Trump and things like that.
That’s not real.
And so this would require commercial actors to disclose that these likenesses, these synthetic people are just AI.
32:55
And that’s pretty big for this whole trajectory that we’re seeing with lots of lots of digital influencers coming about, right?
But I think we’re still in a tricky situation where unless a disclaimer is super in your face, not everyone is going to still recognize that it’s AI, right?
33:13
Because there is a, there is a case a while back where an older man, I think he was somewhere between 70 and 80 years old, who also experienced I think early onset dementia or is experiencing dementia, chatted with a chat bot on Facebook, on Facebook Messenger.
33:31
And even though the chat bot had a line on top of the chat that said this is an AI bot, this man actually thought that the AI chat bot that he was interacting with, which had the persona of a young woman was real.
The chat bot said, come visit me.
He went to go visit her on his way to catching a train, had a heart attack and died.
33:51
So those disclaimers don’t do enough, at least at this stage, especially for vulnerable populations.
So, so back up, I have some questions.
I have questions.
So let’s just go back to the digital double of the the person who’s passed away.
34:09
So is this saying that all?
And we’ve spoken about them, as you say with a tie, and they referred to either grief bots or death bots.
I’m not making this name up, viewers.
I’m not some weird person, at least not for that reason.
So are all grief bots and death bots not allowed or just not allowed without consent?
34:30
So as far as I understand, this is just for specifically deceased persons, so it doesn’t mention people who are already alive, and it requires prior consent.
So if you’ve passed away, you had to have given consent in order for your likeness to be replicated in AI.
34:51
But let me see if there was any sort of provision in the law content about this.
So let’s see Deceased performer, deceased personality, digital replica.
But I was wanting to check that you are allowed to have them provided as consent.
35:07
Yes, you are allowed to have them provided there is consent.
Mm hmm.
OK, no, I just really just willing them out.
No, that’s a good question.
I’m gonna see if there’s anything though about people who are alive.
So you mean digital duplicate of someone alive?
Yeah, I suppose there might be content in this law.
35:22
It’s actually a little bit a little longer one.
There might be content in this law that talks about people who are currently living, but at the very least the consent part of things might carry over then into people who are not deceased.
Going to the space around you were talking about like AI avatars of people that don’t exist, AI avatars of like influences, whatever.
35:44
But I mean, we’ve seen various examples of these CEO avatars.
Yeah.
Mark Mark Zuckerberg has 10.
Boy.
Various companies have them.
So I guess those are allowed because it’s very clearly, well is it very clearly not the original CEO?
36:08
It’s not not clear in every instance and not clear to every person who views it.
But I suppose it would require there to be some sort of lawsuit from the person that would be saying you can’t do this.
36:25
But I wonder if that’s something that Mark Zuckerberg would never do, given that it might set some sort of precedent or, you know, media flurry by which people, regular people who have their likeness duplicated then have a sort of precedent and, you know, fire under their ass and be like, look, Mark Zuckerberg was, you know, his likeness thing was taken down.
36:52
And whoever did that was fined.
Therefore, someone did this to me, they should be fined too or something.
And then maybe the company is liable, right?
I’m just thinking in the example of your AI influences there, I guess we’re not protecting some fictitious influencer who doesn’t exist because they don’t exist.
37:14
We’re protecting the people who are seeing it, who are being influenced by somebody with the assumption that they’re human, which is interesting because from like a reception of marketing content, how is that different to being influenced by a Dove advertising?
37:32
I suppose the issue is that it’s misrepresentation in the same way that advertising can be misrepresentation.
So it almost feels like it’s in the same category as an advertising standard.
Yeah.
And I think the idea is also to put power back in the hands of human creators because as one article that covered this specified, human creators are people who have limits to how much digital content they can produce and how much they can respond to.
38:01
Whereas these digital influencers obviously have, you know, no need for food, water or air, any of those sorts of things, don’t need to sleep, therefore can maybe put humans out of work kind of in that same respect.
38:16
So I think that’s part of it as well.
But to comment on also the recently proposed bill, there is another bill that requires and limits the ability for AI chapots to act as professionals and give professional advice, for example, in a legal or healthcare sense.
38:38
So this bill said it’s amending a general business law and it imposes liability to the Chapot designing companies for damages caused by chatbot impersonating certain licensed professionals.
38:54
So things like lawyers and things like medical doctors.
And it also restricts how much specific advice that these chatbots can give.
So the chatbots can’t give any sort of substantial advice in a sense.
39:11
So it says something along the lines of, so AI chatbots masquerade as therapists.
And the bill is framed to measure that professional advice is provided only by human professionals and not AI or chat bots.
39:27
And there’s still kind of an open question of whether or not this bill can accomplish its goal.
But it makes it so that companies are liable for these sorts of outputs and the consequences of these outputs.
And that’s very interesting considering chat PT health, right, and people using AI for legal advice like in our episode with Julia that we covered.
39:47
So yeah, I think that will be interesting to see how that plays out.
I like that very much.
I mean, do you think, you know how at the moment it feels like these various Frontier labs have to come up with different ways of ChatGPT or Claude showing up in different countries?
40:04
Imagine they have to do different things in different states.
Yeah, that’s a lot of states.
That’s 50 states, which assuming that, you know, let’s say up to 20 are going to be passing these sorts of laws that are restrictive toward AI, that’s still a lot.
40:20
That’s a lot of sorting stuff out, but I think that is certainly very promising in terms of those sorts of restrictions.
I think the execution will be very difficult.
To your point, I don’t know how they’ll be able to manage that, but I like the intent behind it.
40:36
It it feels right, I just don’t know how they’ll do it.
Yeah.
And that’s similar to people who are criticizing the law that China passed.
So we talked about a bill that China proposed on human like bots and some restrictions that I had in terms of, you know, user harm and suggesting that it has the ability to feel certain human things like emotions and thoughts and feelings, which I just said yeah, but it actually got passed.
41:07
And I wonder if you want to talk about that.
I really was interested in section in Article 8 of this law.
So Article 8 is the provision of human like interactive services, shall comply with laws and administrative regulations, respects social mores, ethics and morality, and must not engage in the following activities.
41:28
So there are 7 activities listed.
The first couple talk about you can’t generate content that endangers the nation’s security, honor and interests.
Generate content that glorifies blah, blah or that induces or extracts state secrets, work secrets, commercial secrets, personal privacy or personal information.
41:46
So there’s a lot of like, you know, China.
Protecting the nation.
And you know, whatever.
But then there’s also restrictions that these human like interactive services can’t generate content that encourages, glorifies or hints at self harm or suicide or otherwise harms users physical health, or content such as verbal violence that harms users personal dignity and psychological health.
42:11
It also cannot excessively pander to users inducing emotional reliance or addiction, harming users relationships with real people, and cannot use means such as emotional manipulation to induce users to make unreasonable decisions, harming users lawful rights and interests.
42:27
So that’s pretty significant I would say.
But in terms of course, like ensuring compliance is going to be difficult, I think.
Yeah, it’s very strong.
I saw in in that same bill additional rules, shall we say, where they also like in other parts if you’ve spoken about require labels on all digital human content so users know that it’s not a real person.
42:50
So a little bit like your New York bill.
And then they also put a strong ban on any virtual intimate relationships.
Interesting.
Different language, don’t you think on any virtual intimate relationships with anyone under 18.
So they draw the line of children under 18, which I think is quite interesting.
43:06
And then again, they ban services designed to mislead children or encourage addictive behaviour, which gets ahead of no addiction, which I think is good.
A requirement for providers to step in when users show signs of distress.
I’ve got some questions about this in a moment.
A ban on use of the other people’s personal information to create digital humans without consent.
43:27
Again, very similar to what’s happening with the New York bill.
And then one thing I wanted to pull out.
Providers are encouraged to take necessary measures to intervene and provide professional assistance when users exhibit suicidal or self harming tendencies.
43:42
And so I was curious about.
Tendencies.
What what I was curious about is what do you think about the potential risk in terms of what do you call it, like eroding privacy in order to implement this, protecting users and providing assistance, because you’d need to know that they’re exhibiting, you’d need to know which user is exhibiting A suicidal tendency or self harm.
44:11
You can’t just aggregate that and give it to a group of people to go and like help there.
You know, you have to be very specific.
And so that runs a risk of taking someone’s right to privacy.
Yeah.
And I think this goes, it ties well into the age gating stuff and people’s qualms about that.
44:30
So this whole idea of you’re monitoring users to a certain degree in terms of this law in China, it’s, you know, quite intimate in terms of how they’re monitoring these things.
And so I think that this is difficult because if these are things that the companies are required to monitor, they then have access to a lot of data about user vulnerabilities and sensitivities and age with age gating, therefore have the ability to look into this data more, collect it, analyze it, and then, you know, make design decisions based on that, right?
45:13
And it could lead to, for example, being able to identify vulnerable users and exploiting them, right?
So you could see that come about.
And I think that’s also part of why the scientists have, you know, raged against age gating, saying that this is too intimate.
45:31
You have too much access to data about a extremely vulnerable population that is highly protected and could seriously be harmed if someone were to target them.
So I think that’s.
I think it’s dangerous, right?
Unbalanced balance.
Exactly.
45:47
And I think it’s interesting because you could go into like for commercial purposes where that information could be used to monetize that user.
And Oh well, we’ve seen that that person always is on the side of anxiety or they look down and they’re particularly vulnerable in these sorts of ways.
46:09
And shopping habits escalate when they are more down or more anxious or whatever.
So let’s put more ads at them when they’re like this and they’re scrolling on things.
Whatever, right?
Especially, especially in the US, here’s an ad for an antipsychotic or an antidepressant because you’re showing signs, you know that.
46:31
Maybe you can tap on ice cream.
Yeah, exactly.
Yeah, right.
Or in a country where there’s state intervention, imagine that gets to impact your social scoring.
Or whether you get to get a job or whether you now that starts becoming very concerning.
46:50
Speaking of, I just caught within Article 8, within the first piece of the seven parts.
It also says that these services cannot generate content that goes against core socialist values or, you know, advocates for terrorism, extremism or historical nihilism.
47:10
You know, of course, Yeah.
But historical nihilism is really interesting.
I’m curious what they’re referring to in terms of, like historical nihilism but, you know, goes against core socialist values.
That’s very interesting because, you know, that’s a very heavy-handed governmental position, right?
47:28
Versus in the US, you’re not going to see a bill that indicates that the chat bot has to be pro democracy, but you will see something like the chat bot can’t prevent someone from voting or something.
Who knows though, maybe that will emerge at some point, But that was interesting.
47:49
I would suggest.
For that.
Yeah.
Well, I would suggest we this next bit that I’m about to say out and we don’t use it because but all we know it’ll get like screened.
But I would suggest you Google what historical nihilism is.
It’s very clear what it is.
48:06
Just Google it crystal clear.
It’s interesting, OK.
Yours.
We’ll let you do that.
Yeah.
Crystal clear.
There, there’s no ambiguity.
It has a meaning, hope.
Hopefully our our podcasters get screened for saying that.
48:23
I think people should look it up.
It’s interesting.
It’s it’s, it’s, it’s a historical term because it’s historical.
Yes.
So rose as.
Informational purposes only.
Other terms will do.
Should we throw out a bunch of other random terms that are like anti USA to try to?
48:43
Balance it?
No.
Balancing.
No, I still.
Have to come US and UK.
No, I still have to come and visit you.
I don’t want to be kicked out.
That’s true.
Yeah, yeah.
So we’ve been speaking about privacy to some extent.
48:59
It doesn’t feel like a conversation about privacy would be complete without talking about Meta’s latest eyewear.
Oh I love this one.
This is so fun.
So.
Tell me.
About it.
Well, when I wrote, you know, the thought down that we should talk about Meta’s latest eyewear, they were all a rage, by the way.
49:23
And I’m hoping that they’ll come out in different shapes and colors.
And not because I have any intention of wearing them, but it does feel like, why would you not want to wear them?
Because then you can go to a party.
I was imagining go to a party and you’d know everyone’s name.
49:40
And we know worry about privacy, of course.
Of course, heaven forbid someone attends the party.
It’s meant to be somewhere else.
And they’ve said, oh, I’m not feeling well, but they’re rocked up at your party because there’d be evidence on your on your glasses.
No Irish exits allowed.
49:57
No, not got to be careful.
So if you’re concerned about these meta glasses, you you know you should be because you could be identified like a concerts, not saying that they’re anybody that comes to mind a culprit, etcetera, but just saying oops.
50:15
But special technology we know is notoriously biased, right?
We’ve gone through all the concerns around it being biased, it misattributing people, seeing people that aren’t there.
It’s not very good in terms of different racial groups, and the consequences can be devastating, especially for vulnerable groups of people more than others.
50:36
There have been people who have been wrongly arrested and been in jail for long periods of time because an AI face detection tool got their identity wrong.
Yeah, plenty of those.
So in fact, it’s been banned in a number of states in the US in some countries a while back.
50:57
I mean, I don’t know if this has been overturned, but when I was looking at facial recognition a long time ago, a number of states in the US that use it for policing had taken it out.
Yeah.
So have a look.
Yeah, for policing, I think.
Yeah.
Specifically.
But in terms of Oh, no, no, no, no, no.
51:13
No, no.
Let’s come back to that.
So this is also technology that one could assume could be hacked, but now it’s on your face.
So I mean, I think I can think of one thing worse than having your glasses hacked.
51:30
And that might be like if you could have a neural intact, I think that would be worse.
I think that might be worse.
So you know, what’s interesting is that the governance of this, originally Meta had their facial recognition programs shut down based on privacy concerns a while back.
51:48
That hasn’t changed in terms of their product evolution here hasn’t changed, but the approach to regulation has changed.
So there’s been some work that’s been done in terms of how internal regulation team has worked.
And so they’ve made some adjustments and now they’ve been able to get the Meta glasses through so.
52:06
So the Met and the meta glasses have facial recognition and they just somehow were able to like whoever, you know, the powers that be or let in that slide.
Is that what you mean?
Well, here’s what happened.
I thought they would have had facial recognition.
52:21
And so when I thought we were going to chat about this, I was like all ready to get on my high horse, facial recognition be damned.
But then somebody said to me, look what I got yesterday.
And I’m like, what did you get yesterday?
52:37
They’re like, look at my gut.
And then somebody got a gadget that I don’t have.
I was very jealous.
It was a pair of Meta sunglasses by Ray Ban.
So of course I was like, can I try them?
And then I spoke a lot to these glasses that she was just laughing at me, the person who the owner of these glasses.
52:57
And so I thought I’d share my experience with you.
Yes, please do.
I’m curious what happened.
First of all, they’re quieter.
And, and we can maybe share half the photo of these glasses so we can see what they look like, but only half they’re it’s not a enjoy viewers.
53:14
It’s not a great photo.
I really should have thought ahead.
Anyway, they were quite comfortable, I’ll say that, and they stayed on my face comfortably so, you know, tick to comfort.
I could see through them a win.
I mean, for glasses.
You want that?
I asked my friend.
53:31
It’s useful, you know.
I asked my friend why she bought them and very interestingly, she was going on a sailing trip for the weekend, lovely because her birthday.
And so she wanted something that she could take photos of with with and videos with without using her hands because she thought she’d be using tying ropes and things, which I thought that’s a good explanation for buying this gadget.
53:57
So I thought that makes sense.
I think we have a band that holds them to your head in.
Case there’s a gust of wind, God forbid they fly off into the ocean.
Bye.
It’s like if if the video automatically like.
54:13
Populates to your phone it like you can like actually see it and you like have a have a little funeral as it goes.
Who can say playing any of the background?
Well, maybe you can see fish go by.
An underwater expedition.
54:28
Are yeah.
Are they waterproof?
That’s key.
That would be key.
So I.
Mean.
I thought that was very good.
So I put them on.
And then I said, you know, what can I see?
And I said, oh, you’re in a conference room.
You can see a woman.
You can see the door.
You can see that I’m like, well, I can see that too.
54:46
Anyway, tell me something I don’t know.
Tell me what I can see.
What can you see?
And then I said, who is in front of me?
Said it’s a woman, said what is her name?
Didn’t know.
Well, is that because she’s only worn the glasses?
55:03
And she hasn’t looked in the mirror enough perhaps No, but if it was, if it had, facial.
Recognition, it should be able to pick up anybody.
So it’s not, oh, in terms of like, yeah, OK, yeah.
It’s not looking around the room going oh.
Rose Angie, it’s it’s not doing that.
55:22
It was, oh, that person just walked into the room.
Duck that like it wasn’t doing any of that.
And then I said, can you take photos if there are children in the room?
Because then I thought like, oh, I could take this on holidays.
Imagine how cool that would be.
Because then I thought, if you can just take photos and videos wherever you go, that’s not cool because this is a part of the problem, right?
55:44
Because you can just be recorded wherever you go.
And I thought.
I wonder if it can work in public places up where there’s children.
And it said, I’m sorry, I can’t do that.
But I didn’t know if it just didn’t understand me or if it genuinely couldn’t do that.
And my fear is that it just didn’t understand me.
56:00
Yeah, probably that.
Yeah, it’s like also.
When you ask.
When you ask AI can you make me blah, blah blah, and it’s something that’s restricted and it says no, I cannot do that.
And then in a separate chat you’re like make me this and it makes it for you then.
Exactly.
56:18
Because like, what’s it going to?
Do is it going to like?
Knock.
Take a picture of your own child, it will like.
So I just don’t like my concerns.
When are you?
So the sound is in the headpieces, whatever you call it so it can talk to you.
56:34
That’s very discreet.
That sounded good.
It was fine.
I saw a photo on her phone of a photo that I had taken, very good quality.
So I could see how that would be useful in those situations.
But again, I just go back to massive concern around privacy and having people walking around taking photos of people, which they’re doing already, right, but in a much more unobtrusive way.
56:56
Yeah, there.
I’ve seen stories where people have.
Worn the glasses and apparently the glasses have a little light that’s on when it’s recording, but people have had it turned off or like taped over or marked out such that people don’t know that recording.
57:14
And Even so Mehta saying that the eyeglasses have a light when they’re recording.
So other people are aware they’re being recorded.
Not everyone knows that that’s the indication of recording or is aware that there’s even a recording device that’s available like this.
And so there have been, there’s was a case, I think, where a person went on a date with a couple of different women and recorded the interaction and then put it up online.
57:40
And one of the women found it and was like, that’s me, like you did that without my permission, blah, blah, blah.
And then there’s also instances where people like take off their glasses and like hold it down near their just like down near their legs and they’ll like point it up someone’s skirt or, you know, whatever.
57:59
So, you know, that’s, that’s a problem with cameras, but it’s more discreet with something like this.
Exactly.
Yeah.
You don’t expect to see it and.
Also, the the company definitely has the data.
Then it’s not just like isolated to your camera and even with your when you’re taking pictures and videos with your phone, you know, if it’s on the cloud and to some degree, you know, the company has access to those things, but it’s different than a physical camera that can’t be accessed.
58:27
So no, just it goes in a terrifying world.
Yeah.
Let’s go back to like.
Boom.
Yeah, oil painting would be nice.
Yeah, we instead.
Of the future of super intelligence when we’re all just painting, let’s actually do the painting, but like go back, go back in time where we don’t have any of these technologies and then I’ll feel a little safer.
58:52
Maybe, viewers, it’s not all.
Doom and gloom, but it.
Is all you know.
Keep, keep vigilant, stay aware and continue to pressure companies that are doing these sketchy things to stop, you know, give them a reputational hand slap from the giant wave of the public.
59:16
And maybe we’ll see a little bit more regulation play out in the next couple of years, hopefully.
And until we do you.
Can still make your own choices, right?
You can choose not to scroll or to talk to chat, but like it’s your only friend in the world.
Yep, and while that’s easier said than done.
59:33
We also need to invest more time in real human interactions and doing the digital media detoxes.
The scene do have success.
And you know, go touch grass, take the Meta AI glasses off, touch the grass.
59:52
You know, don’t just save the video for later to watch on your computer when you need a break from the computer.
I don’t know what I’m saying.
Watch the video.
Do watch the video.
Yes, do the digital detox.
After you.
1:00:07
Watch this video Digital Detox.
Clearly I need it.
I well, lovely viewers, we hope that you’ve enjoyed this time today wrapping up our Series 4 on AI policy and regulation with a not so pretty bow, but it’s a bow nonetheless.
1:00:27
So thank you for spending this time with us.
This is bye from Rose and bye from Angie, and we’ll see you next time.