WEBVTT
00:00:03.839 -->
00:00:06.400I've got smart friends who just won't believe it.
00:00:06.400 -->
00:00:15.359You're giving these companies pretty much everything that uh that they could possibly know about you.
00:00:18.480 -->
00:00:19.760Hello, Hugh.
00:00:20.239 -->
00:00:21.280Hello, Hannah.
00:00:21.519 -->
00:00:24.719Welcome to season two of the Tech Overflow podcast.
00:00:24.719 -->
00:00:26.960You might notice that I'm sitting in your house.
00:00:27.199 -->
00:00:28.079You are in my house.
00:00:28.079 -->
00:00:34.399You're in what used to be my study, and one day it will become a place where I can listen to music, hence all the vinyl records.
00:00:34.719 -->
00:00:44.960But for now, it's very much a I think saying it's like a homegrown podcast studio undersells it because you've definitely gone full tech guy with all of the equipment that's surrounding us.
00:00:45.200 -->
00:00:52.880Yeah, and I've learned a lot about how to record audio and how to record video and video capture cards and all kinds of things, lighting setups.
00:00:52.880 -->
00:00:55.679Hopefully I've done an okay job for a complete amateur.
00:00:56.000 -->
00:01:00.159Someone did ask me for a podcast episode about how you record a podcast.
00:01:00.159 -->
00:01:04.719So it's a little bit meta, but it sounds like you're now fully qualified to lead on that as well.
00:01:04.959 -->
00:01:05.120Yeah.
00:01:05.120 -->
00:01:05.519Yeah, yeah.
00:01:05.519 -->
00:01:11.200And then we can next season do one about how to record a podcast episode about how to record a podcast episode.
00:01:12.079 -->
00:01:15.599Okay, so it is February 2026.
00:01:15.599 -->
00:01:16.319It is.
00:01:16.319 -->
00:01:18.879How have you been since we closed our season one?
00:01:19.120 -->
00:01:19.760I've been great.
00:01:19.760 -->
00:01:23.519I did the promised high rocks that we talked about in the season rap episode.
00:01:23.519 -->
00:01:25.920Family came for Christmas, all those kinds of things.
00:01:25.920 -->
00:01:29.840Obviously, summer here, plenty of time in the pool, down the beach, all that good stuff.
00:01:29.840 -->
00:01:32.640Um, and then season two crept up on us very, very quickly.
00:01:32.640 -->
00:01:33.359How about you?
00:01:33.359 -->
00:01:36.079How was your host season one experience?
00:01:36.319 -->
00:01:40.560Uh well, it was November, December, and January UK.
00:01:40.560 -->
00:01:44.640So I can't conjure up images of sitting by a pool.
00:01:44.640 -->
00:01:55.200Um, I worked a lot on the the chief of staff and lead the strategy team for a UK tech company, and we were pretty busy through the end of the year and start of 26.
00:01:55.200 -->
00:02:02.159So I basically did a lot of work, made the journey down a couple of days ago, ready to hit the ground running on season two.
00:02:02.319 -->
00:02:02.959Awesome.
00:02:02.959 -->
00:02:06.480So season two, it's gonna be a little bit different to season one.
00:02:06.879 -->
00:02:08.400I would say it's a build.
00:02:08.639 -->
00:02:09.120A build.
00:02:09.280 -->
00:02:09.520Yeah.
00:02:09.599 -->
00:02:10.240Yeah, I like that.
00:02:10.479 -->
00:02:17.599Season one, we were figuring out how you actually record a podcast, both in terms of content and the technical side of it.
00:02:17.599 -->
00:02:20.639I learned a lot about time zones in season one.
00:02:20.639 -->
00:02:24.800And we pushed out a bunch of episodes and you guys listened and loved them, which was great.
00:02:24.800 -->
00:02:27.199And Hugh has been breaking down the stats on them.
00:02:27.199 -->
00:02:28.319So, what have we learned?
00:02:28.319 -->
00:02:29.680What are the numbers telling us?
00:02:30.000 -->
00:02:36.879Numbers are telling us that we're building, and I thank everybody for following and subscribing to the show, engaging with us on socials, but we are we are building.
00:02:36.879 -->
00:02:41.199What I also learned is that we do best when one of two things is true.
00:02:41.199 -->
00:02:45.840Either we cover a topic that absolutely all of our listeners are interested in.
00:02:46.000 -->
00:02:46.639Like AI.
00:02:46.879 -->
00:02:48.960Like AI did super, super well.
00:02:48.960 -->
00:02:53.280Driverless cars, our interview with Nick Pally, everybody was interested in that.
00:02:53.280 -->
00:02:56.159So when we do that, things go super, super well for us.
00:02:56.159 -->
00:03:01.759And then also, if we tell a great story, that really resonates with our listeners too.
00:03:02.000 -->
00:03:05.360Okay, so relevant content, great stories.
00:03:05.360 -->
00:03:06.319Sounds fair.
00:03:06.639 -->
00:03:07.199Sounds fair.
00:03:07.199 -->
00:03:09.759So more relevant content, more great stories.
00:03:09.759 -->
00:03:10.960That's the focus for season two.
00:03:11.280 -->
00:03:18.719So I've been uh scrutinizing Hugh's little black book of contacts, and we've got some, no spoilers, but we do have some pretty epic guests lined up.
00:03:18.960 -->
00:03:19.680We do, we do.
00:03:19.680 -->
00:03:24.879And I know our two episodes in season one where we had guests were two of our best rating episodes.
00:03:24.879 -->
00:03:25.520No surprise.
00:03:25.520 -->
00:03:28.719Uh, I thought Jonathan Bedeen and Nick Pelle were fabulous, fabulous guests.
00:03:28.719 -->
00:03:31.680So, you know, more guests, I think uh I think they'll be super popular episodes.
00:03:31.680 -->
00:03:33.520I can't wait to hear what uh all of our listeners think.
00:03:33.599 -->
00:03:36.319And when we're not with guests this season, Hugh, what are we going to be talking about?
00:03:36.560 -->
00:03:38.560Well, I think we're gonna lean a little bit more into AI.
00:03:38.560 -->
00:03:42.080I mean, it's certainly uh it's certainly the topic de jour at the moment.
00:03:42.080 -->
00:03:50.639So we'll definitely lean into a little bit more around AI, a little bit more around sort of tech as it impacts uh everybody, smartphones, those kinds of things.
00:03:50.639 -->
00:03:55.680So a little bit more of that, home devices, things that I think all of our listeners are encountering every day.
00:03:55.680 -->
00:03:58.800We'll try and demystify those for uh for our curious listeners out there.
00:03:59.120 -->
00:04:01.680And guys, I'm already loving Australia.
00:04:01.680 -->
00:04:04.800I've been here about 48 hours, not including the flight down.
00:04:04.800 -->
00:04:09.439So if we could just nail season two, I'm inviting myself back, Hugh.
00:04:09.680 -->
00:04:11.280Yeah, we need that podcast deal, Hannah.
00:04:11.280 -->
00:04:12.879And then you your flights will be paid for.
00:04:12.879 -->
00:04:14.080This will be a professional thing.
00:04:14.879 -->
00:04:15.840Won't be back of the plate.
00:04:16.000 -->
00:04:17.199Yeah, we need that podcast deal.
00:04:17.360 -->
00:04:17.600Yeah.
00:04:17.600 -->
00:04:18.000Yeah, yeah.
00:04:18.000 -->
00:04:18.879Let's make it happen.
00:04:18.879 -->
00:04:21.439I think we've got a good set of listeners and a good set of topics.
00:04:21.439 -->
00:04:22.319I think we can make it happen.
00:04:22.560 -->
00:04:27.120And if you a listener want to help us, I'd say the most important thing you could do would be review our podcast.
00:04:27.120 -->
00:04:29.439So go to Apple Podcasts and write a review.
00:04:29.439 -->
00:04:30.480Be super, super great.
00:04:30.480 -->
00:04:34.160And of course, follow and subscribe to the show and engage with us on socials.
00:04:34.160 -->
00:04:34.959We'd love all of that.
00:04:35.199 -->
00:04:38.399And please, please recommend us to your friends, family, colleagues.
00:04:38.399 -->
00:04:42.000Um, that is a great way for us to grow, is word of mouth.
00:04:42.000 -->
00:04:44.240If you like what you hear, please share the love.
00:04:44.240 -->
00:04:46.319I'm sure your friends and family will be grateful.
00:04:46.319 -->
00:04:48.639All right, so season two, episode one.
00:04:48.639 -->
00:04:50.879Is Alexa really listening?
00:04:50.879 -->
00:04:55.839And I'm sure other people have had this moment where they open their phone.
00:04:55.839 -->
00:04:57.519For me, it's often on Instagram.
00:04:57.519 -->
00:05:03.439And like Instagram knew I was coming to Melbourne, like the day we booked, maybe even before we booked the flight.
00:05:03.439 -->
00:05:08.639And I mean, I personally find it quite useful because it fed me lots of tips and recommendations, but that's kind of crazy.
00:05:08.800 -->
00:05:09.360Yeah, it is.
00:05:09.360 -->
00:05:19.199And uh, you know, if I think back to um my early days at Microsoft and, you know, the great people that I worked with, I'd say if I fast forward a little bit, some significant fraction went to hedge funds.
00:05:19.199 -->
00:05:32.160Um, so some of these smart people off uh off making money, and I'd say some significant fraction went off to ads and are making sure that uh highly relevant ads end up in front of you and uh that these these feeds work really, really well.
00:05:32.160 -->
00:05:36.399So some of the greatest minds of our time are definitely making sure that you get ads for Australia.
00:05:36.639 -->
00:05:43.519Well, controversial take, but I'm kind of grateful because the ads are they're feeding me stuff that I want.
00:05:43.519 -->
00:05:49.120They recommend clothes mostly that suit exactly what I didn't even know I wanted.
00:05:49.120 -->
00:06:02.800Sometimes I even like if I want something, I want a particular item or I want like a new bag or something, I'll search it a bit on Instagram just to get the algorithms going because I want it to feed me inspiration.
00:06:03.120 -->
00:06:08.240Yeah, and um, you're doing all the work you need to do to make sure that uh you get those ads for a while.
00:06:08.480 -->
00:06:09.199Well, yeah, that's the thing.
00:06:09.199 -->
00:06:11.439And then sometimes it keeps going, and I'm like, I've already bought one.
00:06:11.439 -->
00:06:13.439I don't want to, I don't want to know what I could have done better.
00:06:13.680 -->
00:06:15.040It's one of the hardest problems in ads, actually.
00:06:15.199 -->
00:06:15.360Really?
00:06:15.360 -->
00:06:15.600Okay.
00:06:15.759 -->
00:06:16.319Yeah, absolutely.
00:06:16.319 -->
00:06:28.079Is uh, you know, you're on Amazon, you've you've got a friend who's had a baby, you do some shopping for baby goods, you buy them the baby goods, you ship them the baby goods, and then uh, you know, you get baby goods forever.
00:06:28.079 -->
00:06:32.800Which I don't know, maybe for a certain demographic is fine, but for me, I'm like enough of the baby goods.
00:06:32.800 -->
00:06:34.639Uh I'm not in that phase of my life.
00:06:34.639 -->
00:06:38.879Um, but it's a hard problem to figure out when somebody's what we call intent has has actually changed.
00:06:38.879 -->
00:06:41.120And say it was, you know, ephemeral.
00:06:41.120 -->
00:06:42.160Hard to figure it out.
00:06:42.399 -->
00:06:54.720Okay, so we've kind of got into the subject matter, but I mean there's this age-old myth that our phones or our Alexa or Google homes are listening to us, and for the record, that is technically possible.
00:06:54.720 -->
00:07:04.240Like, I'm not saying that that's what's happening, but you have a device in your home with recording device on it, and it's got access to the internet.
00:07:04.240 -->
00:07:08.000So, like someone could be hacking that and like tapping in.
00:07:08.000 -->
00:07:14.399Not that I'm having particularly interesting conversations at home, but that is a technical possibility, right?
00:07:14.639 -->
00:07:16.560Totally, totally, totally is a possibility.
00:07:16.560 -->
00:07:20.639And and look, you know, we'll we'll probably talk about it a little bit later on in the in the show.
00:07:20.639 -->
00:07:34.639But you know, folks like Amazon do pay people and they sign an agreement so that they can record them, you know, 24 hours a day across their house to try and understand, you know, what they're doing, the things they say, helps them develop a better Amazon echo, those kinds of things.
00:07:34.639 -->
00:07:41.439But by and large, um, what is supposed to happen is that these devices will only activate when you say the keyword.
00:07:41.439 -->
00:07:42.879And I'm a little bit hesitant to say it because I'm not sure.
00:07:42.879 -->
00:07:43.600Yeah, the wake word.
00:07:43.600 -->
00:07:56.319We but yeah, we're pretty, probably pretty uh but they're supposed to, you know, you've got uh you're supposed to say hey Siri and your um Apple device starts listening to you and listens until it believes that you've finished whatever it is you're saying, and then obviously, you know.
00:07:59.600 -->
00:08:03.120I wonder how many times in this record will activate the week word.
00:08:03.120 -->
00:08:04.879So they will literally pay people.
00:08:04.879 -->
00:08:05.759How do I get this?
00:08:05.759 -->
00:08:09.199I want to be paid for just being at home using Amazon Alexa.
00:08:09.360 -->
00:08:11.680Yeah, there's actually um there's a couple of companies out there.
00:08:11.680 -->
00:08:15.120Um, there's one that's been around for a long time called Appen, A-P-P-E-N.
00:08:15.120 -->
00:08:17.199It's actually an Australian company based in Sydney.
00:08:17.199 -->
00:08:30.079And a lot of the large tech companies, you know, whether it's Meta, Amazon, Google, Facebook, Microsoft, a lot of those folks will actually hire Apple to go and source humans from around the world to carry out tasks that they want carried out.
00:08:30.079 -->
00:08:40.399So if, for example, you're working on the Amazon Echo and you're working on the Alexa product, you might go to Apple and say, look, I want to recruit some people across the world to, you know, do some task.
00:08:40.399 -->
00:08:41.919It's going to help me improve the product.
00:08:41.919 -->
00:08:51.440And then the Apple folks will go and recruit those people, train those people, get them to perform the task, do all the legal agreements, pay them, all those kinds of things, and then eventually send you back the back the data.
00:08:51.440 -->
00:08:55.840But they've got this huge flexible workforce across the world that can do any task you you want done.
00:08:56.080 -->
00:09:06.159This is, I keep learning about this whenever we talk about stuff, and you'll be like, yeah, there's a whole industry behind it that can link people up to do whether that was like human in the loop checking for the LLMs or this kind of thing.
00:09:06.159 -->
00:09:10.240And I guess they'll have to do it in like most major languages, right?
00:09:10.240 -->
00:09:18.960Because Alexa's the same, at least in every language that I've seen people use it, which admittedly is probably less than 10.
00:09:18.960 -->
00:09:22.240Um, but I assume it's pretty universal, Alexa.
00:09:22.559 -->
00:09:23.360Yeah, that's it.
00:09:23.360 -->
00:09:28.639And they actually have naming teams as well, and some of them hire an agency to make sure the name works across the world.
00:09:28.639 -->
00:09:33.200So I actually had a friend who came up with the name Pentium when uh when Intel invented the Pentium chip.
00:09:33.200 -->
00:09:43.279And one of her jobs was to make sure that you know, Pentium was a fine word uh for Spanish-speaking people, Italian-speaking people, Arabic-speaking people, what whatever it is, and make sure that it didn't offend anybody.
00:09:43.279 -->
00:09:46.559And uh, you know, it sounds pretty high-tech and sounds pretty good in all languages.
00:09:46.559 -->
00:09:52.080And uh, I'm sure that the folks at Amazon probably did that with Alexa and the folks at Apple did that with Siri.
00:09:52.320 -->
00:09:52.799Amazing.
00:09:52.799 -->
00:09:54.320And it must be annoying though.
00:09:54.320 -->
00:09:59.840Imagine if your name is Alexa, you know, you were named before the famed Alexa.
00:09:59.840 -->
00:10:01.519That must be slightly irritating.
00:10:01.679 -->
00:10:02.080Yeah, yeah.
00:10:02.080 -->
00:10:10.320One of my uh youngest daughters um friends' sisters is her name is Alexa, and she was quite annoyed by it, but you can actually go into the settings and change it to a different keyword.
00:10:10.320 -->
00:10:12.480That didn't make it easy when she visited her friend's houses.
00:10:12.480 -->
00:10:18.559But you know, back to your key question, which is a great question, is um they should not be, they're not designed to record at all times.
00:10:18.559 -->
00:10:28.159They're only designed to record when you say the keyword, and then they're designed to continue until they believe that you finish whatever it is that you're going to say, and then they should turn back off again.
00:10:28.159 -->
00:10:41.440Though there has been a couple of legal cases, um, Apple was taken to court, Amazon was taken to court where you know people said, hey, you know, it sometimes records when it shouldn't be recording because it it thinks it should be.
00:10:41.440 -->
00:10:43.039So it thinks it's been triggered.
00:10:43.039 -->
00:10:44.960Um, it's this gray area, right?
00:10:45.039 -->
00:10:58.080It's like I say I used to work with someone called Shiri, and then every time I said Shiri, Siri would think that I activated it or, you know, anything that my brother's called Alex, so I'm sure they accidentally set off his Alexa all the time.
00:10:58.080 -->
00:10:59.200Hi, Al, if you're listening.
00:10:59.200 -->
00:11:10.320Um, but yeah, I mean that's it's a funny one because it means that they're collecting, like in the case of my brother, it would mean that they could be collecting a bunch of stuff that was really not intended to be for the Alexa.
00:11:10.320 -->
00:11:13.600But what can they do if if the words sound similar?
00:11:13.600 -->
00:11:16.159I mean, I guess it's it's kind of fair enough.
00:11:16.159 -->
00:11:17.279Is that controversial?
00:11:17.600 -->
00:11:20.240Well, as I say, you can go into the settings and adjust the keyword.
00:11:20.240 -->
00:11:24.960So you could pick a different keyword and uh, you know, pick something that's less likely to be triggered.
00:11:24.960 -->
00:11:28.559And, you know, there's people working hard to make sure that it's not triggered unless it should be.
00:11:28.559 -->
00:11:33.120And I'd say the 99 point something percent case is it should only turn on when you say the keyword.
00:11:33.120 -->
00:11:36.320So Alexa is listening only when you say the keyword, it's not listening on.
00:11:36.720 -->
00:11:49.600But they are, I guess is a key point that we've maybe not stated, they are sending those clips back to head office to check for like quality of answers, and and that's like a product development.
00:11:49.600 -->
00:11:56.000Is that kind of like when an app asks if it can send the logs back to the exactly that, exactly that.
00:11:56.159 -->
00:12:02.480So obviously, if you think about what's going on, you're saying some words, those words are translated into a digital form.
00:12:02.480 -->
00:12:17.360That digital content is going back to the head office or to a data center run by that company, and then that gets uh you know transcribed out as text, um, which you know we obviously we use in the podcast to transcribe out our recordings, and you might be looking at the transcription now, that's automatically done.
00:12:17.360 -->
00:12:31.440So the same kind of tech is used in the data center, and then you know, try and recognize what the question is or what the statement is, figure out what the answer should be, synthesize that answer as voice, send that voice back to the device and play that back to you.
00:12:31.440 -->
00:12:33.440So they they need an internet connection to work.
00:12:33.440 -->
00:12:38.159I don't know if you've ever been in a flat spot and tried to get Siri to work and it says, Oh, sorry, I'm not available right now.
00:12:38.159 -->
00:12:49.519So it needs an internet connection and you know, your voice becomes text, something happens, some text is generated, that text becomes a voice, that voice you've chosen, and you hear that back on you hear that back on your device.
00:12:49.519 -->
00:12:53.519But certainly whatever it is you've said has ended up in a data center somewhere owned by one of these companies.
00:12:53.759 -->
00:12:55.200And there's no way of turning that off.
00:12:55.440 -->
00:12:56.639No, no, it just wouldn't work.
00:12:56.639 -->
00:13:05.600Um, so you know, obviously a lot of folks um who've listened to the show have listened to our large language model episode and realized that a lot of things going on in the cloud that power these.
00:13:05.600 -->
00:13:11.600And you know, if you squint enough, it's similar kinds of technology that needs a lot of computing power and models and those kinds of things.
00:13:11.600 -->
00:13:14.399So this kind of technology is really only possible in the cloud.
00:13:14.399 -->
00:13:19.679Of course, you could you could probably get it to give some simple answers, like tell you the time or whatever it is just on your smartphone.
00:13:19.679 -->
00:13:24.240But by and large, these are internet-enabled um services that operate in the cloud.
00:13:24.480 -->
00:13:27.360I think that's the number one question at home we ask Alexa.
00:13:27.360 -->
00:13:28.480It's always what the time is.
00:13:28.480 -->
00:13:29.279It's like the laziest.
00:13:29.279 -->
00:13:33.919We both wear watches, we're never far from our phones, and yet I think it's got to be the number one.
00:13:33.919 -->
00:13:34.960Alexa, what time is it?
00:13:34.960 -->
00:13:35.360Always.
00:13:35.519 -->
00:13:49.600Yeah, well, in prep for this episode, I looked back through my Alexa chats or the family's chats because I've got all of those, so I can find out what my daughters were asking and things, which again, you know, you can ask whether that's a that's a fair thing or not, but I've got the whole family's Alexa chats, and I'd say the 90% case is people setting timers in my house.
00:13:49.600 -->
00:13:51.200Yeah, setting timers, that's another one.
00:13:51.200 -->
00:14:01.120That's the 90% case, and then the other 10% is probably a little bit of telling the time and converting temperatures between, you know, Fahrenheit and Celsius, but some pretty basic stuff for some pretty advanced technology.
00:14:01.440 -->
00:14:13.679And I think you said in the prep for this episode that like that's one of the biggest challenges for the product folks that look after Alexa, that it's I remember when it landed, it was like it was a huge new product.
00:14:13.679 -->
00:14:17.679It was all these sort of projections that it would become another member of the household.
00:14:17.679 -->
00:14:20.480And it's not really developed much past me.
00:14:20.480 -->
00:14:22.000Uh, we use it for shopping lists.
00:14:22.000 -->
00:14:25.600You know, if I run out of pepper or whatever, we'll say, Oh, Alexa, add pepper to my shopping list.
00:14:25.600 -->
00:14:30.240And that is helpful, but it's that, it's the time, it's very little more than that.
00:14:30.480 -->
00:14:37.600And look, you know, it's a difficult product to work on, I think, because you put yourself in the shoes of the product managers who work on Alexa, right?
00:14:37.600 -->
00:14:42.639The problem you've got is that lots of people out there have asked Alexa to do something.
00:14:42.639 -->
00:14:43.440It didn't work.
00:14:43.440 -->
00:14:46.159And so what you've learned as a user is Alexa can't do that.
00:14:46.159 -->
00:14:51.200Now there's some product manager who works for Alexa, they're doing their research, figuring out what features to build.
00:14:51.200 -->
00:14:53.759They they figure out that that's a common feature that everybody wants.
00:14:53.759 -->
00:14:59.919They go and build the feature, they ship the feature, and now Alexa can do that, but you don't know that because you tried it once and it didn't work.
00:14:59.919 -->
00:15:01.519So why would you try it again?
00:15:01.519 -->
00:15:08.080So it's very, very hard to educate the users that the product is actually evolving because people learn by trial and error.
00:15:08.080 -->
00:15:11.120And once they have an error, then they never try it again.
00:15:11.120 -->
00:15:13.679So you'll notice now that sometimes it'll promote things.
00:15:13.679 -->
00:15:18.080So, you know, you'll say, hey, set a timer for 10 minutes and it'll say timer set for 10 minutes.
00:15:18.080 -->
00:15:22.480By the way, did you know that I can play your playlist from Spotify?
00:15:22.480 -->
00:15:24.000And you and it's incredibly annoying.
00:15:24.000 -->
00:15:28.720But that's the that's the product managers trying to advertise the fact that they've built a new feature.
00:15:28.720 -->
00:15:31.519But it, you know, it creates this sort of cognitive dissonance, right?
00:15:31.519 -->
00:15:35.679Where it's suddenly sort of almost putting an ad into a task that you want it to do.
00:15:35.679 -->
00:15:48.320So it's sort of unfortunately got a little bit marooned by that because it started off with a basic set of features, people assume that's all it can do, and so you know, it's a basically a fancy egg timer, even though it's capable of doing a lot more today.
00:15:48.559 -->
00:15:52.240Yeah, that happens and it's so sometimes I'm a little rude, I have to admit.
00:15:52.240 -->
00:16:01.600Um that's I'm not proud to admit this, but when Alexa can't get something simple right, if they're recording my reaction, it's probably not very flattering.
00:16:01.600 -->
00:16:05.440So we know that Alexa and Siri, etc., aren't listening.
00:16:05.440 -->
00:16:08.000So what creates this phenomenon?
00:16:08.000 -->
00:16:17.840Because I could say that, right, to some of my friends and family, and they just simply won't believe you based on the precision and relevance of the ads that they get.
00:16:17.840 -->
00:16:21.120So, like, convince me that my phone isn't listening to me.
00:16:21.440 -->
00:16:23.759And look, I've got smart friends who just won't believe it.
00:16:23.759 -->
00:16:24.799They say, yes, it is.
00:16:24.799 -->
00:16:27.440It's the only way that it could be serving ads like that.
00:16:27.440 -->
00:16:32.960But let's let's just zoom out a little bit and think about the device that you've got in your pocket, right?
00:16:32.960 -->
00:16:40.080So we talked about that back in season one, you know, what smartphones are capable of and things, but but let's just let's just think about a couple of the basic things right now.
00:16:40.080 -->
00:16:47.120So, first of all, you're carrying it around um and you're going to different locations, and that location data is being sent back to somewhere.
00:16:47.120 -->
00:16:53.440So if you're a user of Google Maps, you've agreed to share your location with Google, and now Google knows where you are, right?
00:16:53.440 -->
00:17:08.480So if, for example, um, you know, you go to a surf shop, you want to buy some board shorts, a surfboard, whatever it is, and you're walking around the surf shop, you've just given, you know, and you've navigated there, you used Google Maps, you've just given Google the signal that you're interested in surf shops.
00:17:08.720 -->
00:17:12.880Does it know here if I because I go to the gym several times a week, does it know I'm in there?
00:17:12.880 -->
00:17:16.240Not or only if I'm yeah, yeah, it does pretty accurately.
00:17:16.319 -->
00:17:29.839So there's a team at Google called the Blue Dot team, and you know, it's quite a you know, moderate size team, and their job is to make the blue dot as accurate as it can possibly be and try and you know give pinpoint precision to locating where everybody's but but what happens if I'm not?
00:17:29.920 -->
00:17:31.759Because I don't need Google Maps to get to the gym.
00:17:31.759 -->
00:17:32.400I know where it is.
00:17:32.400 -->
00:17:37.920Is it is it still it's running in the background if I unless I've agreed only to share my location when I'm using the app?
00:17:38.319 -->
00:17:38.960Yeah, that's right.
00:17:38.960 -->
00:17:41.519So you've probably said always allow.
00:17:41.759 -->
00:17:48.960I try and set it off for apps unless I need to, because I know that it's tracking, but I mean it's you can't have maps without location on.
00:17:49.200 -->
00:18:01.279And look, you know, things like Instagram, for example, maybe shifting to a different company, Meta, you know, it's it's convenient to have the the location on because it it does do a better job of of serving content to you when it knows where you are.
00:18:01.279 -->
00:18:07.279It also makes it easier for you when you're posting something on Instagram to select the location that you're coming up.
00:18:07.519 -->
00:18:14.160Well, that's clever that they've done that because everyone loves to post an Instagram story with the location or I guess a grid post.
00:18:14.160 -->
00:18:19.440And when I've turned it off on Instagram, it's annoying because then you you have to turn it on every time.
00:18:19.440 -->
00:18:22.480So they've they've kind of hooked you into sharing where you are based on that.
00:18:22.880 -->
00:18:24.640And um, you know, lots of apps do it.
00:18:24.640 -->
00:18:28.960You're owning up to where you are, and so there's a lot of intent that comes from that.
00:18:28.960 -->
00:18:31.519So if you zoom out, let's go back to Google for a second.
00:18:31.519 -->
00:18:38.880If you think about sort of the Google ad ecosystem, you know, so when you run a search on Google, you get back ads at the top of the search results page.
00:18:38.880 -->
00:18:47.200But number one, not unsurprisingly, perhaps the number one, if you type in surfboard, it's gonna serve you ads that are about surfboards.
00:18:47.200 -->
00:18:48.319So that's the number one clue.
00:18:48.319 -->
00:18:50.319The number two clue is location.
00:18:50.319 -->
00:18:54.160That's the number two thing that Google uses in their ad ecosystem.
00:18:54.160 -->
00:19:07.839And so, you know, Google cares a lot about Google Maps and building a great product that we all love and invest an enormous amount of money into that product, you don't pay for that product, but in return, you're giving Google your location, and that location is used to improve the ads that they serve to you.
00:19:07.839 -->
00:19:10.480So you can learn a lot about a user from their habits.
00:19:10.480 -->
00:19:19.519So they know you go to the gym every day, they can figure out a new intent that you have, you know, suddenly you're in a surf shop, you haven't been in surf shops before, you're in a new location that you haven't been in.
00:19:19.519 -->
00:19:30.079I mean, you can imagine all the things that you can distill from a person's location if you have it all the time that they have their smartphone, which frankly is pretty much all the time that people are on the move these days.
00:19:30.319 -->
00:19:34.799Okay, so the Google example is good because you've linked Google Maps to Google search result.
00:19:34.799 -->
00:19:37.440But for me, the big one is Instagram.
00:19:37.440 -->
00:19:41.519And I guess maybe I guess TikTok is the same for people, or maybe Facebook if you're using it.
00:19:41.519 -->
00:19:44.640Like, and is that that's because of Meta, right?
00:19:44.880 -->
00:19:45.519Yeah, that's right.
00:19:45.519 -->
00:19:48.960And the meta, you know, the meta ecosystems are great too, right?
00:19:48.960 -->
00:19:55.920So if you think about something like Facebook or Instagram, there's a lot of intent data coming from that, from how you use it.
00:19:56.240 -->
00:19:57.759And should we just break down for the use?
00:19:57.759 -->
00:20:02.480So Meta is maybe not the biggest tech company in the world, but definitely top five, top ten.
00:20:02.480 -->
00:20:03.519Yep.
00:20:03.519 -->
00:20:08.960And they own Facebook, Instagram, WhatsApp, which is a big one.
00:20:08.960 -->
00:20:09.839What else have they got?
00:20:10.000 -->
00:20:12.799They're they're the three primary, you know, consumer-facing products.
00:20:13.039 -->
00:20:14.000For me, it's WhatsApp.
00:20:14.000 -->
00:20:20.480Like if you know what I'm not that it's particularly interesting, but it's probably stuff like, oh, I'm at the gym again.
00:20:20.480 -->
00:20:25.119Or like, you know, I'll meet you when I'm home from the gym, or Yeah, Hannah's arriving tomorrow.
00:20:25.200 -->
00:20:28.319We need to buy a present for Bob, you know, all of these kinds of things, right?
00:20:28.319 -->
00:20:35.039So you're you're providing lots and lots of context, whether through your use of Instagram, through your use of Facebook, through your use of WhatsApp.
00:20:35.039 -->
00:20:40.000So that's a that's a big signal, and you can figure out a lot about people's intent by their location, right?
00:20:40.000 -->
00:20:48.079So I don't have to say out loud, or maybe I've said out loud in my house, oh, I'm I really want to buy a new surfboard, and then suddenly I'm getting surfboard ads.
00:20:48.079 -->
00:20:52.240It could also be because I searched for surfboards, or I went to a surf shop.
00:20:52.240 -->
00:20:55.680It's not that Siri or Alexa or Google Home is listening.
00:20:55.920 -->
00:20:57.599Yeah, we're we're putting out signals, right?
00:20:57.599 -->
00:20:58.319That's the big thing.
00:20:58.319 -->
00:21:06.559And and we're putting out signals in what I'm texting you or WhatsApping you, what I'm searching for.
00:21:06.559 -->
00:21:09.759And then they're like, it's like how I'm scrolling through Instagram, right?
00:21:10.000 -->
00:21:10.880Yeah, absolutely.
00:21:10.880 -->
00:21:11.599Absolutely.
00:21:11.599 -->
00:21:16.799If you think about any feed-based tool, right, there's lots of clues you're giving it by interacting with the feed, right?
00:21:16.799 -->
00:21:18.799The most obvious one is if you tap on something, right?
00:21:18.799 -->
00:21:24.319So if you tap on something and expand the picture, you're you're showing pretty strong intent that that's something of interest to you.
00:21:24.319 -->
00:21:27.359If you scroll through the images, even more intent, right?
00:21:27.359 -->
00:21:28.960Like, hey, I'm super engaged with this.
00:21:28.960 -->
00:21:30.400So you're giving that signal back.
00:21:30.400 -->
00:21:33.519If you skip something really fast, it's a great signal, right?
00:21:33.519 -->
00:21:34.799I'm not interested in that.
00:21:34.799 -->
00:21:40.799If I pause on something, watch, you know, more than a reasonable amount of a video, tells it something.
00:21:40.799 -->
00:21:41.599If you like it.
00:21:41.599 -->
00:21:43.599If you like it, obviously massive signal.
00:21:43.599 -->
00:21:44.000Yeah, yeah.
00:21:44.000 -->
00:21:47.359If you share it with somebody else, huge signal and reshare something.
00:21:47.359 -->
00:21:52.240And message it to somebody else, you know, all of these kinds of things.
00:21:52.240 -->
00:21:59.759So you can imagine working on that team and building a model that understands people's intent based on positive things they do and also negative one of.
00:22:00.240 -->
00:22:05.680Things I learned early on in my career was this whole difference between reported behavior and actual behavior.
00:22:05.680 -->
00:22:13.119Like we can say that we do XYZ or eat XYZ, and actually, like that's not how we tend to behave.
00:22:13.119 -->
00:22:15.200We tend to report aspirational behavior.
00:22:15.200 -->
00:22:20.000But my Instagram scrolling, that'll reveal all actual intent.
00:22:20.000 -->
00:22:20.799Yeah, yeah, yeah.
00:22:20.880 -->
00:22:21.440That's it.
00:22:21.440 -->
00:22:22.160Absolutely.
00:22:22.160 -->
00:22:29.200And and sometimes, you know, these companies know what you're interested in before you even do, based on your behavior.
00:22:29.200 -->
00:22:37.839The other thing that, of course, your smartphone knows is its IP address, which is, you know, basically the internet connection that it's it's attached to right now, right?
00:22:37.839 -->
00:22:43.039So you're in my house right now in the podcast studio, you're connected to my Wi-Fi.
00:22:43.039 -->
00:22:46.559My Wi-Fi has a you know is connected to a cable modem.
00:22:46.559 -->
00:22:50.240That cable modem has a unique identifier that's a lot like a phone number.
00:22:50.240 -->
00:22:52.480Obviously, I'm also connected to that.
00:22:52.480 -->
00:22:56.160Your husband's connected to that, my wife's connected to that, right?
00:22:56.160 -->
00:23:02.559And so you've got other people in the house doing things effectively on the same phone number, on the same IP address.
00:23:02.559 -->
00:23:04.559And so that's a pretty interesting clue as well, right?
00:23:04.559 -->
00:23:14.160So if somebody in the family gets interested in surfboards, so suddenly, you know, I'm interested in surfboards, there's a fair chance that maybe you'd be interested in seeing a couple of surfboards as well.
00:23:14.160 -->
00:23:16.559And probably a better example might be something for the home.
00:23:16.559 -->
00:23:18.160I'm shopping for a car.
00:23:18.160 -->
00:23:24.240Uh might be a good idea to serve my wife pictures of cars because that's probably a family decision.
00:23:24.240 -->
00:23:29.759So there's lots of signals that are coming out of these devices that are really helping Target.
00:23:29.759 -->
00:23:41.680And uh, you know, it might put something in front of you that you've only just started to think about, or maybe you mentioned, and uh, and you know, you might come to the erroneous conclusion that the device was listening.
00:23:41.680 -->
00:23:48.799But the reality is, you know, there's a lot more signal coming out of your smartphone than probably, you know, you'd get out of recording your voice.
00:23:49.039 -->
00:23:49.920So here's a question.
00:23:49.920 -->
00:23:56.400One of my biggest actions on Instagram is like sharing reels with probably like three to five people on repeat.
00:23:56.400 -->
00:23:58.319You know who you are if you're one of them.
00:23:58.319 -->
00:24:05.920But so sometimes what they're searching or what they're signaling comes up on like my feed.
00:24:05.920 -->
00:24:09.920So a really good friend is like, oh, I'm I'm really getting into cold water swimming.
00:24:09.920 -->
00:24:13.519And then I'll be like, I have been getting served cold water swimming content.
00:24:13.519 -->
00:24:20.240Do you think it it knows that we're a very tight user group and that's oh my gosh, it's crazy.
00:24:20.480 -->
00:24:29.920And if you think back to, you know, look, I don't know what was in Mark Zuckerberg's head when he built Facebook, but one of the great innovations of Facebook was first of all, you had to be a real person.
00:24:29.920 -->
00:24:32.240When you sign up, we have to you have to be a real human.
00:24:32.240 -->
00:24:37.359So we're identifying you as an as an individual, which is unlike Twitter that's now become X, right?
00:24:37.359 -->
00:24:40.720Where you could sign up with multiple accounts, make up fake names, whatever it is.
00:24:40.720 -->
00:24:44.559So there's actual real humans was sort of the beginning of Facebook.
00:24:44.559 -->
00:24:46.079And then there was the friend graph.
00:24:46.079 -->
00:24:53.359You made friends with people, and those friends were friends with other people, and and so suddenly you had a graph of how the world is related to the world.
00:24:53.359 -->
00:24:53.839Yeah, right.
00:24:53.839 -->
00:25:00.319And you'd even you'd even say, This is my wife, this is my husband, it's my brother, you know, my relationship is complicated.
00:25:00.319 -->
00:25:03.359You're telling it all kinds of things about how that graph works.
00:25:03.359 -->
00:25:06.319You put people in lists called close friends, all those kinds of things.
00:25:06.559 -->
00:25:07.519Yeah, the top eight.
00:25:07.680 -->
00:25:08.000Yeah.
00:25:08.079 -->
00:25:11.519And you know, you interact with their posts and you see their posts.
00:25:11.519 -->
00:25:17.759And so basically you're you're giving the company you know incredible information about how you're related to other people.
00:25:17.759 -->
00:25:20.720And obviously, you know, you can be related through different paths.
00:25:20.720 -->
00:25:23.519If you're related through lots of different paths, then probably you're pretty close.
00:25:23.519 -->
00:25:24.079Yeah, yeah.
00:25:24.079 -->
00:25:30.240You engage with their posts and eventually you're able to infer things like if they're interested in it, then probably you are as well.
00:25:30.240 -->
00:25:32.240And again, remember location's a thing too.
00:25:32.240 -->
00:25:37.680So that probably knows that you go already go to the gym with this person and now suggesting open water swimming to both of you, right?
00:25:37.839 -->
00:25:43.359So there's nothing more jarring than seeing someone else's explore page on Instagram.
00:25:43.359 -->
00:25:48.240Like if I opened yours or you opened it, you'd probably be really disappointed if you open mine.
00:25:48.240 -->
00:25:53.200It's just like fitness content and videos of dogs or something.
00:25:53.200 -->
00:25:57.759But yeah, they're they're so tailored and it becomes so real to you.
00:25:57.759 -->
00:25:59.920It's like your whole world and your whole network.
00:25:59.920 -->
00:26:08.319But you pick two different people from two walks of life, then they'll have potentially super different interests and be pulling different content down from the algorithms.
00:26:08.559 -->
00:26:09.599Yeah, absolutely.
00:26:09.599 -->
00:26:15.519If you think about these these companies too, you know, like Google's ad business is an incredible business, right?
00:26:15.519 -->
00:26:20.319That's first of all built on the keywords you're searching for and secondly on where you are.
00:26:20.319 -->
00:26:27.759Metas is a much, much more effective ads business per dollar spend because they've got this graph of who you're connected to.
00:26:27.759 -->
00:26:35.519They've got lots of engagement data of you just using this thing for hours and hours a day, interacting with things, not interacting with things.
00:26:35.519 -->
00:26:38.319They've got the graph how you're connected to people, all of that.
00:26:38.640 -->
00:26:44.240It's like the world's biggest user group study that's going on at all times, everywhere.
00:26:44.480 -->
00:26:45.359Yep, absolutely.
00:26:45.359 -->
00:26:48.960And so it doesn't need to listen to uh to what you're saying.
00:26:48.960 -->
00:26:51.519It can probably do a better job than that.
00:26:51.759 -->
00:26:52.799Yeah, it definitely can.
00:26:52.799 -->
00:26:56.240What am I saying out loud versus what am I mindlessly scrolling?
00:26:56.240 -->
00:26:56.559Yep.
00:26:56.559 -->
00:27:03.759And Facebook less so, but WhatsApp, like these are apps I'm using more than I'd care to admit, and they're free.
00:27:03.759 -->
00:27:10.480I think I heard once on a documentary, it was like, if you're not paying for a product, you are the product and someone else is basically paying for you.
00:27:10.799 -->
00:27:12.480Yeah, I guess that's one way to think about it.
00:27:12.480 -->
00:27:15.359I mean, the other way to think about it is look, it's a straight trade here, right?
00:27:15.359 -->
00:27:23.440You get amazing apps in your pocket, things like Google Maps or Instagram that, you know, help you navigate the world, provide you with enjoyment, connect you to your friends.
00:27:23.440 -->
00:27:24.799You get those for free.
00:27:24.799 -->
00:27:34.000And so in return, you're agreeing to give these companies lots and lots of data that they can use to advertise things to you and you know, help advertisers sell their products.
00:27:34.000 -->
00:27:40.960And also, you know, you're giving it data that can help it do a better job of advertising to your friends, your family, everybody that you're connected with.
00:27:40.960 -->
00:27:45.519So it's an incredible ad network and probably one of the best ad products that's ever been built.
00:27:45.759 -->
00:27:52.799Well, I and I controversially love it because it helps me find the content and the products that I want to buy.
00:27:52.799 -->
00:27:54.640Um, sometimes I get carried away.
00:27:54.640 -->
00:28:01.759And controversially, like if Amazon was listening in my home, I mean, I'm not a very interesting person, and so they wouldn't be getting much out of me.
00:28:01.759 -->
00:28:07.200And maybe if they were listening into someone else's home, they'd discover criminal activity and disrupt it.
00:28:07.200 -->
00:28:09.839So I'm like pretty easygoing on privacy.
00:28:09.839 -->
00:28:14.079You'll probably tell me why that's a bad approach when we finish recording.
00:28:14.319 -->
00:28:17.440Well, I just say we're all past the point of no return, perhaps.
00:28:17.440 -->
00:28:26.000Yeah, like like we're all we're all using these products and we're using them an enormous amount of the time every day and giving these companies a huge amount of data.
00:28:26.000 -->
00:28:34.079And so, you know, whether or not we let them record the occasional sentence that we say or whatever it is is, you know, maybe by the by at this point.
00:28:34.079 -->
00:28:45.839And you know, maybe that's a maybe it's a sad situation we've let the world get into, but the reality is, you know, you you're giving these companies pretty much everything that uh that they could possibly know about you, and they probably just don't need a voice at this point.
00:28:46.160 -->
00:28:58.319No, that's my biggest takeaway of this episode so far is like what I say out loud in conversation at home is actually much less useful than what you hover over as you scroll down your Instagram feed.
00:28:58.319 -->
00:28:58.640Yeah.
00:28:58.640 -->
00:29:04.319And there's no way of how could I cut that feed if I really decided I had a problem with it?
00:29:04.319 -->
00:29:10.319I mean, Instagram I could I could give up and probably read a few more books if I did.
00:29:10.319 -->
00:29:12.400And I don't really use Facebook.
00:29:12.400 -->
00:29:16.319I maybe write down a few people's birthdays or something and probably just get rid of that app.
00:29:16.319 -->
00:29:28.559But WhatsApp's a tricky one because that's probably my most used and the one that I would pay a subscription to use WhatsApp because I use it to I use it to talk to you, I use it talk to everyone all the time at all times, right?
00:29:28.720 -->
00:29:32.799Yeah, I'd say it's less of an Australian thing for um probably for our Australian listeners.
00:29:32.799 -->
00:29:34.640I'd say there's a lot more folks use iMessages.
00:29:34.960 -->
00:29:37.440That's like in the US, they find it really cute that we all use.
00:29:37.440 -->
00:29:38.240Yeah, yeah.
00:29:38.319 -->
00:29:46.480But I'd say for you know, Europe, South America, um, probably Africa, I'd say WhatsApp's probably dominant, being a bit different in different jurisdictions.
00:29:46.480 -->
00:29:47.839But you know, the point stands, right?
00:29:47.839 -->
00:29:51.359Is that these that these have become a primary way that that people communicate.
00:29:51.359 -->
00:29:54.799I know when I lived in the US, nobody called me and I never called anybody.
00:29:54.799 -->
00:29:56.160I mean, we just messaged the whole time.
00:29:56.160 -->
00:30:04.559I'd say, I'd say Australians pick up the phone a little bit more, but but still, you know, the primary way of communicating is is using texting of some sort, right?
00:30:04.799 -->
00:30:10.160And so the the takeaway here is we're feeding it's it's called metadata, right?
00:30:10.160 -->
00:30:15.200Metadata is also just like the whole of all the data that we're pumping into these apps at all the time, right?
00:30:15.359 -->
00:30:23.039Yeah, look, probably the definition of metadata is to say it's it's you know additional information that's added in addition to the actual core data that you're sending.
00:30:23.039 -->
00:30:35.519So let's imagine, for example, that um I type in a search into Google, the data would be the search, and the metadata would be things like where I'm located, the time of day, how fast I type the words, you know, these kinds of things.
00:30:35.519 -->
00:30:35.839Oh my god.
00:30:35.839 -->
00:30:37.279That'd be the additional data.
00:30:37.759 -->
00:30:41.039It will know what I'm doing as soon as I wake up.
00:30:41.039 -->
00:30:42.319That is crazy.
00:30:42.319 -->
00:30:46.720I I wonder if if you could pay for like a report back.
00:30:46.720 -->
00:30:51.920Everyone obviously likes to think about themselves, myself included, but you could get that back.
00:30:51.920 -->
00:30:56.480Kind of like how you on the Amazon app, you can like get all of your prompts back.
00:30:56.480 -->
00:31:14.799If I could get the sum total of like Instagram, WhatsApp, et cetera, like meta could give me a breakdown of like my habits and like where I can improve, and like you need to not get up and open Instagram, like you need to be getting up and getting out of bed, or something like that or you spend a lot more time on fitness videos than you see seem to be doing at the gym or whatever.
00:31:14.799 -->
00:31:17.680But that they could, I mean, people would probably buy that, right?
00:31:17.839 -->
00:31:32.559There's some probably some product managers, you know, have thought about those kinds of things, and I guess that's the basis of companies like Whoop, um, you know, Garmin's features, those kinds of things really is trying to change people's habits based on, again, collecting an enormous amount of data about people and metadata about people.
00:31:32.880 -->
00:31:42.319We should do more on on like fitness data later in this series, but um, it also makes me think of Spotify Wrapped, which has been going for years now, and everyone loves to.
00:31:42.319 -->
00:31:44.880I mean, mine always just says that I listen to Tailswift.
00:31:44.880 -->
00:31:46.160I'm so basic.
00:31:46.160 -->
00:31:53.200Um, but they also this year, I don't know if it was there's something about this year, every app was doing a rapt, yeah.
00:31:53.200 -->
00:32:00.240Which is it's kind of cool, although some of them were a little bit of a stretch, but it's that whole idea of like, here's you played back to you.
00:32:00.240 -->
00:32:07.599And it's I think no one really cares about your Spotify wrapped other than you, and yeah, everyone's like pushing it out onto their Instagram or whatever.
00:32:07.759 -->
00:32:08.559Yeah, absolutely.
00:32:08.559 -->
00:32:12.000And uh the folks at OpenAI build chat GPT did one too.
00:32:12.160 -->
00:32:12.799Oh, yeah, that's right.
00:32:13.039 -->
00:32:19.440Yeah, I think I made the mistake of um sharing the screen with you as I looked at it for the first time and probably learned more about me than you than you wanted to know.
00:32:19.519 -->
00:32:25.039But uh I learned that Hugh likes protein recipes but doesn't like eggs for breakfast.
00:32:25.039 -->
00:32:25.440Is that right?
00:32:25.599 -->
00:32:26.480Yeah, that's exactly right.
00:32:26.480 -->
00:32:27.200Yeah, yeah, yeah.
00:32:27.359 -->
00:32:28.079I need to look at mine.
00:32:28.079 -->
00:32:33.759It's that's the thing that we'll just be like my chat GPT is gonna tell me that I'm like a hypochondriac.
00:32:33.759 -->
00:32:34.799I'm just always Googling.
00:32:34.799 -->
00:32:35.599Medicol.
00:32:35.599 -->
00:32:36.160Yeah, okay.
00:32:36.160 -->
00:32:37.440So much medical chat.
00:32:37.440 -->
00:32:40.720I've revealed a lot about myself on this episode already.
00:32:40.720 -->
00:32:48.720So thinking about Spotify, do they have, do you think they have my wrapped from like 2018, or do they just get rid of that data after a year or something?
00:32:49.039 -->
00:32:55.359So it really depends on the jurisdiction as to what how long you can keep the data and and what data you can keep.
00:32:55.359 -->
00:33:01.359Um, but I'd say, you know, if you work at one of these large tech companies, you know, you want as much data as you can possibly get your hands on.
00:33:01.359 -->
00:33:03.680I mean, I've been in this situation when I worked at Microsoft.
00:33:03.680 -->
00:33:13.680I mean, I wanted to keep every single query that every user had ever typed associated with that user and all of their behavioral data of what they clicked on and what they didn't click on.
00:33:13.680 -->
00:33:27.920And the reason I wanted to keep that's not so I can study the user or particularly know anything about a particular user, but it's because if you train with all of that data and build an AI model, um, you can have a better search experience for future users if you've got more data.
00:33:28.160 -->
00:33:30.319Okay, so it's the aggregate that's what they're really after.
00:33:30.480 -->
00:33:30.640Yeah.
00:33:30.640 -->
00:33:36.160So the more data you've got, the better your search algorithms will be, or your Instagram feed algorithms or whatever it is.
00:33:36.160 -->
00:33:43.599So all of these tech companies want to keep all of the data forever, but you know, different jurisdictions place different constraints on the organizations.
00:33:43.599 -->
00:33:50.000And the way the world works, at least at the moment, is that the EU probably has the strictest privacy standards.
00:33:50.000 -->
00:34:00.400I think, you know, you typically see the EU move first and say, hey, we don't want you to keep data for more than this period of time, or we want it to be more anonymized, or we want to give the users more options, more control.
00:34:00.400 -->
00:34:05.359The tech companies, I guess, come to the table and try and negotiate a position that suits them.
00:34:05.359 -->
00:34:09.199But again, in their minds, they want to keep as much data as they possibly can for as long as they can.
00:34:09.440 -->
00:34:17.199But it's interesting that you say, for people listening who are getting freaked out by this episode, it's really not about what I, Hannah, or you, Hugh are doing.
00:34:17.199 -->
00:34:22.239It's about the collective habits of a number of people over a long amount of time.
00:34:22.480 -->
00:34:23.119Yeah, exactly.
00:34:23.119 -->
00:34:26.400And it's particularly important for what we call in the tale.
00:34:26.400 -->
00:34:33.280So, you know, it's pretty easy to figure out things that are popular and serve popular things to people, right?
00:34:33.280 -->
00:34:37.199So if there's a, I don't know, popular pair of shorts right now that lots of people look like.
00:34:37.760 -->
00:34:38.719They don't need to put it on.
00:34:38.719 -->
00:34:48.639Like, that's like Marks and Spencer's who we did a an episode on last season about their hack, but like one of the things that really annoys me that they do is they push a product and I love it.
00:34:48.639 -->
00:34:51.119And then I click through and I can't buy it because it's sold out.
00:34:51.119 -->
00:34:57.360So it's like, don't push me the stuff that's selling well, because then I'll just get frustrated when I click through and it's sold out.
00:34:57.599 -->
00:34:58.880Yeah, yeah, that's definitely a problem.
00:34:58.880 -->
00:35:05.840Though a lot of the companies do push things that are sold out because it helps them get a signal about how many users are still interested.
00:35:05.840 -->
00:35:08.800So when they do their next production run, they know how many to make.
00:35:08.800 -->
00:35:14.559So having worked in a fashion company, I know that pushing things that are actually sold out is a good way to get a signal of how it's gonna be.
00:35:14.559 -->
00:35:15.599Scarcity, okay.
00:35:15.840 -->
00:35:17.360Yeah, it makes me want it more as well.
00:35:17.360 -->
00:35:20.480Because I'm like, oh, I knew it was nice, and everyone else thought it was nice too.
00:35:20.639 -->
00:35:20.960Yeah.
00:35:20.960 -->
00:35:22.320Back to the EU, right?
00:35:22.320 -->
00:35:26.719So the EU wants to protect user privacy, wants to give users more control.
00:35:26.719 -->
00:35:31.599You know, the tech companies come to the table and they try and advocate a position that suits them.
00:35:31.599 -->
00:35:36.480But what's in their mind, of course, is we want to keep the data for as long as we can and as much as we possibly can.
00:35:36.480 -->
00:35:41.119And there's a bit of back and forth with, you know, different EU privacy commissioners or whoever it is.
00:35:41.119 -->
00:35:43.599You go back and forth, eventually you reach some agreement.
00:35:43.599 -->
00:35:48.320What typically then happens is that the US kind of fast follows, maybe a couple of years later.
00:35:48.559 -->
00:35:50.639Even under this administration, you think they would?
00:35:50.880 -->
00:35:51.920Well, I'm not sure.
00:35:51.920 -->
00:35:54.960I mean, I I I think, you know, historically that's been the case.
00:35:54.960 -->
00:36:00.159I guess we'll we'll know in a we'll know in a year or two how the current administration, you know, thinks about these things.
00:36:00.159 -->
00:36:02.880But I'd say in general, the EU is the trendsetter.
00:36:02.880 -->
00:36:09.760You know, you've got places like Australia who are pretty aggressive, but the EU tends to set some precedent and then eventually everybody follows.
00:36:09.760 -->
00:36:15.119So you find the tech companies working pretty closely, pretty aggressively with the EU to try and advocate their position.
00:36:15.360 -->
00:36:19.599And even China, like China's obviously a little bit more lax on that.
00:36:19.599 -->
00:36:21.840They probably heard this me say.
00:36:22.960 -->
00:36:26.559I'd say, uh, I'd say that's just a completely different space.
00:36:26.559 -->
00:36:35.920You know, the the EU, the UK, Australia, the US, you know, have sort of rules-based systems where they create laws about what these companies can and can't do.
00:36:35.920 -->
00:36:44.880I'd say in China, you know, largely the government does what the government wants, and largely the government has access to most of the most of the data that most of the companies have there.
00:36:44.880 -->
00:36:46.880So it's quite a different sort of environment.
00:36:46.880 -->
00:36:50.320I want to just finish off that point about the head and the tail and things, right?
00:36:50.320 -->
00:36:59.519So, you know, I gave a bad example of saying, look, you know, it's not that hard to figure out what shorts to show you, you know, we know where you've been, what the weather's like, the kinds of things that you're searching for.
00:36:59.519 -->
00:37:00.719So it's not a hard problem.
00:37:00.719 -->
00:37:10.400What's a hard problem is is when you've got that sort of one-off, really detailed requirement that you've never had in your life before, that not too many people have.
00:37:10.400 -->
00:37:23.519So, you know, imagine I know you don't own a car, but you know, imagine you had a car, it's an old car, you're having problems with some component in the car, you know, the cassette radio is not working, and you want to buy a new knob for the cassette radio, right?
00:37:23.519 -->
00:37:29.679So you go into Google and you type, you know, cassette radio, Nissan Pulsar 1983 knob.
00:37:29.679 -->
00:37:38.400That's a pretty obscure query, but there's gonna be some user somewhere in the history of time who searched for exactly that query or something very, very close.
00:37:38.400 -->
00:37:43.760Maybe they search for the 1982 model and they've clicked on some results and they've skipped some results.
00:37:43.760 -->
00:37:50.400If we're able to keep that data, then we can do a pretty good job of answering that query for you because we know what's a good answer and what's not.
00:37:50.400 -->
00:37:55.119But you need a lot of data to answer a query like that to train a model and deliver great answers.
00:37:55.119 -->
00:37:57.119Because people aren't searching it that much.
00:37:57.119 -->
00:37:57.760That's right.
00:37:57.760 -->
00:37:59.039So it's very, very rare.
00:37:59.039 -->
00:38:14.400You know, that's one of the things that makes Google an incredibly successful company is that they've got lots of data and they're able to answer the hard queries that you really care about, whether it's about, you know, buying an obscure part for your car or some medical condition that, you know, a friend's got that's really obscure.
00:38:14.400 -->
00:38:15.679Um, you need lots of data for that.
00:38:15.679 -->
00:38:20.719And so that's one of the reasons that in the ads business, one of the reasons why these companies keep a lot of data.
00:38:21.039 -->
00:38:29.519The revenue that a company like Meta is that must be like the lion's share or at least the majority of their revenue income, right?
00:38:29.599 -->
00:38:34.480Maybe we should we could we should break this down in a future episode, but we should talk ads one day and sort of exactly how they work.
00:38:34.480 -->
00:38:35.519But but yeah, absolutely.
00:38:35.519 -->
00:38:45.519I mean, these are you know, Google and Meta, who own Facebook and Instagram, have two of the best businesses that have ever been built because their ads businesses are are incredible.
00:38:45.519 -->
00:38:49.119They do a great job of serving content to you that perhaps you don't even know you want.
00:38:49.119 -->
00:38:51.119You click, you make the advertisers happy.
00:38:51.760 -->
00:38:54.320You hear people say it like, oh, I love that jumper.
00:38:54.320 -->
00:38:56.639Oh yeah, Instagram served it to me.
00:38:56.800 -->
00:38:56.960Yeah.
00:38:57.440 -->
00:39:02.400And then they use the money they get from the ads to pump in to make the product even better, so you keep using it.
00:39:02.400 -->
00:39:02.639Yeah.
00:39:02.800 -->
00:39:03.360Yeah, yeah.
00:39:03.920 -->
00:39:05.119It's pretty obvious when you think about it.
00:39:05.119 -->
00:39:07.519And to be honest, like I've said a few times, I don't really mind.
00:39:07.519 -->
00:39:16.960I find it to be quite helpful in the modern world that we live in, but that's probably one fairly lax viewpoint on a spectrum where other people would feel very differently.
00:39:17.519 -->
00:39:18.239Yeah, absolutely.
00:39:18.719 -->
00:39:19.760Can I ask a question then?
00:39:19.760 -->
00:39:33.679So I know that companies aren't interested in necessarily you as an individual, but like if I was a criminal and the government wanted to like, or the police or whomever wanted to know about my habit, can they get that from the tech companies?
00:39:33.679 -->
00:39:35.440Can they like is that called subpoenaing?
00:39:35.440 -->
00:39:38.880Is that is can they just say you need a handover so we know what they've been doing?
00:39:39.119 -->
00:39:47.360Yeah, the companies are approached by law enforcement all the time, you know, asked to provide data of different types and they have a lot of processes that they have to follow.
00:39:47.360 -->
00:39:49.599Again, they have to operate within the law.
00:39:49.599 -->
00:39:53.679The law enforcement organizations have to provide the right warrants or whatever it is.
00:39:53.679 -->
00:40:00.960But yeah, absolutely, you know, this data is often used to solve crimes, um, prosecute people, whatever else it is.
00:40:00.960 -->
00:40:10.239And I guess it's you know it's similar to the old days with the paper version, you know, where somebody knock on your door with a warrant and say, We're gonna we're gonna come in and take your office away in a truck and look through all of your papers.
00:40:10.239 -->
00:40:11.920So there's a digital version of that as well.
00:40:12.159 -->
00:40:18.800Would a company like Meta ever have flags that meant that they would want to turn someone over to law enforcement?
00:40:19.360 -->
00:40:22.159Um no, no, I don't think so.
00:40:22.159 -->
00:40:24.400I'd say, look, these companies don't think they're the police.
00:40:24.559 -->
00:40:24.800Yeah, yeah.
00:40:25.199 -->
00:40:27.360They leave they leave law enforcement to law enforcement.
00:40:27.360 -->
00:40:33.360Obviously, they have to make sure that illegal things don't happen on their platforms within whatever the legal constraints are.
00:40:33.360 -->
00:40:38.320So, you know, they have to prevent, you know, people selling weapons online or whatever else it is.
00:40:38.320 -->
00:40:44.480So there's all sorts of, you know, interesting legal constraints around these companies, and they have to put in their best effort to do that.
00:40:44.719 -->
00:40:52.800Okay, but there there are examples of some stuff being done for the greater good, not just the greater good of whoever I'm buying my workout clothes from.
00:40:52.800 -->
00:41:15.280But uh, I have a friend who worked in marketing for a cancer charity, and she was telling me about how they did like a collab with Boots, which is the biggest pharmacy in the UK, and they were able to discern from people's shopping habits that they needed to be checked for symptoms before they even realized there was a problem themselves, based on like what medication they were buying to alleviate symptoms.
00:41:15.519 -->
00:41:16.000Yeah, absolutely.
00:41:16.000 -->
00:41:23.519And again, you know, you've got all this data about people's behavior, you know, when they wake up, what they do, where they go, the things they search for.
00:41:23.519 -->
00:41:28.239I mean, you can use that data to to figure out all kinds of things about people, including medical conditions.
00:41:28.239 -->
00:41:36.239In fact, um, I don't know if I told you this story, but when I worked at Microsoft, early days of Microsoft working on search, it's probably about, I don't know, let's call it 2007.
00:41:36.239 -->
00:41:39.039I was really interested in queries and clicks.
00:41:39.039 -->
00:41:42.559So what I mean by that is when somebody runs a query, what do they what do they click on?
00:41:42.559 -->
00:41:47.760And how do you use that click data to inform, you know, better answers for future customers?
00:41:47.760 -->
00:41:49.360So I was really, really interested in this stuff.
00:41:49.360 -->
00:41:52.800And um, I thought as an interesting experiment, we had a list of data.
00:41:52.800 -->
00:42:00.320I thought I'd look for people who were searching for particular drugs, you know, prescription drugs that that you can that you can legally buy.
00:42:00.320 -->
00:42:06.880And then I'd have a look at the symptoms that they'd search for in the same session or in other sessions, right?
00:42:06.880 -->
00:42:09.039So let's imagine you buy some amino flames.
00:42:09.360 -->
00:42:11.119Welatonin is a great example.
00:42:11.119 -->
00:42:20.480We were chatting this morning and my husband was probably yeah, and he was like, I took some melatonin on the plane, and then I've got some chest pain, and then I Googled it, and it turns out that that can be a side effect.
00:42:20.480 -->
00:42:26.239So good news, I don't have a chat, a heart problem, but I am not I'm gonna stop taking the melatonin from my jet lag.
00:42:26.400 -->
00:42:26.960Yeah, absolutely.
00:42:26.960 -->
00:42:27.519Great example.
00:42:27.519 -->
00:42:30.159Or, you know, you take anti-inflammatories and you get stomach crank, yeah.
00:42:30.159 -->
00:42:38.880Yeah, and so I basically looked for some brand name drugs, and then I looked for symptoms, you know, write some simple code to go and figure these things out.
00:42:38.880 -->
00:42:46.480And then I thought, I'm gonna go and look up the FDA documentation and try and figure out if these are acknowledged side effects.
00:42:46.480 -->
00:42:48.559And of course, yeah, most of them were, right?
00:42:48.559 -->
00:42:54.159So you know, figure out people who search for a particular brand of anti-inflammatories would later search for stomach cramps.
00:42:54.159 -->
00:42:56.719You say, yeah, well, that's a common side effect of these things.
00:42:56.719 -->
00:43:03.599But I also found some side effects or what could have been side effects that weren't yet recognized side effects.
00:43:03.599 -->
00:43:13.599You know, I said to my boss at the time, um, I bet these really are side effects of these drugs because so many people who are searching for these drugs are also searching for these symptoms, right?
00:43:13.599 -->
00:43:21.039So if you've got millions and millions of people searching for a particular drug and then searching for a side effect, probably a side effect.
00:43:21.360 -->
00:43:21.679Wow.
00:43:21.679 -->
00:43:36.960Okay, so you knew it before I guess because you can only trial a certain sample, which I'm sure is extensive for safety, but actually, like I said up front, like this is the biggest live user trial happening at all times.
00:43:36.960 -->
00:43:41.920And okay, so would they ever, would Google ever like sell that back to the drug companies?
00:43:42.639 -->
00:43:43.199I wouldn't think so.
00:43:43.199 -->
00:43:43.760No, okay.
00:43:43.760 -->
00:43:44.559I wouldn't think so.
00:43:44.559 -->
00:43:46.559You know, that's not their line of business, right?
00:43:46.559 -->
00:43:59.599But I think I guess it just speaks to the fact that if you can keep a lot of this data, keep it associated with a user, you've got enough users, millions of users, billions of users in some cases, you can figure out these kinds.
00:43:59.599 -->
00:44:03.440of things in a way that you know no clinical trial will ever figure out.
00:44:03.679 -->
00:44:03.840Yeah.
00:44:03.840 -->
00:44:11.599Or like it will, if I search a brand, I'll get similar brands that aren't the brand I searched for on my ads, which is again super useful.
00:44:11.599 -->
00:44:17.840Because it's like, oh, we hear you and now we're serving you something similar based on what other people were interested in.
00:44:18.159 -->
00:44:19.840The Instagram ads team's going to love you.
00:44:19.840 -->
00:44:21.840Like uh because I love them.
00:44:21.840 -->
00:44:22.400Yeah.
00:44:22.400 -->
00:44:25.119And then and if you work on an ads team, that's what you want, right?
00:44:25.119 -->
00:44:32.239Is you you don't want to be the annoying, you know, TV ad that's shown 10 times every hour in the middle of a sports event that everybody's sick of.
00:44:32.239 -->
00:44:37.039You want to be an ad that's highly relevant, highly targeted, timely, and exactly what you're looking for.
00:44:37.039 -->
00:44:38.639It's on the story and that's what you want to be.
00:44:38.960 -->
00:44:39.440That's what I want.
00:44:39.440 -->
00:44:43.199I want on the stories when I can either flip past it or engage with it.
00:44:43.199 -->
00:44:47.679Instagram ad team if you're listening, I don't like when you serve me ads on the feed.
00:44:47.679 -->
00:44:51.519So if you could just sort that out, that'd be great since I've given you so much good promo.
00:44:51.519 -->
00:44:52.079Awesome.
00:44:52.079 -->
00:44:57.199Okay, so before we wrap up, I have revealed quite a lot about myself to you and the listeners.
00:44:57.199 -->
00:45:00.480I would love to know who are do you have you use Instagram or not so much?
00:45:00.480 -->
00:45:02.559Like what are your targeted ad groups?
00:45:02.800 -->
00:45:04.079Oh that's a great question.
00:45:04.079 -->
00:45:16.159I would say I follow a lot of musicians uh on Instagram and so I get a lot of ads for you know vinyl records that reissues of things, you know, bands I'm interested in.
00:45:16.159 -->
00:45:22.400So I'd say look, you know, I can when I do go use Instagram, I'd say the ads are highly relevant to me.
00:45:22.400 -->
00:45:24.639But I'd say I'm kind of disciplined with Instagram.
00:45:24.639 -->
00:45:26.719I um I actually only use it when I'm traveling.
00:45:26.719 -->
00:45:34.559So whenever I'm traveling I try and post one picture a day sort of make a visual record of the trip that I'm on and then I delete the app and wait till my next trip.
00:45:34.719 -->
00:45:38.000Guys Hugh is more disciplined and clearly much cooler than me as well.
00:45:38.000 -->
00:45:40.480Oh not sure final records and live meetings.
00:45:42.400 -->
00:45:57.519But uh we can talk about my use of LinkedIn which is extensive and often so I wouldn't say I'm social media free but I probably uh you know hang around more on the professional networks which makes me a lot less cool than you actually well swings and roundabouts I guess but I've revealed myself to be incredibly basic.
00:45:57.760 -->
00:46:11.199So in summary don't worry about Alexa or Google or your phone listening to you but maybe think about how much other data you're giving them via pretty much everything else you're doing on your phone.
00:46:11.440 -->
00:46:14.079Yep absolutely Hannah so they're really not listening.
00:46:14.079 -->
00:46:39.360Alexa's not listening Siri's not listening Google Home is not listening but you're probably giving more data and more metadata than you realize with things like your search queries, your use of LLMs, scrolling through feeds, your location is a huge factor and you know you're handing all of these companies all of your behavior and your family's behavior and your friends' behavior and that's how they they can do such a great job of building the businesses that they've built.
00:46:39.599 -->
00:46:56.320But I like how you framed it when you described it as like it's an exchange things like Google Maps, WhatsApp, those have drastically improved my day-to-day life and so them extracting my data back seems fair enough, but it'll be interesting to hear back from the listeners on how they sit on that specific point.
00:46:56.559 -->
00:46:57.280Yeah, absolutely.
00:46:57.280 -->
00:47:01.519It's been great to be back in the chair Hannah welcome to season two this is great fun.
00:47:01.760 -->
00:47:10.159Season two I've learned an incredible amount already and I'm sorry that I revealed so much about myself to listeners but uh it gives you a sense of who we are.
00:47:10.320 -->
00:47:15.440Yeah and if you're excited about the Tech Overflow podcast you can find us at techoverflowpodcast.com.
00:47:15.440 -->
00:47:16.880We're also on the socials.
00:47:17.199 -->
00:47:25.760We're on Instagram indeed and X and Hugh's favorite LinkedIn you can also find us wherever you get your podcast.
00:47:25.760 -->
00:47:30.239Please please like subscribe review and share with your friends colleagues family.
00:47:30.400 -->
00:47:40.000We're really excited about season two and we want to help as many people learn about tech as we can we do indeed so thanks to all the curious people who've joined us today and we'll see you soon for episode two.
00:47:40.000 -->
00:47:41.599We'll see you next time
