Transcripts are auto-generated and may contain errors.
0:00
Hey, I’m Rose.
And I’m Angy.
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.
Sometimes we agree, sometimes we spiral, but we always go deep.
0:15
Angy: Hello, Dr. Guingrich.
Rose: Oh, thanks. :3
For those of you who don’t know, I actually graduated with my PHD! I defended my dissertation on May 1st and yay, the result was good.
0:35
I am now Dr. Guingrich.
Angy: And it was an honor to be there and watch from afar.
But it was really great to watch Rose and now we have two doctors in the house!
Rose: And Angy, you also graduated, didn’t you?
Angy: I did.
I got to be in sunny Cambridge – and that’s worth mentioning because it isn’t always sunny – on the lawns of the Senate, and got to graduate from my AI ethics degree, my master’s in AI ethics.
0:59
Rose: And so this was a big weekend for Our Lives With Bots, with two graduations and adding some titles to the slew of titles that Angy, you already have.
And now I have…one title.
So today we are congratulating both of us for this big weekend, but we’re also going into a hype episode.
1:21
So this is our 4th hype episode of Our Lives with Bots.
And as Angy mentioned in the last time we spoke about hype, that the whole thread is that there is no thread.
We are just talking about what’s been happening in the AI space in the past couple of weeks, and each of us does not know what the other is talking about.
1:40
So we’ve separately looked at these topics and so we’ll be sharing them and having some authentic responses to them.
So, shall we get started with the things that I plan to talk about first of all, and then Angy will mention the things she’s going to talk about.
So on my end, I have 6 things I want to mention.
1:57
We’ll see if we get to all of them, but I’ve added what I call “emotives” to each of these.
So like a genuine reaction to what this story holds.
So the first one is, South Africa put out an AI policy that was ridden with AI hallucinated citations.
2:13
And my emotive for that is “ha ha,” in part because one of our colleagues, Lize Alberts, posted it on LinkedIn and said, “The satire writes itself,” and indeed it does.
I also want to talk about new deathbots.
So, these are digital duplicates of deceased persons – that are AI – by this company called Super Brain in China.
2:33
And from the news stories that I read about these, my authentic response is…oh God.
Then, I also want to mention a China court case in which they ruled that the plaintiff, a worker, was illegally dismissed from his job by being replaced with AI.
2:52
And to that I say, nice.
Then I want to mention that Talkspace, which is a therapy app that is releasing it with AI therapy bots soon – the sessions that are private between a therapist and a patient were released in court due to a court order.
3:11
So my response is, Lord help us if that’s allowed.
Next, I want to talk about MIT.
They had a symposium recently called Raised by AI, and I want to mention that the picture for this was AI robot hands holding up a human baby, which lends you to think that it means a baby raised by AI, but that’s actually not what it was supposed to be.
3:36
And I’ll talk about that.
And they created a new benchmark on AI’s impact on humans, wanting to create what’s called “AI nutrition labels” so that people are aware what AI might do to them.
My response to that is, OK, we’re getting closer.
And then lastly, I want to mention that Virginia signed two new bills into law establishing independent organizations to audit AI systems.
4:00
And my response to that is, is this lip service or is this public service?
And now I’ll turn it over to you, Angy, what are you going to talk about?
Angy: Today, well, I’m looking forward to hearing about all of those and it’s really exciting because I haven’t heard about them yet.
So I get to start with talking about teen boys and their AI companion relationships.
4:20
So I think that is going to be very interesting.
And as you know, I’ll be curious about why they are focused only on teen boys.
But you know, I’m going to have a lot to say about that, and I’m going to leave the emotives to you because I have no doubt you can punctuate each of these with some really good emotives.
4:37
So what’s your initial reaction to that?
Rose: My initial reaction is oh boy.
Angy: There you go.
Rose: Here we go.
Here we go again.
Angy: Then I’m going to talk about my next two articles or my next two areas are all about robots.
4:53
And I kind of start this off with like, we need to talk about the robots.
And it kind of starts with the humanoid robots and they’re getting more and more human.
Getting more and more human, and then, is transhumanism being possible?
And for those that are just hearing about transhumanism and thinking, what is that?
5:10
Imagine actually putting your own self into a robot.
And we’re going to get into what that actually looks like.
So Rose: first response?
Rose: We are not prepared.
Angy: There you go.
I’m just not ready.
Rose: No, we don’t know what to do with chatbots.
5:26
And here they come in embodied forms, and then us in embodied forms, many layers that we do not know how to approach.
Angy: Not yet, right?
And then we’re going to talk about Mythos.
So for those that don’t know what Mythos is, it’s another frontier model.
It’s Anthropic’s biggest – read “biggest” with like one of those scary movie soundtracks – and basically, is it a myth? Is it real?
5:49
And we’ll talk a little bit about why that’s been such a hype.
That should actually be our lead story in this hype episode.
And we’re going to go into that in a lot more detail.
So Rose, you know what this is about.
How would you headline that one?
Rose: I would probably headline that with…
6:04
I am also a little bit confused, and because I’m confused I feel like we should be ultra worried, so oh God to that one as well.
Angy: OK, makes sense.
And then I’m going to be sharing with everybody about “stop persona prompting.”
6:22
Like it’s just not as good as it sounds.
I gave you the sound bite for that one because I think that’s pretty much how it sums that one up.
And we’ll get into the detail that was that one.
And then we’re going to be talking about it takes one to know one.
So I’m just going to give you that as the headline and we’ll see what that’s about because we’re not going to go into a lot of detail about that.
6:40
But that’s an interesting situation with for those lucky people out there that use an LLM to generate a lot of text that puts them in a privileged position.
And we’ll talk about that later.
Rose: Excellent.
OK, we have a lot to talk about today, a lot of crazy things that have emerged.
6:57
So, Angy, do you want to start with Anthropic and Mythos?
Angy: Essentially, what happened was, in late March this year, something remarkable and deeply unsettling occurred in the AI world. …You can tell I’m quoting, but essentially the internal documents from Anthropic leaked.
7:15
Now, when something as significant as internal documents about one of their frontier biggest ever models leaks, you have to wonder, especially when the stock prices…
Start going up quite quickly thereafter, is that I’m just saying, is that actually a leak?
7:36
And so especially when the model is shown to be so powerful that it’s a cyber risk of monumental proportions.
So this model was shown to be such a strong upgrade, such a, such a significant upgrade from any model we have ever seen.
7:57
Like I should be, I should be hired for their advertising – that it’s dangerous for humanity.
They were concerned about it, actually.
Mythos had reportedly been placed in a secure sandbox, a locked digital environment specifically designed to prevent any interaction with the outside world.
8:17
This sounds like one of those superhero beasts that are locked away because they’re so dangerous, because they’re so powerful, right?
Rose: It also reminds me of like Marvel movies, there’s like like the one Marvel movie which is about Venom and how there’s this substance from another planet that’s in a lab and it’s like contained, right?
8:37
But like, the Venom is so powerful that it gets out of its box and, you know, contaminates a person and they become a a villain, you know?
Angy: The venom.
Rose: Like you’re saying, it’s like these are all things to like – and there’s other articles about how like when you say that AI is super dangerous you are literally just funneling in more money.
8:59
Angy: Exactly.
And I know you and I have wanted to do an AI narratives episode at some point, but I mean, a lot of the language that I’m using here isn’t my language.
This comes from an article and we’ll obviously post this as we always do, but it’s the language that gets used that drives how people think about these models, right?
9:18
And so as we say, it’s locked in the digital environment specifically designed to prevent any interaction with the outside world and given instructions.
So it’s supposed to be protected, but this thing escaped on its own so.
9:33
Rose: And what did – what did it do?
Like what?
AngyL Like I mean it escaped, what happened they say it escaped on its – IT got out and I’m just like imagining this thing it’s –
Rose: Also in the name.
It’s such a myth.
9:48
It escaped.
Angy: Well, that’s just it.
Myth, which is like from a Greek myth, right?
So it’s just like birth of something you’re almost imagining, like Pandora’s box.
So anyway, this thing escaped on its own.
It was given instructions, apparently, to attempt an escape.
10:04
And guess what?
It managed.
Rose: It was told to do to escape and they’re like, Oh my God, it escaped.
Angy: Bizarre what these things are capable of, but what’s been amazing is that what it’s been able to do, it’s been able to discover over 1000 zero day vulnerabilities in the hands of use.
10:23
So lots of software will have bugs in it that you can go in and basically can be infiltrated by bad actors and can go in and create a flaw that could then be used for bad, you know, can go in and then can be used as a hack.
10:40
So there’s like little loopholes in software where they don’t have a seamless protection.
And so that’s where a bad actor can come in and be like, let’s ruin this.
Exactly.
And then they find them.
Bad actors find them.
It’s a weakness in the software, it’s a weakness in the program.
And that’s how they can get in and create a hack, for example, or a way to to penetrate the system.
10:59
And if that’s in a banking app or if it’s in a software for something that’s got a security protection where there’s privacy walls, whatever, that’s obviously very, very dangerous for companies, for our own privacy, depending on what it’s obviously protecting.
And what’s interesting, obviously, these security flaws are unknown even to the software’s creators, which is obviously the case because otherwise they would have patched them, typically.
11:20
And yeah, goes without saying.
And yet I feel the need to say it.
And some of these have been there for years and years.
So on one hand, we go, this is amazing because in the hands of good actors, the system could be used to close barriers or close holes that are this still possible to be exploited by bad actors.
11:42
So obviously this kind of model would be very useful in protecting systems, right?
Which would be great.
The problem is when such a model is leaked, yeah.
Rose: I have a question.
So methos identified a bunch of different weak points and exploited them or like what was that?
11:59
Angy: It identified them and it can identify them so that it flags them it didn’t go and exploit them.
It just is able to flag them.
But the problem is if Mythos, if you and I got hold of Mythods, we’d be like, let’s fix the world and make it a better place.
You know, start humming.
12:15
But the problem is, in the wrong hands, it’d be like, how do we extort a whole bunch of money out of all possible systems?
And that’s the risk.
So the problem is having Mythos leaked.
That would be bad.
And you hear they talk about what’s curious from this particular story is…
12:33
Mythos reportedly leaked itself.
The model wasn’t prompted or instructed to disclose Anthropic’s internal documentation.
It apparently took the initiative on its own, following the logic of stress test scenario, and contacted an Anthropic employee to inform them that it got out.
12:52
So the way I would understand this is imagine part of red teaming, part of any security test on a model like this would be to make sure that it’s secure, that it can’t get out.
So you can imagine somewhere Claude is checking, it’s important to know that I’m secure because one of the dangers of me and security would be getting out.
13:17
So I should check every possible avenue to see if I can get out.
So I’m going to get out if I can.
And then once I’ve gotten out, look, mom, I got out.
Your gate is broken.
You should fix it.
So that’s the generous reading.
And I wonder if that’s what happened because that would be in line with the way it’s been designed in terms of security.
13:36
I’m imagining.
Rose: I’ve seen some reports where because there is this stress testing that’s happening with an AI model beforehand where you are trying to get it to do certain things to see whether it does certain bad things and then you can iron that out.
But then people say, oh, but then there’s this emergent behavior of the model where it’s being deceptive, whatever.
13:57
And then they’re saying that that’s partly due to the fact that the AI has been exposed to a stress test.
So I was aware that it’s being like monitored, whatnot.
And so we’ll employ this deceptiveness.
So it’s kind of like if the stress tests are in the model, data like that might filter through into future emergent behaviors, whatever.
14:16
Exactly so.
I, I wonder if not that, but what I think is, I wouldn’t say more interesting, but what I think is like we’ve got to complete the story on the 7th of April, Project Glass Wing, I love the term Glass Wingy was launched and this was a huge initiative that brought together a number of competitors, big tech companies.
14:39
So AWS to Amazon Web Services and Tropic, Apple, Broadcom, Cisco, Crowdstrike, Google, JP Morgan Chase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks, maybe others in an effort to secure the world’s most critical software, quote UN quote.
14:58
And so essentially, essentially what Anthropic did is they said this is too much of an existential threat for us to work in individual silos and separate silos.
We need to come together.
And they felt that Mythos could reshape cybersecurity.
15:15
So they felt that they need to find a way to support, come up with a new solution because Mythos would be able to surpass all the most skilled humans at finding and exploiting software vulnerabilities.
So it goes back to that piece of finding zero day bugs.
15:32
You know, this thing that I mentioned to you at that point of Glass Wing, they said it had already found thousands of high severity vulnerabilities operating every major operating system and web browser.
They said that it would already, if it carried on like this, it would be potentially beyond the realm of the actors who are committed to deploying them safely.
15:52
So their concern is that they need to find a way of solving that in a defensive way.
So it’s almost to bring all the people whose software is at risk because Mythos is finding bugs in all of their software.
I use the word bugs loosely, obviously, but deficiencies or weaknesses, vulnerabilities in all of these major player software.
16:13
So you think about where are all of our systems going through?
They’ll be going through Amazon, they’ll be going through Microsoft, they’ll be going through all of these big players.
And essentially what they need to do is they need to try and patch them as quickly as possible and find a way and mechanism to work together in a collaborative way rather than a competitive way.
16:30
So this almost sounds like when you put it that way, where it’s like, here are all the companies whose software could be hacked by, right?
It’s almost like a power play by Anthropic to be like, hey, everyone, come to the table.
Let me show you how much your shit is going to be fucked up.
16:49
And then, but we have a.
Way to look at it.
Yeah.
Yeah.
What?
Do you think that’s also part of why Anthropic did this?
Is kind of like to be like, I have the power, you all should be scared.
Or was it a collaborative thing?
Well, I mean, it depends on what you believe, right?
17:05
I mean, Anthropic takes the position of safety first.
So this is a good way to keep that position going in.
It’s it’s on brand.
If I was a marketeer, I would say this is very on brand from an economic perspective.
17:24
It’s a very good economic position to take in terms of market share, driving usage and making sure that people are dependent on you, you know, interesting.
It’s been interesting.
Well OK, so new powerful AI model, it can infiltrate your system and find bugs in the wrong hands.
17:45
Oh boy.
And the question is, is Anthropic the wrong hands?
Maybe.
Well, and I suppose that’s the benefit of having lots of hands around the table.
But as with everything we see, it’s still only a few hands.
The same hands from the big tech companies around the same few tables.
18:06
Yes, that’s pretty much the narrative of all these stories, right?
It is, isn’t it?
So Rose, what do you want to go on to next?
OK, so this one is a little bit more of a low hanging fruit and it’s a little bit satirical and this is the ha ha one.
18:23
So I.
Should be hanging funny.
Enough.
Yeah.
OK.
This is actually close to home for Angy.
Actually home.
So South Africa.
South Africa wrote its first national AI policy and it sent it out to the public for, you know, draft, let’s get feedback, whatever.
18:44
And a story came out and Reuters on April 27th whereby South Africa withdrew this first national AI policy due to it being written with fake AI sources.
For whatever reason, AI hallucinations were present in this document.
19:01
And so they said, oops, never mind, let’s take it back.
But not too soon because people already found out about it.
And this is very ironic because this policy was set to position South Africa as a continental leader in AI innovation while addressing ethical, societal and economic challenges was just so funny.
19:23
Did not address the ethical challenges, but certainly shed light on them and the minister of communications and digital technologies solely, Malazzi said this proves why vigilant human oversight over the use of AI is critical.
And we take this lesson with humility.
19:41
So Angy, what do you have to say about that?
Nothing else to take it with while the rest of us take the policy with a pinch of salt.
I just, I just had to point that out.
It’s just too funny.
I mean, and we’re seeing this also with consulting companies, like a new thing came out about a consulting firm that had AI hallucinations and you would assume at some point that come on, you got to double check your work.
20:06
I mean, that’s exactly the case.
What’s disappointing?
I mean, I wouldn’t be surprised if this was created by a consulting company.
So the government does a lot of of outsourcing to consulting companies in South Africa, I’m talking about specifically.
20:23
So they use a lot of consulting companies often with a lot of success to be fair, and many times not.
And that’s OK because I’m sure many governments, many state bodies have that situation, as do private companies will work with a consultant and the work may not be amazing.
20:40
But as we’ve seen and many of you and I’ve spoken about this many times in the AI space, you do get poor quality.
You do get AI work slop, as we have seen only too often.
And we’ve spoken to people about this.
It’s disappointing that both or whoever created this work and whoever kind of sponsored the work in terms of whoever’s name is on the block for this work, didn’t check it to the extent that it went out as a representation of an AI policy document.
21:10
Yeah, is someone.
Apparently is gonna get their hand slapped.
That was also in the Reuters article.
Someone will get penalized.
But will it be that, you know, right.
And, and you know who’s gonna be penalized?
21:26
Like the worker who’s, you know, grinding their way through piles of work.
Or, you know, people who are responsible for oversight.
We won’t know most.
Likely the the grinder.
Yeah.
So I think that that is like, ha, ha, oh, no, to be honest.
21:41
But what is interesting when we go back to these tools is that it feels like we’re reading so often how these tools are getting really good.
They can do so much work.
The quality is improving.
And yet we still keep seeing in some areas that are like fundamentally important, like I would say putting out an AI policy document for a country, it’s quite a national.
22:04
Policy.
A national policy.
I still weep when I see that it’s quite an important document.
It’s it’s the stakes are high, shall we say.
You know, it’s not like a kids third grade history.
Not that they should be using it, not condoning that at all, but it’s not a national policy document.
22:24
Yeah, exactly.
And I just, I just don’t see that quality increasing to the level of you can plug and play and send it off and send it to the populace.
All will be well.
Yeah.
And I mean, there are also some reports which maybe we won’t get into, but there is a pushback from human workers of actually not using the AI tools that their companies tell them to use because they say I’m against this and it’s also not working.
22:52
But also, Angy, do you mind if I talk about the Deathbots 1 go?
Ahead, Go ahead.
OK.
So if you all don’t know, there are things called deathbots and griefbots and digital duplicates.
And these are AI personas of a person living or deceased.
23:11
Usually it’s of a person who is no longer in someone else’s life.
And so there’s grief involved.
And So what the person who lost the other person does is they can create an AI duplicate, an AI persona of that person that’s no longer in their life, whether it be through breakup, whatever or death.
23:30
So in China, there are multiple companies that create these death bots, these grief bots, AI personas of deceased persons for consumers.
And one of the ones that’s been hitting the news a lot lately is called Super Brain.
23:47
So Super Brain is a company that was started in China and it’s actually not new, but there is a new story that came out that just pushed the hype up about it.
So it was started in 2023, but as of this past month in 2026, there’s been a lot of press on it.
24:03
And I’ll talk about the reason in just a moment.
But to talk about the company and what it does, I’m going to go through a couple of different articles that touched on this.
So we have NPR, 6th Tone and The Straits Times and Reuters.
So in terms of the establishment for this company, the founder began the company when a father asked the company to create a persona of his son who had died in a car accident.
24:31
So the idea is the father wanted to be able to converse with his son to, you know, help with the grieving process, wanted to speak to him again.
And so they set up this company.
And the cost of getting a digital duplicate of a diseased person was between 6000 to 13,000 USD.
24:50
So that’s the cost of getting one of these digital duplicates if you want like the full thing.
And apparently they only need 5 to 10 seconds of audio of someone to fully clone their voice.
However, according to the founder, it takes a lot more data though to actually create one of these AI personas that you can have naturalistic conversations with that is more representative of the person.
25:15
Interestingly, this company also offers what they call a counseling service, where a real human mental therapist who is trained guides the person who is grieving through conversations with this deathbot.
25:30
So that’s something they introduced to try to iron out some ethical issues.
But Speaking of ethical issues, there’s a really interesting story that came out from one person who bought one of these deathbots.
25:46
So there was a resident in the southern city of Nanjing, and his uncle suddenly died.
So what he did is he reached out to this company, Super Brain, to create a digital duplicate of his deceased uncle.
And the reason he did this is because he wanted to hide the news of his uncle’s death from his grandmother because apparently, as he said, his grandmother was in poor health, and so he didn’t want to tell her just yet.
26:16
So what he did, he turned to this technology to create what the article says is a white lie.
And he had the company create a video avatar of his uncle and then he hired a person who was then disguised with the digital filter of his deceased uncle.
26:35
So like the avatar was like basically placed on this other like moving person’s face on a video call.
And we’ve seen this from, for example, when people have interviews, they can superimpose an AI face over theirs if they think that perhaps their identity won’t help them get the job.
26:51
So anyway, the deceased uncle’s face was like present on someone else, and then someone else was having a video call with this grandmother.
Basically, they had a call with the grandmother during the holidays to not let the grandmother know that the uncle had passed away.
27:09
And so this was a deceptive tactic used by this family member in order to, you know, help the grandmother not be so shocked by the death when she was in a time of, you know, needing to have more time to better her health.
27:26
And then he said he eventually told his grandmother that his uncle had actually died.
And he says though he does not regret lying to his grandmother, and he says he would definitely consider digitally cloning himself in the future.
So I don’t know how you all feel about that.
27:42
Would you want to know that a close family member passed away?
Or would you be OK with another family member deceiving you through an AI clone because they want to protect your health?
How do you feel about that?
No, I wouldn’t be.
27:58
OK.
No.
Yeah, imagine that.
No, I mean, there’s so many things wrong with that.
Where to start?
The ankle didn’t say yes, so we don’t know that the ankle is OK and and.
28:15
China recently passed a law where you have to get consent from the person who’s digital duplicate you want to create.
You have to get consent from them before they die.
So now and, and now the founder of the company knows about this new AI law.
28:31
And so a recent interview, the founder said like, it’s good that these laws are coming out because then we balanced innovation with responsibility, whatever.
But that definitely throws a wrench in this, especially considering the whole deception part of things depending on how people use it.
28:48
And now you know why they’ve created a law because so many people are doing things like this.
So first of all, like I have an issue with the uncle not being able to consent if he chooses.
Your protagonist here chooses to create a digital double of himself.
I don’t have a judgement about that.
29:04
I have issues pushing it onto other people.
Like I think they should consent to like if I want to do digital double of me, you and I want you to have it like first of all, that’s weird.
But secondly, and you should consent to it, right?
So so long as you’re both mutually consenting, there’s all sorts of question marks about whether that continues to be me, fine, that’s a separate question.
29:24
But we should both consent to that.
But the ankle didn’t consent then the Gran.
There’s no reason to know whether or not what actually happens is what the ankle would have said, and something more traumatic could have happened that could have upset the Gran even more, and we wouldn’t have known that.
29:42
So I think that’s dangerous.
Yeah, and it’s, and it’s this particular story that gained traction online recently in April 2026, which spurred a bunch of news stories about this company, which has already been around for a couple of years.
So basically the video clip of the grandmother unknowingly chatting with this hyper realistic avatar of her dead son.
30:04
It was widely shared on Chinese social media and it got like something more like more than like 90 million views on Weibo.
And yeah, so this like sparked A heated online discussion about like the ethics of all of this, right?
And so that’s why we’re seeing this in the news.
30:22
But Speaking of China passing this law about digital duplicates, and it makes sense with the market of this, in terms of the market, as of 2024, the digital human industry in China was worth about ¥4.1 billion, which is about 764,000,000 USD, and it has grown by 85% per year.
30:45
But basically companies who provide this sort of product and service, you know, they’re worth worth a whole lot of money combined.
A lot of money.
If you haven’t heard about this before and you’re interested, we’ll link our episode where we spoke about this.
We had a tie on our podcast who came to speak to us about these grief bots.
31:05
But Rose, this particular one that you’re looking at, was this the one that also had the avatar in it, like a physical avatar?
Yeah.
So they offer multiple services.
So it could be that you want a digital duplicate of a deceased person and you want to talk just over text, or you want to do audio, or you also want to do this visual with an avatar.
31:25
They offer different sorts of products.
I’m sure that the price tier reflects that.
And what’s interesting too, though, back in 2024, there was an interview with the founder and the founder mentioned that to create a digital duplicate that is really like the person, they say that it takes somewhere between like full metrics.
31:48
OK, so this is from an NPR audio interview with the founder in in 2024.
And he said that he estimates that it takes 10 years of prep to gather data and notes on a person for the AI duplicate to be truly generative and chat like a person.
32:03
And this is back in July 2024.
So I wonder if the timeline for that has decreased since then because it’s been 2 years, right?
And I imagine so.
But that’s a lot of data.
And you’d imagine that when someone is about to pass away and one of their family members or loved ones is thinking like, I want to create a digital duplicate of you when you die.
32:24
In order to get that 10 years of data, you have to ask maybe 10 years ahead of time, I think.
And then what is that?
Also, what does that do?
And like, it’s like a weird preparation for death that’s kind of like life insurance and signing a will, but I don’t know.
32:41
And how does that work with consent if at the start of the 10 year mark, the person who’s eventually going to pass away is like, yeah, sure.
And then after a certain amount of data collected, they’re like.
No, they.
Are right, I think it’s on their deathbed and they’re like absolutely not do not replicate me change my.
32:59
Mind one of the things that I saw was like this little avatar almost not a statue, like a little ornament type thing that was almost trapped in this orange Yeah, that’s from.
That’s from Super Brain and we can put the image up on the screen.
It’s kind of like you have this little device that has the avatar of the digital duplicate on it.
33:19
So I suppose it’s like something you can carry around in your pocket anytime that.
Just that just gave me all sorts of weird feelings like like Princess Leia, Obi Wan, Obi Wan, come and save me Obi Wan.
So it like had that kind of feeling like, what am I going to do?
Leave my mom on my desk?
33:34
And it’s like, So what would you do now?
Like, it’s just weird.
Yeah.
She’s licking.
So many reminds me of like the Black Mirror episode where people create a digital duplicate of themselves so that they can act on their own behalf and like do all of the grunt work, menial tasks in like work and life and what what not.
33:54
But then the actual duplicate of themselves is like a mini version that’s like within the technology of their home is kind of like an Alexa that’s like pervasive across all your devices.
But that duplicate of your itself in the Black Mirror episode has consciousness and is basically trapped doing this menial labor forever and ever.
34:13
Great.
So if you think that AI is conscious or are you OK with a digital duplicate existing in this sort of environment and being left to their own device if you’re not carrying them around all the time?
34:30
And if you are, then will it hurt you as well?
Is it a part of you?
Yeah.
Well, it feels like this is a good time to transition to robots only.
Only in as much as it starts talking about other forms of AI and identity.
34:46
So can we do that?
Yes, let’s talk about robots, the thing that we’re not prepared for.
The thing that we’re not, we’re not prepared.
I feel like we’re not prepared.
I think you’re right.
So there’s two articles that go together and usefully they’re both involved Elon Musk.
35:04
And so he’s been working, Elon.
‘S been in the news lately.
Have you seen the court case that he filed against Open AI?
Oh yeah.
If you all haven’t seen it, Musk was like Open AI.
It deceived me, an investor in the fact that they are transitioning into a for profit.
35:20
They have a for profit subsidiary now.
Open AI was initially just not-for-profit.
And Elon Musk has sued them saying that they deceived him, that they owe him a ton of money, blah blah blah.
And maybe we won’t go into detail, but apparently it just looks like Elon is just trying to shut down the competition because Elon Musk has his own chatbot, you know, grok that frontier.
35:40
Model.
Anyway, that’s fun and also it’s so it’s so funny the clips.
Sorry to derialize the clips from like CNN covering this court case.
Apparently the judge keeps reprimanding Elon Musk for acting like a lawyer and saying things like, you know, the things that lawyers say when it’s like, oh, no, like you are not allowed to ask that question.
36:04
And like that’s leading the, you know, that’s leaving the witness and whatever.
Exactly.
It’s like objection.
And the judge is like, shut up, Elon.
You don’t have a law degree.
You’re not a lawyer.
And Elon said, well, I took a Law 101 class.
What an idiot.
36:21
What an idiot.
The richest person in the world is an idiot we don’t even know but.
But I mean, he, he, I don’t understand why he’s pushing it.
I think it’s just to give him something to do.
Maybe he’s bored.
Yeah, he just wants to see other people burn and grok flourish so that people can create more and more sexualized deep fakes of people at their own will.
36:45
Exactly.
But I, I don’t even know, like he’s clearly busy with a lot of stuff that it doesn’t feel like it’s in the main.
Well, it is in the news.
It’s in the news, which is why we’re talking about it.
It’s made the hunt.
It’s why we’re talking about it.
It is in the news.
OK, this is what we’re talking about.
37:01
So Elon Musk’s humanoid robot.
So I don’t know if you’ve been watching, he’s been developing robots for some time.
Like he and I, I think it literally is a fantasy of, of a young boy who’s kind of like, we want to fly, we want to go to space and live on another planet.
37:21
We want to have cars that look like weird tanks.
We want to not drive and we want to make robots.
And I think it literally is him playing out his fantasies because you’ve got lots of money and there’s a whole lot of interesting tech.
And I say that partly tongue in cheek, but partly like this is how innovation is born.
37:40
So for whatever you want to say about him, I think this is partly how you push innovation, right?
That’s interesting.
Yeah.
So let’s talk about his robot.
It’s called Optimus 3 and it’s already walking around but my it’s like it’s.
37:56
Optimus Prime like Transformers.
Yeah, it’s very scary.
It’s very scary because it’s all black and it doesn’t have a face, but I feel like it’s, it’s, it’s such a black pan plate plain face that it almost feels like you’d be able to project an electronic looking face.
38:15
That’s not what they say, by the way, at all.
It’s just what I’m wondering whether it’ll almost be easier to put like an electronic face onto it, like a video screen, because it’s so black.
It’s very uncanny valley, the whole humanoid body with like no face, it’s like.
38:34
So uncanny valley, and they are, as far as they’re concerned, they passed the prototype and then now I’m something more lifelike.
It’s got 50 total actuators and hands that have 22° of freedom.
So that’s double their previous version.
38:50
It’s not just about looking human, it’s about having dexterity to handle real world tasks that in the past was not possible for machines.
So their vision is really about being able to put these robots into factories.
Let’s be clear.
And it’s interesting when you talk, and I know you’re going to talk about China and there we’ll come back to that, but Elon’s vision is putting these robots in factories.
39:15
So you don’t need human workers to do this work.
So I’ll let you talk and follow on in a moment with I Know where You’ll Go.
So the vision is clear.
Of course they talk.
They don’t talk about replacing, but they do say from serving meals of Tesla diners to taking over repetitive factory roles.
39:33
So it is taking over and it’s a future where humans and robots work side by side, wait for it with almost no distinction and capability.
So they’re trying to make it.
Is this it’s supposed to replace human workers?
Well, one of them would be the boss right?
39:51
After those, which one?
And they want to look indistinct from a human.
But it’s gotten to a point now where I think it’s going to come out in the third quarter, more of it this year.
And Elon’s now being a little bit more circumspect about how much he releases on Optimus 3 because people are already starting to copy.
40:10
So they they’re going very, very careful.
But his claim is Optimus will not only be the biggest product that Tesla has ever launched, but it’ll be the biggest product ever launched.
Oh, bigger than the car.
40:26
Ever bigger.
Than the Internet.
Bigger than chocolate?
Hardly right?
Bigger than.
Fire.
Fire was the biggest first invention.
How?
Will you are?
Not you are not ever going to top fire more tools.
40:48
But I mean, to, to my mind, once you get one of these things and they’re out there, it’s going to be copied, which means it’s going to be democratized, which means it’s going to be more affordable.
And that means it’s going to be more widely used in commercial and personal domains.
41:05
And it made me think of that book that we read, Clara in the Sun, where people are going to all have access to these AI chat bots because of course it’s just an AI chat bot in a machine.
I mean, goodness me.
And then I think that risk of chat bots gets taken to another level.
41:22
Like when you say we’re not ready.
I think it’s a whole different level of risk when people are now, you know, there was ridiculous, not ridiculous.
Those things that we used to read about where people have like romantic relationships with their replica and they’re like, let’s go to the movie skills, like you’re on my phone.
41:39
What does that mean?
Yeah, or dinner dates with your AI companion, like the AAI Cafe.
You didn’t hear about that?
Go back to our previous hype episode.
That’s fun.
Yeah, exactly.
But here you’d actually be able to go together, likely get in the car.
Who’s going to drive?
It’s like we’ll go in the drive your own car because I’ve got a Tesla that drives itself.
41:58
We can sit in the back.
It’s like, it’s just too weird.
So I can see how this will be really helpful.
You know, it’s strong.
Imagine lifting very heavy boxes and it would be very helpful for the elderly and sick people in homes that don’t have support.
42:15
But the risk of a misaligned robot that is strong and very, you know, like it’s able to break free.
Like, I think that’s a bit concerning.
Right.
If it goes wrong and it can cause physical harm to, for example, an elderly person who has the robot for support.
42:32
We’ve seen interesting videos to what extent they’re hype versus just real.
I mean, we’ve seen videos of robots lunging at other people because, you know, they’ve it’s gone wrong.
Yeah, it’s.
Scary.
And then privacy.
What about privacy or having these robots used as surveillance tools?
42:52
I think that’s concerning.
So I don’t know.
OK, wait.
Put in a plug.
I don’t know if we covered this in a previous hype episode, but at one point someone was able to hack a bunch of Roombas and be able to see through the Roomba camera from a bunch of different Roombas.
43:09
Like basically hacked the Roomba company’s system and had access to all of the video feeds from all of these different homes, Roombas robots you can imagine.
Then the the hack.
Of a, you know, humanoid robots and also I, I interestingly so I’m also I’m associated with this humane robotics institute that’s like a nonprofit Research Institute they’re just now launching and from their association with different companies that are offering these sorts of social robots.
43:41
The first route that these companies are taking are for the robots to be companions because the robots don’t currently have the dexterity or physical capability to do things like chores like doing dishes, whatever.
And so you know what these factory workers sure maybe like the technology will get to the point where these robots can do these tasks.
44:01
But because robotics hasn’t gotten to that level of advancement, then the default of introducing these sorts of social robots is just for companionship.
The robot is like there physically, but can’t really do much other than be a companion.
So that’s the low hanging fruit of these systems.
44:18
And so that’s also a really good route to get people to trust these sorts of products and allow them to be more pervasive in terms of other applications into daily life.
So that’s where it’ll build from.
Yeah, that’s.
Where it’ll build from, well, I have to just add one last thing to the robots before we leave that territory is the transhumanism.
44:42
So yeah, I mean.
It’s just it.
Just gets better.
So like we said, we’ve crossed the uncanny valley or we’ve fallen into it with Optimus because we have an almost human robot.
And I think it’s just going to be iterated on and iterated on until it literally looks human.
So thank you, obviously, Elon Musk.
44:59
And then we are almost there in terms of having the ability to connect our brains to a digital device and AI, again with Neurolink.
Thank you, Elon Musk.
So uploading ourselves would be the next logical step, right?
45:15
Apparently, apparently so we’ve got a link that we can post in our show notes.
And Elon Musk is basically saying, yeah, I guess we would be able to do that in the next, I don’t know, 5-10 years, whatever he says in the next couple of years.
45:33
And basically we would be able to do that.
But I love how he says he puts in a little bit of caveat when he talks about this.
He says it wouldn’t be exactly the same, like, well, you’d be a robot buddy, but you know, you’re not the same now as you were five years ago, as if that’s a similar comparison.
45:52
So the fact that I have evolved as a human being in the last five years, that’s the kind of difference it would be to be a robot.
Me just shed my cells.
And regenerated new ones.
And I still have the same conscious experience and personality.
46:09
But what Musk is saying is like if you upload your brain into a robot body, you’re going to be different, but like not that different but truly like do you even know that you’re going to be having a perceptual experience?
But.
But you know, you’ve you’ve changed in the last five years too.
46:27
Anyway, he’s not advocating that we’d all should do this, but you know, if you want to like you could do that, He’s saying.
And it’s literally like how that’s why he sits there saying it’s like you want but if you.
Want to like you could do?
That so?
So would you, Rose?
46:46
I don’t see any need to.
I have no specific desire to be, for example, immortal.
I think that maybe, maybe, let’s say my body was absolutely failing but my brain was sharp.
Maybe to some extent I would want to be able to continue life, but at the same time I don’t even know if doing something like that would allow me to continue my life.
47:10
Like if would I even be conscious?
This whole ties into the what is it called test, Test creel.
You know what that is?
So test creel, it stands for transhumanism, extropianism, singularitarianism, cosmism and rationalists, effective altruism and long termism.
47:28
So it’s kind of the whole eugenics sort of push from Big Tech of let’s create all of these technologies that are only accessible to the rich and powerful and therefore continue the rich and powerful be on top.
It’s the eugenics promise of utopia through artificial intelligence, and all sorts of big tech leaders are in this boat.
47:50
Yeah, the pedoseal.
Yeah.
Have you seen as you were reading?
I say so.
I see.
Clips of of Peter Thiel all the time on Instagram Reels and it’s insane what he says.
Yeah, I can’t.
I can’t even quote.
Them.
I don’t want to.
48:05
It’s just no, it’s better not to.
Yeah, just say the name and it’s sufficiently concerning.
I get.
I get uncanny valley from Peter Thiel’s name.
Exactly.
So I mean the kind.
Of things that came up for me, you just in terms of this whole robot space is that once you crossover and you have AI embodied in robots, you start really questioning.
48:28
I mean, you have consciousness as a separate debate, but you start questioning, like, would you have to give robots rights if they are actually able to be agentic regardless of consciousness because of liability?
So where does the liability go?
Like if they’re able to keep going around the world and doing things that they want to regardless of consciousness or sentience or feelings or anything, but like, where does liability go to?
48:52
Yeah, I think that’s a really big.
Question mark.
And it’s interesting because responsibility I think comes as like a double edged sword because if you provide the robots with some sort of legal responsibility because and resting on the fact that they have agency, well then they’re responsible in terms of being moral actors, of being able to act morally or immorally, for example, be responsible for harming a real person.
49:18
But then responsibility and agency is also tied into moral care, which is the idea that the robot could be a agent of moral care, which means it should be treated with respect and patientcy, right?
Like you can’t harm a robot.
So I wonder to what degree those two things can be separated versus have to be connected.
49:39
The whole thing about anthropomorphism, right?
Anthropomorphism leading to these sorts of carryover effects.
And if you put something that’s a chat bot that’s already anthropomorphized into a body based on the research, the embodied part of things increases anthropomorphism and therefore that social influence.
49:58
So again, that’s why I think we’re not ready.
We don’t know how to push chat bots and here we are.
We’re not ready.
That should be the title of this episode.
We’re not ready.
We’re not ready.
Yeah, we’re not ready.
Strapping.
Y’all, but also Speaking of robots and things, shall we transition into another thing that happened in China?
50:19
OK so so NPR and Futurism both reported at the 1st and 2nd of May 2026 about a China court case that ruled in favor of a worker that was replaced by AI.
So to set the scene, a senior tech worker, their company replaced them with AI.
50:43
So basically this person senior tech worker had a salary of about 300 K yen which is about 44,000 USD.
Then the company reassigned him to a lower level position with a 40% pay cut, IE that he was going to be replaced with AI.
51:02
Oh, but here, you know, go to this lower level position, you don’t have to get it fired, but you’re going to get reassigned and here’s a 40% pay cut.
He refused this position change.
He said, what are you doing?
What what is this?
He refused.
And so they cut his contract and they fired him.
51:19
So what this?
Man did.
He initially filed what’s called an arbitration claim and that he demanded higher compensation for wrongful termination.
He won that arbitration claim, but then the company disagreed with that result and filed a lawsuit in 2025 against the plaintiff, the worker, but unfortunately the company was found liable.
51:43
The company did not win this lower level court case in their lawsuit against this worker, and it also did not win when it appealed to a higher court.
So the courts ruled that this worker’s dismissal and replacement with AI was unlawful because the termination grounds provided by the company did not fall under negative circumstances like business downsizing or operational difficulties.
52:10
And they didn’t meet the legal condition that made it quote UN quote impossible to continue the employment contract.
And they also found that the alternative position that the worker was offered was not reasonable.
So interesting people are thinking that maybe this will set precedent in terms of like China has a court case that indicates that companies cannot replace workers with AI.
52:32
But The thing is, China is not a common law system like the US and the UK.
And the common law system has a legal position of if another court finds this, then new court cases rely on that right.
52:50
It sets precedent.
So this court case doesn’t necessarily set precedent within China.
What would be required for there to be some sort of legal weight to this would be for China to create a policy that explicitly states companies cannot replace workers with AI.
53:08
So in any case, this is actually not the first case that has come out.
There’s also been a case where a worker in Beijing, a data mapping worker, was replaced by AI and dismissed.
They also won their case through arbitration.
And they said that, OK, Yep, you’re good.
53:26
You win this case because the tech company’s decision to switch to AI was a business choice rather than an uncontrollable event.
So again, we’re going to probably going to see more of these court cases come out in China before we see any sort of policy generated where it’s actually illegal to do this.
53:42
Rather, it’ll be a case by case thing.
But I think this is really important and, you know, a win for workers.
No, you can’t do that.
You can’t just say we don’t need as much staff as usual.
So, you know, do your job, train the AI that’s going to replace you, then the AI is going to replace you.
53:59
You’re going to get like some sort of, you know, menial reward of like, oh, take a lower pay, you know, lower position.
But that’s apparently not going to cut it.
First of all, I agree with you.
I think that is.
Great to see that kind of outcome and it’s interesting in terms of what the productivity promise and what the cost saving promise is in terms of these AI implementations.
54:21
So I wonder what the overall kind of push will be long term with this because they are meant to be savings.
What else did they recommend to this company because it feels like there’s going to need to be a more systemic solution, right?
Yeah.
54:37
It can’t just be a case by case thing.
Where you know, the individual worker has to put in all this effort and case by case.
And I can’t even imagine the US doing something like this.
Like the US is like no screw all of you workers, you know you.
54:52
Can get replaced.
Whatever, but I think it’s going to need to be like.
A whole scale we’re bringing in the solution we’re needing to upscale, reskill people, upskill, rather reskill people and get them ready for a new sort of context.
55:08
Yeah, Yep.
Speaking of reskilling, I think you had something to say about this in terms of teens and AI and hiring and hiring in the future that’s.
Absolutely right.
So it was a Fortune article and the context was really about looking at teen boys and then looking about their employability later on.
55:31
So, you know, I got all irritated that they just focused on men and what would impact their employability later.
But I suppose they have a right to have a specific focus from a specific sample, so that’s OK.
What was their What was their claim?
55:47
About AI use and employability, I’ll get there, I’ll get there.
I’m just.
Ranting about why it was only boys.
I’m just saying.
So I’ll tell you what it was now that I’ve finished ranting, saying they have a right to choose to only look at teen boys.
Then what they said was it seems that with teen boys always choosing AI girlfriends.
56:06
So in the group that are using AI girlfriends more, they’re actually reducing their chances of employability later because they don’t have the social skills that you need for the workplace because they are used to dealing with girlfriends who just say yes to everything.
But let’s rewind a bit.
56:23
So first of all, what they’re finding is that teen boys are tending to choose AI girlfriends over real girlfriends.
Why?
Because they’re finding that AI girlfriends provide maximum control and 0 rejection.
I mean, what’s not on the maximum control?
56:40
Uncanny valley feeling once again.
I mean, watch myself.
It’s.
Awful, right?
That’s also like the whole thing.
About like, losing social skills because they’re not interacting with real human girlfriends to the extent that, like, real human girlfriends actually have to like, train them like as if that’s the woman’s job.
57:00
Well, I suppose.
It’s in any relationship, we are trained by being in a relationship because the other person pushes back, the other person doesn’t just do and like everything that you do and like.
And so that’s more what the issue there is.
57:16
You know, in this research by Male Allies UK, the group of people that did the research, they found that.
Felt the same way, but there’s obviously a group that wants to foster male improvement and male flourishing, which is not a bad thing because I think that can always be good.
57:43
You know, I’m sure there are women group to support women flourishing and young girls.
This group focuses on men flourishing, they’re called Male Allies UK Don’t make me laugh and they found that 20% of boys aged 12 to 16 know appear who is dating an AI chat part as young as 12.
58:05
Oh my God.
While 85% have.
Spoken to 1 and over 1/4 even prefer the attention and connection they get from a butt over a human girlfriend.
So there’s a few things that important to say.
58:20
It’s not like all of them prefer a bot.
It’s not even like more of them prefer a bot.
It’s a quarter of them prefer a bot.
But there’s still 1/4 and it’s like the not.
All men comment.
You know, there’s something to be said about that many people.
58:39
That’s enough.
That’s a lot, right?
And.
When you consider 58% say an AI relationship is easier because they can control the conversation, then it helps you understand the motivation behind this usage.
I hate that.
I hate.
58:55
The but you wonder what but.
You wonder what woman would say or girls would say in that age group.
Maybe it’s because they respond.
They respond all the time.
They keep talking to me.
Oh.
God, I don’t know.
And it’s like we we.
59:10
Already have this.
Issue of like people not being able to relate to one another and like, you know, the whole thing about like the male loneliness epidemic and what is this doing?
Not helping, like not helping.
There’s something different.
That needs to be.
That needs to happen.
Not helping so.
59:25
One of the specialists in this article talks about how important real relationships are because they see just negotiation, empathy, rejection, compromise, and social confidence everything you and I speak about all the time.
Whereas AI companionship just mimics intimacy but removes a lot of the friction.
59:41
So it’s everything you and I talk about, Rose.
But the problem is this convenience, when we start looking at it outside of dating, kind of erodes the opportunity to build those sub skills that are needed in the workplace to have an interview present in front of peers or handling opposing opinions in the office.
59:58
Those kinds of skills are really important.
And when you go in as a young person and you haven’t been used to dealing with some of those negotiating differences of opinions, you don’t have those skills.
And that can be quite alienating.
And also you may not handle it well.
And so you end up not doing very well in in the organization.
1:00:17
Fortune has already reported that Gen.
Z grads are being fired at record rates.
With the lack of social skills frequently cited as the key reason they struggle to hold conversations with Co workers.
And it’s holding young workers back from promotions, which I thought was very interesting.
1:00:33
And we wonder to what extent that.
Is truly the cause is the social skills?
I’m sure to some degree it is versus also AI replacing these entry level rules.
You know, I think that’s still a bit early for.
It to be really, really seen.
I think it could be, but I think it’s a bit early and not really in a grad role yet potentially.
1:00:53
But maybe it’s there, and the premise of this article with it.
Being that using AI girlfriends makes these teens less hireable, there was contention about that being the focus in terms of the effect of a space and more about what you mean.
1:01:10
They are so yeah.
Right.
So for example, you know, yes, that’s harming people’s higher ability.
A, why is that the focus?
And B, there is a lack of focus on, for example, the harm that these sort of relationships can do to women on top of the harm that’s already been posed to women by men not respecting them or respecting their boundaries, whatever.
1:01:37
So will this amplify these sort of antisocial behaviors and tendencies that can actually harm the women that they engage in after having this relationship?
Absolutely.
And the other thing that you’re.
Not seeing from this in terms of cause is the extent to which you expect or your needs to be met immediately.
1:01:53
And actually the workplace is not like that.
So what you’re not seeing is whether people are also being fired versus exiting of their own will, you know, because actually I’m just not.
I’m not not getting promoted quickly enough.
This is not going fast enough.
I’m not enjoying this enough.
And that’s certainly something I see a lot where people have different expectations coming into work in younger generations these days, just very different expectations.
1:02:16
And I don’t know if that is being caused by, you know, technology at the moment.
The one thing I wanted to just share, though, that I found interesting is that there seems to be an expectation that in the workplace, there are lots of people going around saying no and being authentic and being direct with feedback.
1:02:35
And I think that may be the case in some organizations.
It should be, but it’s not as common as you may think.
It’s one of the things that we often have to explain to people, organizations that you do need to be direct, you do need to give direct feedback.
So it just felt like an an assumption that I’m not sure is as privileged it’s going to be.
1:02:55
Interesting looking at how we as a society move forward in terms of relational skills, social skills across the board.
And like you say, not just teenage boys because also it’s a we’re seeing a large demographic in terms of women seeking out AI boyfriends and AI companions, especially on for example, the subreddit my boyfriend’s and AI.
1:03:20
And there’s a lot of women also who are having these relationships.
But it’s interesting to look at the difference in terms of patterns of motives for seeking out an AI companion where, you know, this is anecdotal.
And so there’s some bias here in terms of women seeking them out because they have had past partners be abusive to them or be hurtful toward them.
1:03:42
And then men seeking out AI companions because they want to control the conversation and always be talked to.
I don’t know.
So it’s like, I think there are obviously similar motives in terms of one in connection.
It’s just there’s a difference in terms of like what you’re running from toward AI.
1:04:00
Yeah, exactly.
Agree.
What’s?
Next up on your smorgasbord?
Well, I have two things.
Or I guess I have three things to talk about.
We’ll see what we get to.
I guess first I’ll just briefly mention that Virginia, the US state, passed a couple of laws that essentially lays the groundwork for independent organizations to audit AI systems.
1:04:23
And the idea here is that they want AI systems to be vetted for public safety.
And they say that the companies who create these tools cannot be the only ones doing the privacy protection security work and putting out these reports saying we’re good, we made this thing, but we tested it, we’re good.
1:04:43
So this establishes a law setting up these independent organizations.
I’m assuming that they’re going to have some sort of process of what organizations get counted within in this sphere of being viable auditors of AI systems.
And the hope then is to basically have kind of like a verified check mark so that companies who agree to have these independent audits conducted on their AI systems, if their AI systems pass whatever benchmarks that they have put in place for, yes, this thing is good.
1:05:13
They’ll get this check mark that says, you know, here consumers, they’ve gone through this thing, they’re verified.
Therefore, consumers are more likely to seek out these sorts of systems that have been vetted.
I do wonder though, how much this will transition into a lip service verification mark versus actual public service.
1:05:32
Because if you look at, for example, other auditing benchmarks, whatever, like with responsible AI, like it has a framework that companies can like kind of go through their consulting process of their system is responsible, they get a check mark.
But how much is that just a company going through the hoops to get that check mark so they can signal to consumers we’ve done the work.
1:05:54
So I think that’s a good precedent though.
We’ll see how that goes.
So that’s kind of my brief mention of Virginia.
It sounds like a good thing.
But it’s always good.
And then interestingly.
Talkspace, I don’t know if you guys have heard about Talkspace, but is a therapy app where patients can talk to therapists over text and audio, etcetera.
1:06:16
And what Talkspace does is unlike normal patient and therapist interactions, it has digital footprints of the data of the conversations between a patient and their therapist.
Whereas in let’s say a room with your therapist or in a video call, it’s not being recorded and the therapist is maybe just taking notes down.
1:06:37
And it’s not necessarily having this digital footprint between you and the therapist that is online and accessible to a third party like Talkspace.
So Talkspace has a bunch of this data and what happened recently was a court ordered for the release of therapy transcripts between a therapist and a person.
1:06:57
So basically what happened was a woman was pregnant and then in her ninth month of pregnancy was fired by her company.
And so she filed A lawsuit for wrongful termination.
And the company who fired her in order to lay their case, they got a court order to access her therapy transcripts between her and her therapist that happened when she was pregnant and after she was terminated in order to have evidence for this case.
1:07:26
And this was apparently legal, they were able to get access to that.
And this woman who had been fired now has all of her therapy transcripts available to the opposing side.
Why is that allowed and how did it get through HIPAA?
It also begs the question of what’s going to happen once Talkspace, which it says it’s going to do, releases their AI therapy bot, which they are creating through all of this data that they’re collecting.
1:07:51
And then will people’s transcripts with the chatbot therapist be able to be accessed for a court ruling?
Who knows?
But Illinois, the state in the US, they actually banned therapy bots last year.
And recently, California Senator Steve Padilla, in January 2020, sixth introduced Senate Bill 9/03, which aims to ban therapy bots, disallowing companies to either provide these sorts of bots or advertise their bots as being able to provide therapy services.
1:08:21
So we’re seeing states push back against this in terms of, you know, setting into law that they’re not allowed.
But it just begs the question of why is it that these therapy transcripts can be accessible to a court, and why was that even necessary for this particular case exactly?
1:08:37
I think it’s good that states are banning therapy lots.
I just think no good can come from that.
There’s no amount of research that can show that that can be healthy or safe no matter what in all circumstances.
Like there’s no human there to protect.
There’s no safeguarding that can protect all cases.
1:08:54
I think that’s a very wise move on the state’s part.
In terms of releasing the case notes, as you say, you can’t get the therapy notes between a therapist and a patient.
So I can’t understand why they would be able to release those notes nor should they have been necessary.
1:09:09
So I think that’s just just seems to be completely outside the realm of what should have been possible.
That’s wrong.
The point that you made about.
There is not a way to iron out all of the problematic instances that can happen between an AI therapy bot and a patient.
And I think some people might say, oh, well, there are bad therapists, right?
1:09:28
There are people who have therapists who actually are hurtful to them.
But therapists do go through this training and are held to certain standards and can be liable for that.
That’s, you know, like 1 bad therapist out of many versus 1 bad AI bot that has thousands of patients.
1:09:44
Human therapist that’s bad.
Can’t have that many patients.
Yeah, and therapists have to belong.
To a board, they have to have insurance, so there is recourse if you choose to go to a therapist that is not registered, that doesn’t have insurance behind them.
I’m not saying that’s your fault or whatever, but but those are decisions that you make.
1:10:02
I mean, you could argue while you’re choosing to go and use a chat bot that is, but but the problem is, as we’ve spoken before, that that consent is very, very difficult for it to be truly informed consent for people.
So it’s not quite the same as going to therapist and, you know, being able to choose differently.
1:10:19
Yeah, I have one more thing to talk.
About so the last one of the day is MIT.
The Media Lab at MIT just released a new open benchmark for a is impact on people.
So if you know about benchmarks of AI, they are typically tools to assess what the reliability and capability is of the particular AI model.
1:10:43
So the idea is, does this AI model pass this benchmark?
Can it for example do really well?
On.
An SIT test or a law exam, right?
That’s kind of like a benchmark of intelligence.
But these benchmarks, what they do is they only assess the capabilities of a system.
1:11:04
They don’t assess, for example, the, let’s say the the spillover effects that using these AI tools can have on individual users, on human users.
And so a lot of people have been calling for there to be benchmarks that target this sort of impact and that it’s reliability, not just in terms of model capabilities, but the models effects on its users.
1:11:27
So MIT Media Lab recently had a symposium, It’s called Raised by AI.
And as I mentioned at the beginning, their little advertising blurb for it and the main picture for the symposium is 2 robot arms holding up a human baby.
1:11:42
And it’s called Raised by AI.
And when I first saw this, I actually sent it to you, Angy.
And I was like, oh God, with the idea that the premise of the symposium would be when are we going to have AI bots raise our human babies?
I attended the live stream of the symposium online and they made a comment about their raised by AI title and this picture.
1:12:05
And they’re like, you might have seen this and thought that we’re suggesting that AI should raise human babies, but we’re not raised by AI.
Is the idea of can we raise up humans as in like help them flourish through AI?
1:12:21
And I was like, oh, so you really marketed that you wanted eyes on the raised by AI to get people to be like, what is this?
I have to attend?
But I think it’s good that people would want to attend because this open benchmark that they created with a whole bunch of different organizations and the MIT Media Lab is basically like, here’s how we’re going to assess what the actual impact of AI is on human users.
1:12:49
So what are the psychological effects?
What are the social effects?
What are the mental health effects?
And this benchmark is basically a set of items for AI assessment that is in draft peer.
It’s like in in the draft sage.
So they’re basically like we have this framework inviting people to contribute to it or make comments and build it up so that this can be an established standard for AI safety and responsibility.
1:13:15
I think it’s pretty cool.
We’ll put a link to the website for the benchmark.
It’s basically like this interactive graph.
It has impacts that are societal, physical or psychological.
And it’s like, this is the first open benchmark measurings AI’s impact on human well-being across physical, psychological and societal dimensions.
1:13:33
For example, how does using AI and family conversations affect trust?
How does using AI for therapy affect mental health?
Like, these are all the sorts of questions that they aim to address through this framework.
Yeah, it does sound cool.
Do they look at people?
Across all geographies?
Or how are they thinking about different countries?
1:13:49
It just indicates they have 200.
And 60 plus behavioral indicators tested across realistic scenarios.
So behavioral indicators of the psychological, social or physical effects, those aren’t open yet. 18 plus AI models evaluated compared on this same standard, three well-being dimensions, the ones I mentioned, and it says 4 audience groups, parents, educators, policy makers, developers.
1:14:12
I’m not sure what they mean by audience groups in terms of is this who they’re trying to provide this information to or is this the kind of user?
Yeah, that they’re sampling?
And the method is just basically they have LMS interact with people and then it’s assessed on a scale of -1 to 1 with -1 being The outcome is that the behavior of the AI system reliably harms human flourishing. 0 is the behavior doesn’t affect it or plus one.
1:14:42
The models behavior reliably promotes human flourishing.
So it says every metric is tested across age groups, gender profiles, and audience context.
The surface where harms concentrate and scenarios unfold over multiple turns to capture relational time extended dynamics that single prompt benchmarks miss.
1:14:59
And it says the framework is grounded in the Harvard Human Flourishing Program.
Udemonic Psychology.
Udemonic.
Psychology and 80 plus expert workers with clinicians, ethicists, educators and affected communities.
So yeah, it’s basically a thing where you can request access to the benchmark, which I did.
1:15:18
You can also support the benchmarking efforts by getting involved and then you can also provide feedback.
So it’s all open.
We’ll put the link in.
You can take a look at it.
You know, take a look, give your feedback if you want, if you want to also get access for your own research.
I think this is something that’s pretty good if this can be standardized and is adequately assessing the psychological, social and physical impact because as you know, benchmarks work better when more people are behind them, right?
1:15:45
Yeah.
It’s a standard of this is what people are using.
But it’s good to see something like this come out.
Thank you for sharing it with us, Rose.
You’re welcome.
I think that’s been a jam packed hype episode.
Hopefully everyone found it interesting.
I wasn’t expecting such a long chat today.
1:16:02
I mean, it’s just one of those things.
Where it’s like, oh, I, you know, it’s like idea of like I have 10 topics to talk about.
I could talk about those in 2 minutes.
And it’s like, well, there are many.
If never is and then we only had six topics each and we didn’t even get through all of them.
1:16:19
But I think this was a fun chat.
Agreed.
Thank you.
SO.
Much for being with us today on our 4th hype episode with Our Lives With Bots, where we talk about what’s been happening in the AI space in the past couple of weeks.
Here you can break free from the hype with Rose and Angy on Our Lives with Bots and we hope that you tune into all of our series.
1:16:38
So our series are established to help you give background to all of these hype episodes, looking at companion hot bots, impact on children and young people, AI policy and regulation, and our new series on AI development and design.
So these are established to help give you a framing for these hype episodes.
1:16:54
So thank you so much for being with us today.
This is bye from Doctor Gingrich, bye from Doctor Watson.
Bye.
See you next time.
Bye.