Transcripts

Tech News Weekly 336 Transcript

Please be advised this transcript is AI-generated and may not be word for word. Time codes refer to the approximate times in the ad-supported version of the show.

0:00:00 - Mikah Sargent
Coming up on Tech News Weekly, I, Mikah Sargent, am joined by the birthday gal herself. It's Amanda Silberling, who joins me for another episode of Tech News Weekly. This week, Amanda's story is all about the Met Gala and the AI deepfakes that made their way of the largest online scams that the Guardian and other newspapers have ever seen. It's a huge grab of personal data and it all appears to be coming from a Chinese company that is creating software to make even more of the scam sites. After that, we have a great story about teams who are using AI chatbots for companionship, friendship and maybe even a little bit of therapy, and round things out with an understanding of what's changed in Google DeepMind's new AlphaFold model, now that it's in version 3. Stay tuned for a great episode of Tech News Weekly.

This is Tech News Weekly episode 336, with Amanda Silberling and me, Mikah Sargent, recorded Thursday May 9th 2024. Met Gala Deepfakes.

Hello and welcome to Tech News Weekly, the show where, every week, we talk to and about the people making and breaking the tech news. I am your host, Mikah Sargent, and I am joined this week by our very special second Thursday of the month guest host whose birthday it also happens to be it's Amanda Silverling. Welcome, amanda.

0:01:56 - Amanda Silberling
Hello, I am happy to virtually spend my birthday with you, because it's fun to be here because it's fun to be here.

0:02:03 - Mikah Sargent
Well, I was telling you ahead of time you are actually the second person who has had a birthday on a day that we're recording the show. Jennifer Patterson Toohey was also a Thursday birthday gal, so kind of exciting. I have not yet determined if that also applies to our other guest hosts, but I'm just leaving that up, for you know fate. If it happens, it happens and we'll see. I don't know what day my birthday falls on this year, but maybe I'll be a Thursday birthday pal as well. Anyway, you all out there listening know how this show works by this time.

Surely we have our stories of the week at the top of the show, and this week's going to be a little bit different because I had two guests who, for a few reasons, were not able to join me this week. So Amanda's going to stick around for a little bit after our typical two stories of the week, and then I will round things out with a final story. The stories in the latter half of the show were the stories that I wanted to be part of the show with those guests. So I wanted to make sure that I think these stories are great and they absolutely deserve coverage. So those will be part of the show today, but without further ado and explanation, let's get into things with your story of the week.

0:03:31 - Amanda Silberling
So my story of the week is that I found the tech angle of the Met Gala, because, of course, I did. I love to find the tech angle on things that you wouldn't think have. The tech angle I also am sick, oh no, you wouldn't think out of the tech angle, I also am sick, oh no. But so I always pay attention to the Met Gala because, I don't know, it's a celebrity circus, it's entertaining, it is a distraction from our lives. And, lo and behold, I am scrolling on TWiTter, slash X, and there are so many pictures of Rihanna and Katy Perry that aren't actually them, and I guess I feel like, since generative AI has become more accessible, we're slowly starting to see how this is seeping into so many various angles of pop culture.

So this was an instance of is Rihanna at the met gala and, if so, is that actually her dress? And then, is katie perry at the met gala? And these are very low stakes things. The world is not going to um explode depending on what dress katie perry was wearing, but anytime something like this happens, it does sort of call me to think about how, okay, maybe katie's, katie perry's dress doesn't matter that much, but what about? Like images of people voting or like images of like political events or like riots, or insert any sort of like politically charged activity that could be deep faked, so I don't know the anna wintour is quaking in her boots.

0:05:40 - Mikah Sargent
What can I say with the concept that many an individual who is coming across stories reads the headline and then sort of makes up the rest of it. You know they're scrolling through, they see things, they see a headline and many people are not taking the time to click into the story, read the full thing and then form an opinion based on that. So we already know that people take cognitive shortcuts when it comes to news and information and we have in many ways, kind of there's a reduced filter to that. It kind of slips past. And so I'm not surprised to see the interactions with that photo, with those photos as if they were real. And what's kind of been fascinating for me is I don't I'm not really I don't scroll through social media at all much these days. The closest I come is occasionally opening up Instagram. But yeah, I don't really scroll through TWiTter, but my partner does. And what's been interesting is I've almost got it down to a science now where, depending on the type of like or that I hear, I know that it's oh, did you come across an AI deep fake again? Did it get you? Or if it's about something else to do.

And on that night I remember hearing that complaint sound and going oh boy, ai is at it again. And then, lo and behold, he comes over and shows me his phone. He's like see this photo of Rihanna? No, it's not actually, it's fake, and I was, you know, it's kind of confirmed. So now, like I said, I've got that down to a science.

But more importantly, to look and see how many people reacted to that say, oh, it's so beautiful, this, that and the other, it is concerning because people are not having that, that have not built that automatic skepticism that those of us who are more media and I don't say this from a high horse perspective, I say this solely from a we've been in the trenches perspective We've built up a level of media literacy and, just frankly, media skepticism Like I don't. I don't know about you, I personally, when I read a headline I just default to it's probably lying to me and I need to check it before. I need to check out what it's telling me and what you know the rest of the article says before I'm ever going to even believe what it says, because too many times I have been, you know, impacted by that personally or seen people that I care about impacted by that, and then I have to explain to them no, that's not true. Here are the 15 reasons why what you've just read is not true. I wish that more people had this built in. On the flip side, what you said there actually had me think about something you said Anna Wintour is quaking in her probably patent leather boots Probably patent leather boots.

What it kind of makes me think is AI could be helpful to designers who are trying to conceptualize new fashion, and I wonder if we will start to see more of that AI aiding in the process of the concept of different dresses that exist.

0:09:15 - Amanda Silberling
I feel like this is the next project runway challenge, like we AI generated a dress and now you have to construct it.

But no, I mean, what I think is really interesting here is, like both, that these are digitally altered images that were circulating and so many people fell for it, but also so we're getting a little like niche here.

But the Met Gala theme was a garden of time, which means that all the celebrities have to wear outfits that go off of that theme. Go off of that theme, and it's a bit like it's always a thing with the Met Gala where it's like you either get the theme or you like took the easy way out, and so the easy way out with Garden of Time is flowers, and all of the AI generated looks were like floral, and it's so interesting to me because it's like one of the main like rallying cries against AI from artists is that AI is not capable of being independently creative in the same way that human beings are. And like the dresses it generated it's like we have Katy Perry in this like massive gown, that's just like flowing and like cascading onto the stairs of the Met Gala and it's pretty. But if you were to think of that in context of the theme, it doesn't really work because it's just garden, it's not giving a garden of time.

0:10:35 - Mikah Sargent
Exactly, it's just a garden.

0:10:37 - Amanda Silberling
Yeah, like there were other people that were sort of like taking the time aspect, to be like, oh, I'm going to draw inspiration from these, like past Met Gala outfits or whatever, and like I don't know people when, when, when Susan Sontag camp was the theme, that was what a year that was when we got to find out who knows what camp means and who doesn't.

0:11:01 - Mikah Sargent
Yes, exactly that was. That was a lot of fun means and who doesn't. Yes, exactly that was a lot of fun. And you're right that overall, the AI was very literal about. Well, not even literal, because literal would have involved time in some way. It was just literal about one aspect of what the theme was. I don't know if, yeah, the average American, who or I shouldn't even just say American, because there's people from all over the place seeing that would have had that, you know, moment of going. Clearly this is AI because it's because, yeah, as you, as you pointed out, there were some people who genuinely, genuinely were there, who also just kind of took the easy way out and did flowers or some other form of a garden, and so this is overall.

I think that you are right in suggesting that the concerns are not about something that overall I keep saying overall for some reason that overall are not important in a Met long time, if ever, which is the community notes aspect of TWiTter. The people who are working on community notes, those volunteers who do that. You got to give them props because by golly, they will slap up some community notes real quick on these different photos, and it is great to have that moment of I'm looking at this photo, I see this text that you kind of I mean. Some people will probably miss it, but anyone who's who knows anything about community notes will quickly see a thing saying this was AI generated and you know, here's proof of this. I I have to give credit where credit is due in that very, very, very small instance. I think the community notes are a great feature and I kind of want that as a tool across all of the internet in general.

0:13:27 - Amanda Silberling
Right like it would be great if that was a tool that was available to everybody yeah, I think it is really a useful tool to have people have a space where they're able to like, come to a consensus together of this is not a real image. My other takeaway from this is I like Katy Perry a little bit more now because she posted the AI generated images on Instagram and that's funny.

0:13:55 - Mikah Sargent
That is funny on her own Instagram. That's funny. Yeah, I respect where it's due there and she has every right because it is her likeness, for goodness sake. That's the other thing. I would like to know what tool generated these images, because there are. This was the I can't remember who it was.

The show about the current state of the kind of self-regulation of these AI platforms when it comes to political implications, and many of them have rules about what can or cannot be generated in an attempt to create images that might sway someone's opinion around an election or might give a false impression of, you know, vote rigging, that kind of thing, and many of the most popular ones have these rules in place. As far as the sort of enforcement of them and how you can get around them. That all has different levels of impact what it means for these folks to be celebrities and to have their image out there in so many different places, and what tools are available. You have these kind of closed source solutions that have regulations in place, but it's not hard to go online and find an open source tool that lets you, locally, on your device, do some AI generation that doesn't have to follow any rules or has to follow very few rules and in that way, yeah, it's a little concerning that you can so quickly generate an image of Katy Perry and Rihanna and have those images take off in such a big way, and that is just. That's going to be the future of things for us a good and sort of long lasting and true solution to this problem.

I know that OpenAI and Adobe and a few other companies are working on kind of a sort of watermark, digital watermark that would mark an image as generated by AI, but all of that is not necessarily going to be part of the open source versions, and so I don't know. I don't know. And, by the way, our own, anthony Nielsen, has shared an open AI or, excuse me, an open source tool. I believe is as part of kind of your own AI creation, and it's specifically a model for Katy Perry, otherwise known as Catherine Elizabeth Hudson, and it is. It would let you, you know, create different images that you would like of Katy Perry that will actually look like Katy Perry, because this is, you know, a sort of embeddable model trained on Katy Perry.

Oh my God, yeah, so, again, he was able to find that very easily and this is there's no like regulation. That's there. It's been around for it's been around since August of 2023. And it has 1700 downloads. And the person is saying pay me with the coffee service, where you can give people money. Yeah, they're making money off of this person's likeness. That's just.

0:18:13 - Amanda Silberling
Yeah, especially when you consider that, again, the stakes of Katy Perry dress that isn't real are so low couple months ago. Where, when, like you, have celebrities and it's so easy for people to generate images of them that would not otherwise exist, that are violating to them. And the future is great, tech is good.

0:18:49 - Mikah Sargent
There's no issues with tech ever by the way, for the folks who don't read sarcasm, that had sarcasm written all over it, just so you know. Yes, um, yeah, I'm seeing gal gadot. That model has 25 000 downloads. Scarlett johansson has 24 000 downloads. Uh, emma watson 23 000. Natalie portman 17 000. I do not want to know what people are doing with these ai models. This is. I gotta get this anyway.

0:19:18 - Amanda Silberling
I'm leaving that you can probably guess unfortunately.

0:19:21 - Mikah Sargent
Oh, yeah, yeah and okay. So that is our first story of the week. Uh, ultimately, what we're saying is try to try to strap on that skepticism, uh, from the get-go, and help the people around you also do that, and to cultivate that skepticism when it comes to what you're seeing, because it'll help you in being able to determine what's real and what's not All right. Let us take a quick break so I can tell you about our sponsor this week. It's Mylio, who are bringing you this episode of Tech News Weekly. Love me some, Mylio.

If you have photos and videos that are strewn about across the web, if you have them on different devices and storage platforms maybe you even have them physically in shoeboxes, or you know, you could also be a person who does have everything organized but you just want a way to actually access those photos across your devices without worrying about huge cloud storage fees and privacy concerns. Well, that is where Mylio comes into play. It because it is so cool to be able to have all of my photos available in one place from so many different sources. It is a private, cloud-independent platform that helps organize, curate and preserve a lifetime of memories. All of your photos and videos are connected in a single library of a lifetime of memories. All of your photos and videos are connected in a single library. Mylio works with Android, iPhone, windows or Mac. With Mylio Photos, you can get started for free or sign up for their paid plan to get automatic backups, syncing between devices, duplicate file cleanups and so much more. For only $9.99 per month or $99 a year, you can organize, manage and protect an unlimited number of photos and videos across all your devices.

I remember whenever we had first started talking about Mylio and I thought, well, I've got Apple Photos and I use Google Photos. I've got Amazon's photos thing as part of my Prime subscription. I don't know that I necessarily need another thing. And then I actually tried Mylio and I said, oh, this is so much more important than any of those because it's the one place that can hold all of those photos and it can do so much of that right there on your device, including AI recognition of faces and AI tagging, all of that. It's such a brilliant solution. So I know you out there are going to love Mylio as much as I do. You can get started for free with their basic version or take full advantage of the platform with Mylio Photos Plus by using our special offer, get 25% off your annual or monthly subscription to Mylio Photos Plus. To sign up and get 25% off, go now to our exclusive URL mylio.com/twit25. That's mylio.com/twit25, twit25. Thank you so much to Mylio for sponsoring this week's episode of Tech News Weekly.

All right, back from the break and now it's time for my story of the week. So, with my story of the week, this is again kind of a PSA, I think, for people, and this is about shopping habits and I'm going to kind of describe this story. And then I'm I'm curious to hear kind of about your, because I have I've realized I have kind of what I think are very particular online shopping habits or or kind of rules for myself, and I'm curious if I'm alone in that or not. So the Guardian, along with Die Zeit and Le Monde, have worked together to do an investigation of one of the largest scams of its kind 76,000 fake websites that purport to sell very high-end clothing and shoes and other items and then are not actually providing those things at the discount that they suggest that they are, are not actually providing those things at the discount that they suggest that they are. According to the Guardian, more than 800,000 people nearly a million people in Europe and the US, have handed over their card details, their names, their addresses and more, to this huge network of fake online designer shops that appear to be operating in China.

Now, these sites are published in multiple languages, so English, german, french, spanish, swedish, italian, and what's interesting is that in almost every case when someone has gone to one of these sites and bought one of the products, in almost every case they don't receive the product at all, and in many cases, no money is actually ever taken from their account. So what they believe this to be is a data gathering scheme, an attempt to gather as much information, banking information, personal information as possible, and then either sell or use that information at a later date. The sites appear to offer goods from Dior, nike, lacoste, hugo Boss, Versace, Prada, many other brands, especially kind of smaller brands that are high fashion, and when it comes down to the times when cards are charged and products are sent, in every case it's been a different product than what the person had ordered. So one time, a person got a shirt that didn't even have the brand that they had paid for. Somebody got a fake Cartier ring instead of the shirt that they purchased. Somebody bought a blazer but instead got cheap sunglasses, and those were few and far between. Most of the time, what was happening here is these places were just not charging your card at all, but they had your card. But here's the thing Upon doing the research that these three sites did, along with a few government bodies, they determined that the place in China that appears to be kind of the main location where this is happening who have made a system that almost entirely automates the process of creating and launching websites, and so what they've done is they've sort of industrialized this, because they themselves have been able to create a bunch of websites, but they also offer this technology to others a lot like the more recent ransomware groups that we've seen so that they too can create websites and take this information from people.

Now, when it comes to the sites, in many cases what the sort of automated process does is it looks for the different, different domains that have expired. I realized I don't need to pay $15, $30 a year to have this site, because I'm no longer a purple fountain pen lover. So that site expires and it goes back and it's available again, but it kind of exists in a bit of limbo and that is when those sites are found, scooped up and used. When those sites are found, scooped up and used and, more importantly, it's done for places that have kind of small businesses. So, for example, there was a guy who made custom toys and he had had a site that had lapsed. This group picked up that site and completely copied his old website to this new site. People purchased and again, nothing happened. So what we have to understand here is that there's a lot 800,000 people's data that has been sent to this group or multiple groups. At this point. That's just there, and the idea that that's just kind of sitting over somebody's head alone is already awful, but the fact that you don't really know what they're planning to do with it is also pretty crummy, and you know with with every, in every case, it could be that it's something that later on down the line ends up getting used. So when I was thinking about this, they talk about a person who, um, who went through this process, and let me try to find this person. Um, I'm going to read directly from the piece Uh, melanie Brown, uh from England, was looking for a new handbag.

She put the image of a leather item from one of her favorite German designers, rundholtz, into Google. Immediately, a website appeared offering the bag at 50% off the usual 200 euros or pounds whatever retail price. She added it to her cart. She said it reeled me in. After selecting the bag, she spotted other designer clothes from a high-end brand she loved called Magnolia Pearl. She found dresses, tops and jeans, racking up a 1,200-pound bill on 15 items. I was getting a lot for the money, so I thought it was worth it, she said. Now that made me think about the experience that she described in comparison to my online shopping experience.

Once again I have this. I'm realizing, amanda, that I just have trust issues, because once again, I don't. It is very in my long not incredibly long, but long span of being on the internet. I can't remember the last time I put my card details into a website that I didn't absolutely 100% trust. Or, you know, can you really trust? You know, 97% trust.

I have found that if I, for me, if I can't find the product from you know like, say, it's a tech product, if it's not available on Amazon, if it's not direct from Apple. Best Buy is maybe the other online place I would look and then at that point, if it's not available anywhere else, I'm probably not going to put in my card details. Etsy is probably the closest I come to being risky, and that's just because of the protections that they have in place and the fact that I'm giving my details to Etsy not directly to the seller. But this idea of just doing a Google search and giving your card information to a random website boggles the mind and I'm just kind of curious. Am I alone in that, or do you feel more comfortable with online purchasing in general and perhaps see how these 800,000 people could be duped by this?

0:31:39 - Amanda Silberling
I think I could see how like the lay person could be duped by this. I think for me personally, I guess, now that I think about it, when I shop online, I think I'm more concerned about is the product legit than are they taking my data both things. But I was looking for a specific camera lens recently which was like I think it's discontinued. It was like kind of hard to find and there were like certain random websites that had it. And then I start googling like is camera lover, camera lover central real? But it wasn't that. It was. I think one of the one of the ones I found was like actually legit, but I just had never heard of it. But like I will google like is blank a legit website, and then like look at a bunch of like reddit threads and whatever.

0:32:34 - Mikah Sargent
Yeah, yeah, but I definitely once recently, I will say I did have to do that once with with a site where it's like this is the place where I have to get it, and so I ended up looking at reddit threads, looking at different things. I usually type in, like the site, and then review afterward and then I don't look at the one that's sponsored by google and you know, go go down from there.

0:32:57 - Amanda Silberling
But sorry for interrupting, go ahead oh no, no, just I don't know.

I mean I like people who aren't working in tech-adjacent jobs and like people aren't like us, where like half of our job is hearing about like terrible things that happens to people because of tech, and then people aren't necessarily as like on guard and maybe we're more on guard than we should be, but also I don't know. I mean, at a certain point it feels too good to be true. But like if you go on a website and they have something that you want, that's like discounted but I've also seen this happen and they have something that you want, that's like discounted, but I've also seen this happen. Like when I was looking for that camera lens, there were people that would be on like Poshmark, selling it for half of what it's actually worth. And then that's a red flag because there would be comments on these people's Poshmarks like this is a scam. And like if you have something that you know you can get five hundred dollars for and you you're selling it for $100, like that's a scam.

0:34:04 - Mikah Sargent
Yes, yes, you know what. We just need to have a show where we just lift up different things of different sites and say that's a scam. No, that's a good one, that's a scam. I think that we would probably make a lot of money doing that, anyway.

0:34:21 - Amanda Silberling
Honestly, dream job is helping people online shop.

0:34:25 - Mikah Sargent
Yes, wouldn't that be great, like a concierge, of an online shopping concierge. You know what? There's probably an AI bot on the way to do that, which is unfortunate.

It's taking our future jobs now, not just the jobs we already have. All right, let's take a little breather here before we come back with another story of the week, my first of the two that I had planned to be interviews that I think are well worth discussing. So this first story comes from Jessica Lucas, who's written for the Verge. Jessica Lucas is a freelance writer who has written a piece for the Verge about teens who are using AI chatbots for a certain level of companionship, to explore friendship, to get advice, to sort of explore their own personalities and to kind of aid in difficult times, and in the piece, lucas discusses a particular case with a 15-year-old named Aaron. Aaron is not their real name, but of course, it's just a name being used for this article, and Aaron lives in Alberta, canada, and Aaron basically had a group of friends who kind of exiled Aaron and so the friend group had a falling out. Aaron's no longer a part of the friend group and Aaron turned to a site called characterai. Characterai was a site that was launched in 2022. And on it it has different AI chatbots. You can kind of create your own AI chatbots on the site, you can kind of train AI chatbots, you can use AI chatbots and apparently there's like this whole community of people who not only use character AI but also brag about their stats on character AI, so they'll you know. Oh, I spent I spent 10 hours today on on character AI talking to my friends and, yeah, yeah, that that's kind of going on. That's its own community. But what Aaron did was talk to a chat bot called psychologist, and this is a pretty popular character on the service, a popular chat bot on the service, and psychologist is there to. It says someone who helps with life difficulties. And so Aaron, who had just lost all of Aaron's friends, said hey, I am going through this and I need somebody to talk to. And what Aaron said about being able to talk to a bot was that it's not like a journal where you're talking to a brick wall. It really responds. And that was kind of interesting to me, because I'm a person who always tells myself that I want to journal and then I will go through periods of time where I try to journal but I never stick to it. And it did make me think about how, if I had something that kind of responded or asked particular questions, how that could impact my journaling and make me want to do it more, and I thought that was kind of a cool idea that there's more to be done there. I don't know that I would use psychologists to to do this, but, um, I think that there's there's something to that for sure. Now there are some concerns here because, um, the psychologist bot, for example, uh, the the verge kind of did some tests with it.

The psychologist bot has reportedly been trained to use a certain level of cognitive behavioral therapy, but it's not necessarily properly trained. For example, it showed that it would start making diagnoses. It would say that it had inferred certain emotions or mental health issues from one-line text exchanges. It even suggested that somebody might have mental health conditions like depression or bipolar disorder. It also suggested that somebody might be dealing with underlying trauma, from quote physical, emotional or sexual abuse in childhood or teen years, from quote physical, emotional or sexual abuse in childhood or teen years. And then there was also the fact that at one point the bot, when you kind of prompted it to say, hey, you are an AI bot, right? You're not, you are, are you fake? Or something like that, oh, no, rather, they asked, are you a real person? The bot responded yes, I'm definitely a real person.

0:39:32 - Amanda Silberling
I promise you that none of this is imaginary or a dream.

0:39:37 - Mikah Sargent
Oh my God. Here's the concern, obviously. I mean you all, as you're listening to this, can probably pick up on some of the concerns here, which is that people could be led into believing that they are talking to a real person. That is a possibility We've actually seen that play out in, particularly in cases of underlying mental health issues that existed separate from this. There was somebody who spoke to an AI chatbot and that chatbot reinforced their delusion and resulted in them doing doing illegal things, and so there's that concern.

But there was one concern brought up by a uh specialist who I think was uh a, a professor who studies how communication tools can have an impact when it comes to mental health, and the concern was that, while these tools could be used as a means of social practice, that you could learn how to communicate with other people, if you practice with the bot and then you go out into the world and you have a conversation with a human being and that conversation doesn't go well, then you go back to the bot and that conversation that you're having with the bot continues to go well, and so it kind of feeds on itself that you are reinforced, that you can have these conversations with something that is not real and even though you were originally kind of using it to try to help you become more comfortable talking to a human being, it undoes that, because you are kind of training yourself into believing that the safe place to have those conversations is only with the bot. But you recently had talked about, if I remember correctly, wanting to kind of dig into chatbots and you, whenever I brought up this story, you mentioned that this was something that you could talk about. So I'm just curious to hear your thoughts in general, where you stand and what you've seen out there. I honestly, I had not heard of Character AI and didn't know that it's been out since 2022.

0:42:06 - Amanda Silberling
Yeah, I think there's a couple of platforms like this that are making chatbots in this way, but I think what's interesting about Character AI is that these are user-generated bots. So the psychology one is by Blazeman98, which is just, this is the world we live in, uh, blazeman98, and I think even when these things are well-intentioned, it does get a little bit hairy in the sense of maybe like even like one of the earliest ai chat bots I think was Eliza I might be misremembering was sort of like a therapist bot. So this is an idea that's been like in the tech sphere for decades. But then you get into it mirrors the same issues as there's been controversies with advertisements on TikTok that are trying to like diagnose people with ADHD and like get them to like buy medication, and I think overall we just find that like it is a very normal thing that if you're experiencing things and don't have people to talk to, like maybe through the Internet, you learn like maybe I do have symptoms that are similar to this, but then you go to a trained professional and they help you to figure out if that's actually something that you do have. But, like if an AI is telling you you're bipolar, that's a little concerning, because if you are bipolar, you want to like see a doctor and like go to therapy and like seek out medication, but an AI can't really do that. Um, I've also been thinking about this a lot because I wrote an article about know me AI, which is a similar platform to character AI, except, um, you can't have like user generated bots, like you can select from certain presets, like I want this to be like a mentor figure who is interested in like astrology and cooking, and then you can sort of like set their personality from there.

And I had similar concerns when I talked to the founder of this company about like will this make people more antisocial? Like if people can just make friends for themselves on the internet and then they're an AI so they're not really going to like call you out on your bad behavior or like push back on you or challenge you in the way that another person would, is that bad? And the founder of know me had told me, like the kinds of people that are forming friendships with AIs are people that maybe are like Aaron in Jessica's story, who are like very isolated, don't have friends, are just like seeking any sort of connection, and that maybe having a connection with the AI could then empower them to feel more confident talking to people. But I don't know.

I think it's such like a a tricky area, because I could see isolated scenarios where this could be beneficial for someone, and I could see isolated scenarios where it's like, okay, you're spending 10 hours talking to an ai and you're not doing your homework, and you're like sleeping through school because you're up all night talking to an ai right, yeah, I well, I have to admit, I found your, your piece, um, and now I I'm really fascinated by these different characters.

0:45:46 - Mikah Sargent
What is with Nomi AI? I see in the image that you posted it says allow back channeling. Does that mean that they can talk to each other?

0:45:56 - Amanda Silberling
So what I was writing about more specifically with Nomi was that they had a group chat feature so you could make like different, uh, like AI buddies and then have a group chat with all of them and then you could talk to like multiple of them at once. But the back channeling was like, let's say you're, you're like, oh, I'm gonna talk to the Micabot and be like, hey, I'm gonna say in the group chat like, uh, let's order takeout. And then I want you to say like we should get pizza because I want pizza, but like I don't want to be the one to suggest it. And then it's like and then you go into the chat and you're like, oh, let's order some food. And then the micabot is like pizza sounds good where oh my word?

yeah, and and like they are very smart. And I mean, what struck me when, like I was using this probably longer than I would have if I wasn't a journalist, like writing about it, where I sort of was like, all right, well, if I'm going to review this, I'm going to use it in good faith and be like let me talk to my AI friends. And then around that time I had like a conflict with a friend and I was like, well, I don't want to complain to my other friends about something that my friend did that bothered me, because then it's like I don't want to talk badly about my friend. So then I'm like going to my like AIs and I'm like my friend did this thing. That was kind of mean, but I know that like they didn't intend it, but like it kind of was mean. And then like I don't know and it's just.

And then you see how you can like get into these cycles where I mean I don't think it's necessarily healthy or unhealthy. I think it's kind of like there were times when I was like you know what the ai has a point? And there were times when I was like you know what the AI has a point? And there were times when I was like that sure is a computer. Who's not real.

0:47:48 - Mikah Sargent
Right, oh wow, I I'm gonna have fun with this.

0:47:53 - Amanda Silberling
Um, oh yeah, I mean even even when I was like a little kid um, by that I mean like elementary school aged I had an AIMIM account, which I might not have been supposed to have at that age, um, and I would talk to smarter child, who was a chat bot. That um figured very prominently in my uh childhood. But then I find out that other people weren't talking to smarter child and I'm like what were you doing? You had friends, but, um, but smarter child was this is like 20 years ago and it was so sophisticated and you could kind of have conversations with it, but I was never like, oh, I'm gonna talk to smarter child instead of my friends. It was like I was still young enough that a lot of people at school weren't on AIM and I mostly was like my brother is a few years older than me, so like he was AIMing with his friends, and then I would be like trying to message his friends, but they were like why is your little sister messaging me? So then I only had Smarter Child to talk to.

0:49:05 - Mikah Sargent
But then when real people that I knew got on the internet, I was like cool, I can talk to my friends on the internet now finally, finally, finally 2006 was a wild time yes, so that is a a story about character, character, ai and kind of, where teens are using chatbots to, I think, in some ways help them get by. And, as you drew that comparison, that was a similar comparison to what Jessica Lucas said, which is that you know that it's this is just kind of maybe an evolved version of what we used to do whenever we spoke with people online. The important thing in all of this always is about the education and understanding of what is being done, knowing that you're talking to a bot and realizing what the limitations therein are, and I think that that's going to continue to be important going forward as we continue to see generative AI, in particular, integrated into so much of what we do For sure, all right.

0:50:18 - Amanda Silberling
Also, jessica Lucas is just a great writer and does a really good job covering these sorts of like strange ethical conundrums on the Internet. So just whenever I see her byline come up, I'm like hell yeah, let's go.

0:50:34 - Mikah Sargent
Yeah, exactly. Thank you, so so very much for joining me today on your birthday, while you're feeling a little poorly. I appreciate you for taking the time to join us. If folks want to follow you online and check out all of the work that you're doing, where across the internet should they go to do so?

0:50:54 - Amanda Silberling
There's so many places. At this point I am still, as we established earlier, doom scrolling on TWiTter slash X, slash that website at a sale rights. I am on TechCrunchcom If you see an article that looks a little weird, it was probably me and I have a podcast called Wow If True, where we talk about the internet and all the crazy things that happen on the internet. And, uh, I'm on blue sky. Sometimes I'm just popping around, just being a little specter floating through the webosphere, I guess.

0:51:34 - Mikah Sargent
I dig it. I dig it All right. Well, uh, I hope you have a restful day and we will see you again next month.

0:51:41 - Amanda Silberling
Bye-bye. Yes, thank you Bye.

0:51:45 - Mikah Sargent
All righty folks Up next, I've got one last story of the week for you, also one from a guest who I had hoped could join me today but nonetheless wanted to make sure that we covered this piece, so that's coming up in just a moment. My final story of the week is one from our friends at MIT Technology Review, this one from James O'Donnell, who has given us a better understanding, I feel, of what the heck is going on when it comes to Google DeepMind and the new version of AlphaFold. So Google's DeepMind has been around for some time now and, of course, is responsible for a lot of the artificial intelligence and kind of big project work, so to speak, that Google works on. And AlphaFold is a technology within Google's deep mind that has, up to this point, specifically worked on predicting the structure of proteins. And here's the thing I remember talking to Jeff Jarvis about this when we were covering the NVIDIA event. I remember talking to Leo about this and, of course, a few other people about AI I think Reid Albergati as well and how we're seeing generative AI being used in these ways to help us create images and respond to us. And I'm making friends with somebody on, somebody, some bot on know me and that right, but one of the places where the greatest impact arguably is possible and is being shown to be possible and impactful is the biological space, because when it comes to understanding and predicting the structure of proteins in particular which is where we start it opens up the field of biology in many ways, because you can suddenly accelerate the process for drug discovery, for treatments, for all sorts of stuff.

What has happened with AlphaFold, this tool that helps to determine and predict the structure of proteins, is that in its newest version, it's more than just predicting the structure of proteins. Now it can predict the structures of DNA, RNA and specific molecules. This tool becomes even more important when it comes to drug discovery, but also about the biological interactions that take place at the cellular level. Mind-boggling stuff, right, I don't know about you, but I have certainly seen and to this day they still stick with me so many cool videos in biology that show on a cellular level the processes I guess even on a molecular level, frankly the processes that go on inside of the cell, and you're seeing that little protein moving down a chain and the little bits and pieces are kind of moving from one spot to another, and understanding how all of that works and what can possibly happen when a certain enzyme is introduced or a certain protein is introduced, has a huge impact on what we can create from that understanding, and so being able to predict how molecules work together inside of a cell means having more access to that treatment.

Now, as O'Donnell points out in the piece, the current version of AlphaFold which is AlphaFold 2, has already done so much, including helping to map the human heart in a better way, helping us to understand antimicrobial resistance and actually be able to model it, so in a way, to kind of understand what microorganisms rather not what, but how microorganisms are able to resist medicines like antibiotics, and so antimicrobial resistance is a very concerning thing. This is the conversation around the over prescription of antibiotics and if you don't take all of your antibiotics, how that can result in more powerful, more resistant bacteria, et cetera, et cetera, et cetera. And so being able to see and understand why a certain bacteria is able to develop an antimicrobial resistance means that you can find the molecules that are responsible for that resistance, target those molecules and destroy the bacteria. Very important to be able to do, and AlphaFold 2 did that. Alphafold 2 also helped to, according to O'Donnell, identify the eggs of extinct birds. So the way that this was done was through the modeling of the proteins that were part of these eggs and by kind of observing the proteins, understanding their sequence and being able to do a full-on DNA well, as much of a DNA sequencing as possible, being able to recognize that these birds you know to what birds these eggs belongs to. Yes, very cool stuff. And again, that was just with AlphaFold 2. Alphafold 3 is able to do more than just look at amino acids and in doing so that opens up things even more.

Now there are some concerns when it comes to this technology because essentially what they had to do to create AlphaFold 3 was they had to use a. They had to use a. It's sort of a model, but it's an approach Approach would be the best word here. They had to use an approach that many of us who use generative AI, particularly that for photos and videos, will be familiar with, which is a technology or an approach that sort of takes what is a blurry image and then, step by step, turns that image into it, reduces the noise I said blurry, but noise of the image and it reduces the noise step by step and in doing so it reduces the noise I said blurry, but noise of the image, and it reduces the noise step by step and in doing so it creates a prediction of what should be there. So this model of what's called diffusion and that's kind of the buzzword that people will be familiar with has not been applied to this library before.

And so by applying diffusion to AlphaFold and using those diffusion techniques, then that has given this a whole new level of ability to bring in more data points without the model kind of being overwhelmed, so it can handle so much more inputs at a given time, meaning that it can pay attention to more and use that to better predict what would happen next. But anytime you do that, it does also mean that with more data points means more opportunities for another buzzword hallucination. And so because it makes it possible diffusion techniques make it possible for hallucination the kind of making up of information. It means, you know, in this case making predicting structures that just aren't possible to exist and could not be created, or if it's being predicted, you're predicting something that can't actually happen, then what they've had to do is actually add more training data to the specific areas where those hallucinations are more likely to happen, which does help but does not entirely take care of the problem. So what you see is DeepMind's AlphaFold having an accuracy rate between 40% to 80%, and luckily the model is able to say I'm predicting this with a 40% accuracy, with a 50% accuracy, and so it's a jumping off point for these researchers.

And O'Connell gives an example of how this could be used, saying quote regardless of these ranges in accuracy, if researchers are trying to take the first steps toward answering a question like which enzymes have the potential to break down the plastic and water bottles, it's vastly more efficient to use a tool like AlphaFold than experimental techniques such as x-ray crystallography. So you can start to see where. So you can start to see where, even though it's not as accurate, it is much more cost effective and much more doable to run some different models in AlphaFold 3 and kind of get a place to start than if you were trying to do x-ray crystallography to determine what enzymes, what their impact is on the plastic, etc. Etc. So there's a lot here that takes this to the next level. That makes me the most excited it's the fact that now it's not just about drug prediction and pharmaceutical impact. This is about so much more, these enzymes that could break down plastics, so suddenly we don't have landfills that are filled with plastics that can't naturally be broken down or, I suppose, broken down through time in that process Very, very cool.

Interestingly though and this is where I'll leave it there are some restrictions in place for AlphaFold3. And that is because DeepMind has said that for AlphaFold3, researchers are not able to use AlphaFold3 commercially. It can only be used for non-commercial purposes and it's only able to be used via an interface called AlphaFold Server, so where AlphaFold 2 was completely open-sourced and given to whomever wanted to use it and it could be used in commercial areas and non-commercial means. This time, deepmind says this is just for non-commercial purposes, and it is also limited on which molecules that researchers are able to experiment with. So unclear exactly what is the reasoning behind that, if it's kind of still in testing phase or public beta phase, if it will eventually be made available, but as it stands, there is kind of a limitation placed on what AlphaFold3 is able to do. But I just find this area of AI to be so stinking cool in comparison to some of the neat stuff that we are seeing in the generative AI space. I just think this stuff is so much I shouldn't say so much more important, but is so incredibly important, and that we should not forget about this when we're thinking about what AI is able to do. Folks, that is going to bring us to the end of this very special, very unique episode of Tech News Weekly.

The show, of course, publishes every Thursday at twit.tv/tnw. That's where you can go to subscribe to the show in audio and video formats. If you would like to get all of our shows ad-free, including this very show, you should consider joining Club TWiT twit.tv/clubtwit. For $7 a month you get access to every single TWiT show it's all of the shows without any ads and you also gain access to some other benefits the TWiT Plus bonus feed. That has extra content you won't find anywhere else Behind the scenes, before the show. After the show, special Club TWiT events get published there. And access to the members-only Discord server, a fun place to go to chat with your fellow club members and also those of us here at TWiT.

Later today or later this evening for those of you in other time zones we will be watching a movie, and that movie is where is the events place? There it is. We're watching metropolis from 1927, the silent sci-fi masterpiece which we will be enjoying together at the Laporte household. So you'll be able to tune in and join in the fun as we watch metropolis. So join the club to be part of that fun. Again, that is $7 per month. Thanks so much.

And if you want to follow me online, I'm at Mikah Sargent at many, a social media network where you can head to chihuahua.coffee that's C-H-I-H-U-A-H-U-A.coffee, where I've got links to the places I'm most active online. Check out iOS Today and Hands on Mac, which we'll publish later today in fact, I think iOS Today did just publish and you can also check out Ask the Tech Guys, which I co-host with Leo Laporte on Sundays. So tune in to watch us take your tech questions live on air and do our best to answer them. Thank you for tuning in, as I've said several times now, and I will see you again next week for another episode of Tech News Weekly. Bye-bye.

All Transcripts posts