Is AI pushing our planet too far? | BBC News

0:02 [music]

0:06 Hello and welcome to the program.

0:09 The AI revolution runs on electricity.

0:13 Lots of it, vast amounts of it.

0:15 The data centers being built right now to power the next

0:18 generation of AI consume as much energy as a small city.

0:23 Last year, the world's biggest tech companies

0:25 spent more than 400 billion dollars building them.

0:30 And they need more of them.

0:31 Many more of them.

0:32 But there's a problem.

0:33 In the United States, four in 10 of the data centers that are being

0:37 planned for this year are at serious risk of delay.

0:41 Not enough power, not enough equipment, and not enough people to build them.

0:46 Why?

0:48 Also on the program this week,

0:49 Coachella, the world's most glamorous music festival.

0:53 But some of the online influencers you may have seen

0:56 pictured in the California desert alongside the biggest stars, were fake.

1:01 Not real.

1:01 They don't exist.

1:02 AI generated to promote brands and to make money.

1:06 Plus, we'll also talk this week about the late Val

1:09 Kilmer appearing in a new film one year after his death.

1:14 With us this week to talk about it,

1:16 Palmi Olsen, who is a technology columnist at Bloomberg.

1:20 Welcome to you.

1:20 Thank you.

1:21 Also here, Dr.

1:22 Sasha Luccioni, a computer scientist specializing

1:25 in AI and its environmental impact.

1:28 And also in the studio with us, good to have back here, Dr.

1:32 Stephanie Hare, colleague, author, and AI expert to give you your full title.

1:37 Um [snorts] Palmi, let us start with this issue of data centers.

1:41 We're building lots of them.

1:43 They are powering the AI revolution.

1:46 Why are so many of them on hold?

1:48 Yeah, so this is really an issue of bottlenecks

1:51 and you sort of alluded to it in your introduction.

1:54 There's the issue of power.

1:56 So getting access to actual power on electricity grids that are very,

2:00 very old and haven't received a lot

2:01 of investment over many decades in the United States.

2:04 Um also equipment, getting access to things like transformers or switch gear,

2:09 which are the types of equipment you need to actually run data centers.

2:12 There's also a huge bottleneck getting them.

2:14 It could take up to 5 years to get some of that equipment,

2:17 particularly cuz a lot of it comes from China.

2:19 And recent US tariffs on Chinese goods has made that even more difficult.

2:23 And then there's also just getting the people, the talent, um the electricians,

2:28 um and the people with the skillset who

2:29 can actually uh construct and run these data centers.

2:33 Um so, all those things combined have meant that at a time

2:37 when there is this rapacious demand for um for energy from tech companies,

2:43 it's actually very difficult to build them fast enough.

2:46 Mhm.

2:47 Does that problem become more acute in the energy crisis we're in currently?

2:51 I think it does to some extent and it's

2:53 more of a problem for the so-called hyperscalers,

2:56 the big tech companies like Meta, um Alphabet, Amazon, etc.

3:02 who are actually the ones who have to shell out for these energy costs.

3:06 And so, they're setting up um these kind

3:09 of mini nuclear reactors that can actually

3:12 provide um energy specifically for the data center

3:16 um and use renewable sources instead of gas.

3:19 But at the moment, even getting those up

3:21 and running is logistically very, very difficult.

3:24 It's time-consuming.

3:25 There aren't any actually that are operational just yet.

3:28 And and so, right now I think the main source of energy is gas.

3:32 Uh should we give our viewers just a a scale of what we're talking about here?

3:36 Scott Galloway, who might be coming on the program next week,

3:38 Avery in New York, he says, "Open AI alone, Stephanie,

3:43 will need 20% of current US electricity capacity at 10 trillion dollars." Mhm.

3:50 That's extraordinary.

3:51 It's extraordinary for a company that is yet to turn a profit and which

3:54 is under huge pressure to demonstrate value ahead of an IPO as well.

3:59 So, it's just worth saying that all of these companies, not just Open AI,

4:03 are making very big promises about their data

4:06 structure build out plans as of last year.

4:09 We've already seen the US-UK technology deal.

4:12 That's on hold.

4:13 We've seen Open AI have to pull out of some of its stargate,

4:16 the big US data structure data center infrastructure plan.

4:20 They've had to pull out on some of those things.

4:22 Maybe we are walking back.

4:24 So, that question of will the AI bubble burst?

4:27 It might not burst.

4:27 It might just sort of slightly deflate.

4:29 isn't it?

4:30 When you think that actually if you're going to plug

4:31 these into the grid and you don't know whether

4:35 these companies are going to survive in the scale

4:38 or perhaps they're even bigger than they are right now.

4:40 It It It's a difficult thing to plan.

4:44 It's a difficult thing to plan when you're not a planned economy,

4:46 which is why when we're looking at the data center rollout in a country

4:49 like China and comparing that to the United States or indeed here in Europe,

4:53 we get very different pictures.

4:55 Right.

4:55 Um Dr.

4:56 Luciani, Sasha, I'm going to call you Sasha.

4:58 Does it make sense for the US president to be so

5:01 vehemently opposed to renewable energy given the scale that we're talking?

5:07 Well, the problem is is that the data centers are being

5:10 built so quickly that the renewable capacity has trouble keeping up,

5:13 especially in rural areas,

5:15 especially at outside of places where renewables are are are typically the case.

5:20 So, I think that currently the emphasis is build faster,

5:23 build bigger and they don't want to wait around for solar or wind,

5:27 which is why um essentially most uh of the data centers that are coming

5:31 online as quickly are essentially bringing in turbines

5:33 on the back of trucks, natural gas.

5:35 It's It's like bring your bring your own energy uh

5:37 essentially and most of that is And they have the money.

5:40 I mean, these companies have the money.

5:43 They have the money,

5:43 but actually currently there's a bottleneck even when you have the money

5:46 because there's not enough turbines to to power all these data

5:48 centers because there's there's a backlog nowadays and and even

5:52 these these turbines can't be produced fast enough to to respond to demand.

5:56 Palmi, at Bloomberg, recently you highlighted an issue in northern

6:00 Spain with the data center build out there,

6:03 which has actually been held up as a a model for the rest of Europe.

6:06 But for the people who live around these projects,

6:09 the reality is sometimes very different.

6:11 Why?

6:12 I think it's a common story we're also seeing in the United States.

6:15 A lot of pushback from local residents

6:17 in areas where companies want to build data centers.

6:19 And in northern Spain, um there have been uh the situation is that AWS,

6:24 which is the cloud business of Amazon,

6:27 um sent letters to local people um saying we want to buy your land,

6:32 giving them sometimes 4 days notice to say yes or no.

6:36 Um and some of these people in in in northern

6:39 Spain actually thought it was a scam at first.

6:41 Um one lady went to her local town hall and even they didn't know.

6:45 So, it's a real um kind of land grab almost to try and get land that is

6:50 relatively cheap in an area where energy costs are

6:54 relatively low and that are sparsely populated as well.

6:57 It seems like an ideal situation for building a data center.

7:01 But at the same time, there is the reality for people who do live there

7:04 and there are people who live there um that they have

7:07 to give up that land or maybe the suddenly you've got

7:10 this eyesore in a place that you've lived in for many generations.

7:13 If you're in a community like that though and you've already struggled to to get

7:17 natural resources or to to to get electricity to get yourselves on the grid,

7:22 does the arrival of a big AI company help in that process?

7:26 Perhaps it it could help a community.

7:28 In some respects, and the funny part in that is that governments um

7:33 local governments often uh frame data center build outs as being great for jobs.

7:38 Yeah.

7:38 But I think you're conflating

7:40 in that situation permanent jobs with construction jobs, which are temporary.

7:45 And so, when you build out a data center, you're going to hire maybe

7:47 not Are they not necessarily big employers once the kit is there?

7:50 No, I think in a typical data center,

7:52 you might have about 100 people, most of them cleaners and security people.

7:56 Um whereas for the build-out, sure, hundreds,

7:58 maybe thousands of people, but then that's only temporary.

8:01 Mhm.

8:01 All right, I'm going to bring in an audience question quite early

8:04 in the program this week because it it it it fits what you're talking about.

8:07 It's from James in the UK.

8:08 He says, "Sasha, AI companies continue

8:11 to minimize this environmental impact." Uh he

8:14 points specifically to Sam Altman's recent

8:16 claim that AI's water usage is minimal.

8:19 James says, "That's simply not true." He also tells us

8:22 that younger generations are increasingly

8:24 boycotting generative AI for environmental reasons.

8:27 So, here's his question, "Should mainstream media be doing more

8:31 to hold these companies to account?" Definitely.

8:35 Actually, a recent Guardian study found that uh

8:38 the big tech companies were lobbying very,

8:40 very hard against transparency to make sure that I mean,

8:43 citing confidentiality to not include any uh

8:46 energy figures or water figures about data centers.

8:49 And so, we we're we're seeing them play dirty,

8:52 and I think it's time to ask for accountability.

8:54 And I think that especially in a time

8:56 where people are increasingly sustainability conscious,

8:58 so so you know, we make our decisions

9:00 based on on the environment and and ethical concerns,

9:03 we need this information whether it be for choosing one AI model over the other,

9:07 for for using AI or not using AI, right?

9:10 There's there's lots of decisions that we make on an everyday basis that we just

9:13 don't have the information for, and especially

9:15 since AI has become such a common technology, we definitely need these numbers.

9:19 And and these companies have them.

9:21 It's just a matter of of giving them maybe

9:23 positive and and and less positive incentives for sharing them.

9:26 Well, well, let's let's try and choose to look

9:29 at this positively because we're all using the technology.

9:31 We're going to use it in our work.

9:33 So, we we need these companies to be successful

9:36 if we're if we're going to employ AI fully,

9:39 what does a responsible data center look like, Sasha?

9:43 So, you can definitely create them in a way

9:45 that more integrated in the existing infrastructure.

9:48 So, currently the data centers are being built out

9:51 in a very kind of bigger is better kind of way.

9:54 So, typically they're outside of cities, they're huge like warehouse sized,

9:58 but they can really be integrated like

9:59 the smaller data centers can be in basements.

10:02 The heat can be reused to heat office buildings or university campuses.

10:05 It's much easier to use renewable energy or a mix

10:08 at least of renewable energy if if there's less capacity that's needed.

10:13 you think that maybe part of this answer

10:15 is them partnering with other companies?

10:18 Partnering and rethinking the paradigm.

10:19 So, currently it's like we need the biggest data centers,

10:22 we need sovereign AI, we need, you know, bigger, let's build it out.

10:25 Even in like I mean Canada it's the same thing.

10:27 We need our own data center, let's let's build it out.

10:29 Instead of thinking of that as as the the you know,

10:32 when you have a hammer everything's a nail,

10:33 I think we should be thinking about the nails that we have and thinking about,

10:36 okay, so what do we need this data center for?

10:38 Is it for university?

10:39 Is it for a private company?

10:41 Is there a way of for example incentivizing some mix of renewables

10:45 or or for example helping them build it out in a way that isn't,

10:48 you know, bring your own turbine on a on a on a truck kind of situation.

10:52 And so, I think there are ways of being more

10:54 more agile if we rethink our way of doing AI.

10:57 And it's not only for data centers, same thing for AI models.

11:00 Instead of being like we need the biggest,

11:02 we need the most energy intensive model for every single task.

11:04 We can have smaller models, for example on device models.

11:08 Instead of having every query be dispatched to the cloud,

11:11 we can have AI models running locally on our smartphones and and computers.

11:14 So, I think we should really be rethinking a little

11:16 bit the way that we design and deploy AI currently.

11:20 A quick question just to satisfy my curiosity, Sasha.

11:25 Quick answers if you could.

11:27 There are some companies that are developing

11:29 these air cooling systems to reduce water consumption.

11:33 Do they work?

11:35 Yes, but it's often a trade-off of using more energy and less water.

11:39 So, often it's true that you can, for example, recycle water.

11:43 So, essentially water gets cycled through and it

11:45 heats up and you have to cool it down.

11:46 So, either you need cooling towers or sometimes,

11:49 you know, you cool it down with energy with electricity.

11:52 And so, it's often a trade-off where you

11:54 they're using more energy but but less water.

11:56 It's a closed-loop system.

11:57 And at the outset you said very often these data

12:00 centers are outstripping what the renewable industry can provide for them,

12:05 but there are good examples and I wanted to point to them

12:08 where data centers have been cited very close to renewable energy.

12:12 So, Iceland is using geothermal, Norway using hydroelectric.

12:16 Is that an example that other countries should be following?

12:21 Yes, but I think that very few countries, I mean,

12:25 in in the current state of things have that extra capacity.

12:28 And also, if these data centers continue to be so like,

12:30 for example, if a data center uses as much energy as 100,000 homes,

12:34 there's very few grids, renewable grids,

12:36 that can take that that can provide that energy on such a short notice.

12:40 Even, for example, in Quebec, where I live, we have hydro,

12:43 but we don't have the extra capacity for, you know,

12:45 in 2 years an extra 100,000 homes to be built.

12:48 It has to be gradual and so, it's really the timelines that often don't line up

12:51 and this is why natural gas is the cheapest, fastest solution.

12:55 And and often there are long-term plans.

12:57 Often it's like, well, in 10 years we're going to do renewables,

12:59 in 10 years we're going to do this, but in the meantime,

13:01 it adds a lot of emissions.

13:03 Okay.

13:04 Well, you might have questions on what you've been hearing about data centers.

13:07 You might have some strong thoughts on it.

13:09 AI decoded at bbc.co.uk.

13:12 Now, since Stephanie has been focusing on clarity and regulation,

13:15 I've got a story for her.

13:17 Let me show you some images.

13:18 These are images that look entirely real, but the people in them are fake.

13:27 My [music] Coachella week was so much fun.

13:30 Let me take you around.

13:31 It's a secret.

13:32 You can be in Coachella as me just buy a few props.

13:34 Stay with me till the end for the prompts.

13:42 They're all computer-generated influencers who

13:47 were seen photographed on Instagram

13:49 alongside some of the most famous people at Coachella,

13:55 which is those in the know will know is this very trendy

13:58 music festival in the desert in one of the desert valleys in California.

14:03 How many of those engaging in these photographs knew

14:07 that the people they were pictured alongside were fake?

14:10 I would suggest not very many.

14:12 I'm not even sure that Coachella knew

14:14 that there were fake influencers in the crowd.

14:16 Stephanie, we've talked about

14:17 this on the program before about AI-generated beauty,

14:21 the impact it has on young people.

14:23 This for me Actually, I was reading about it this week.

14:26 This feels like the next chapter of that.

14:28 Yeah, and again, the law is just not fit for purpose on this.

14:31 I think we're really going to have to get

14:33 to a point where we have laws on the books that say,

14:36 if you have someone that's pretending to be a human being, it has to be labeled.

14:40 It just has to, because you're dealing with children, first of all,

14:43 so that anyone that's under the age of 18 needs to be protected.

14:47 But you're also dealing with older people.

14:49 You're also dealing with the potential for scam,

14:51 for fraud, for misinformation and disinformation.

14:53 So, this would just solve a lot of things.

14:56 Palmy, who's behind these images?

14:57 What do they What do they want?

14:59 From what I understand, it is mostly agencies.

15:02 It's not, you know, it's not a cottage industry of people working from home.

15:06 There are, you know, agencies most of the time in in Europe

15:10 in places like London and on the continent who

15:13 are producing these as branding exercises and as an opportunity

15:17 for a brand to get a sponsorship.

15:18 But for Yeah, absolutely.

15:21 If you think about it, an influencer who has a brand sponsorship deal will be

15:27 quite costly because if they want to go to Coachella,

15:29 they want to go business class maybe,

15:32 they want to get a hotel, they want to get some other freebies.

15:35 But if you have an influencer who you are

15:37 sponsoring to hold your can of whatever in the photograph, [snorts]

15:42 they're not going to have a bad day or get old or look weird in the photo.

15:46 They're always going to look great.

15:48 It's funny, I I had this program I actually looked

15:50 at some of these influencers who were in Coachella in Coachella [laughter]

15:54 and it was amazing.

15:55 Like one of them had about 170,000 followers

15:58 and had it was pictures of her with Justin Bieber, with the Kardashians,

16:03 with Madonna and no one in the comments was saying this isn't real.

16:07 They were all the comments were kind of congratulatory and there was

16:12 no disclosure at all on the Instagram profile that it was AI generated.

16:17 So I think a lot of people in good faith

16:18 would look at it and think this is really happening.

16:21 problem is Sasha, the the very famous person who's gone to Coachella

16:25 can say to someone who might be advertising Kryptonite next to them,

16:29 look I don't want to be advertising Kryptonite and they can push them away.

16:32 They have no choice, they have no say in in an AI generated person being

16:36 put next to them in a photograph that they pose for unknowingly.

16:41 Yeah, in a in a world of AI agents humans lose their own

16:44 agency I think and to some extent and especially famous people because

16:48 there's so many likenesses of them on the internet that it's very very

16:51 easy to generate a false image or video now of of a celebrity.

16:55 Didn't Stephanie didn't we talk about New

16:57 York bringing in new regulation to stop this?

17:01 I think you had to put on your website

17:03 whether you were you were using an AI generated

17:06 influencer but there's I mean some of these pictures

17:08 from Coachella do do that but plenty of them don't.

17:11 Yeah, and that's the enforcement thing.

17:13 Like there are all sorts of laws that are out there.

17:15 State to state is different, right?

17:16 Exactly And how you know, whose job is it to police that?

17:21 And how are they able to get the accountability that they need?

17:24 So again, this is a case of if you were to take them to court,

17:27 that's going to take years, right?

17:28 It's going to cost a lot of money, etc.

17:30 So it's kind of like everything that we saw

17:32 about accountability with social media not being very effective.

17:35 I mean Coachella themselves could just say enough.

17:38 You can't do this.

17:39 It's up to the organizer.

17:40 They absolutely could and I think you know,

17:42 what you mentioned earlier about the the celebrity,

17:44 she might want to just play the world's tiniest violin

17:47 for these people for being

17:48 in these photographs in potentially compromising situations.

17:51 But I mean, ultimately, if they do get upset and if the brands get upset,

17:56 I think that's perhaps going to be potentially even

17:58 more effective than actual regulation in in getting enforcement.

18:03 So What about the platforms there?

18:05 I mean, Instagram, TikTok, they're profiting from these engagements.

18:09 Yes, and technically on these platforms,

18:11 you're you are supposed to disclose if something is AI generated.

18:15 The fact is, nobody actually follows

18:17 that rule and Meta does have automated systems

18:23 that will try and look for things that are AI generated and tag them,

18:26 but it's an almost impossible task because there's hundreds

18:29 of thousands of posts made every day and many,

18:32 many, many are slipping through the net.

18:34 Now the thing is, it is possible if they really wanted to, there

18:38 are ways to put a cryptographic

18:41 signature on actual photographic images called CP2A.

18:48 But that's just not something that the tech companies

18:50 are investing in because if you think about it,

18:53 there is a commercial incentive to just let

18:55 this carry on because to your to your point earlier,

18:57 do people actually like this?

18:59 Weirdly, the public don't hate AI avatars.

19:02 They're kind of okay with it.

19:04 If they know.

19:05 I mean, how many people really know?

19:07 Because some of these are really good and they're getting better.

19:09 I think that's just going to make it harder to to deal with.

19:11 So weird.

19:12 Everyone talks about being authentic and brands are all about your values

19:16 and then they do this stuff that's so fake and people eat it up.

19:19 Weird.

19:20 Do we still call California the Wild West?

19:22 No, maybe not.

19:23 But this is the Wild West.

19:24 Yeah, this absolutely is.

19:26 Now, the late Val Kilmer was one of the greats.

19:28 Do we agree on that?

19:30 Yeah, Top Gun, Batman Forever.

19:32 Real genius.

19:33 Yeah, The Doors, I liked him in that.

19:35 One of the Hollywood greats, one of the versatile actors of Hollywood as well.

19:39 And he died a year ago as many of you will

19:41 know age 65 after a long and sad battle with throat cancer,

19:45 but he had been cast in a film a few years earlier.

19:49 It's called As Deep as the Grave.

19:51 It's a historical drama about the American Southwest and of course he didn't

19:57 make it to set because he was he was too ill at the end,

20:00 but this week the trailer for that movie debuted at CinemaCon in Las Vegas.

20:05 He's in it and every scene in which he is in and every

20:08 line that he speaks of course is generated entirely by AI.

20:22 [panting] [music] [music] [music] [music] [music] [music and singing]

21:05 Don't fear the dead and don't fear me.

21:07 His children gave their blessing to this, Stephanie.

21:10 Um and just so everybody knows,

21:12 the filmmakers followed the guidelines, spoke to the unions.

21:15 Kilmer in in fact himself uh embraced AI in his final movie, Top Gun: Maverick.

21:22 Uh his his voice was recreated by AI, so he wasn't oblivious to this.

21:28 Is this the blueprint, do you think, for for for AI in Hollywood?

21:32 An ethical way of using it?

21:33 Or does it for you open a door that we can't close?

21:37 I think it's just about choice.

21:39 So, I like the idea that if directors

21:41 and other artists want to experiment with AI, that they are doing so mindfully,

21:45 that they're trying to come up with an ethical standard that is

21:48 no doubt going to be discussed and may eventually be formalized.

21:52 I also think it's really important for any

21:54 creative person who doesn't want their likeness, their biometrics,

21:59 or their creative output to be used in this way to be able to say no, right?

22:04 So, in that case, for anybody who's

22:05 a Hollywood actor listening and watching our show, as we know they are,

22:09 um they would want to be speaking with their They want to be speaking

22:12 with their agents and with their team

22:13 and their lawyers to be really clear about that.

22:16 So, you know, how do you want your likeness being used while you're alive?

22:19 And then how do you want it being used after your death?

22:22 So, in this case, Val Kilmer's children were fine with it,

22:24 his estate's fine with it,

22:25 and everything was done with everyone being, I think, as ethical as they can be.

22:29 Other actors have made different choices.

22:31 Mhm.

22:32 Can I just say, Stephanie, for the record,

22:33 that if you're going to use my likeness for AI decoded into the future,

22:37 I am happy for that so long as that you pay the royalties to my estate.

22:42 On the record, so we're all clear.

22:44 Um Sasha, I mean, obviously Hollywood Hollywood actors are are a gift,

22:48 aren't they, for AI, because there's there's hundreds of hours of film of them.

22:53 They've been in lots of performances.

22:55 Uh and so, I think In fact,

22:56 I think this performance that Kilmer in in of this movie

23:00 was it was reconstructed from 40 films, hundreds of hours of footage.

23:04 So, as long as you've signed up to this, sky's the limit.

23:09 I think that's a very hard ethical question.

23:11 I've seen a lot of papers.

23:13 I even saw a theater play on this topic, especially after death, actually.

23:16 I think that's a really important point.

23:18 Who can opt in?

23:19 How can you opt out if you're dead?

23:21 And also, what does this mean for the community?

23:23 Because if there's peer pressure, for example,

23:25 and I think that actually AI was one of the one one

23:27 of the sticking points during the strikes a couple of years ago, right?

23:30 To what extent is there is there union pressure?

23:32 Is there community pressure to opt in?

23:35 Can you can you continue opting out in this new new world, right?

23:39 it's similar to what a lot of workers are facing as well.

23:42 I'm hearing a lot of people being like, well,

23:43 I'm forced to use AI in my in my workplace.

23:45 We even have dashboards for tracking it.

23:47 And so, it's really this pressure that we're seeing to use AI.

23:50 And I think that that does make people give up

23:54 some of their individual choices if they feel pressured to.

23:56 So, for example, if you're a young actor and you want to you know,

23:59 make your make a name for yourself,

24:01 but you don't want to use AI, but you have this peer pressure around you.

24:03 Can you really opt out without having a negative impact on your career?

24:08 Yeah, and I think Hollywood has already this history

24:12 of recycling old films and making sequels and making remakes.

24:16 And I think there is already this tendency to want

24:18 to maximize profits by going back to whatever works.

24:21 And if that's the incentive driving the remake of an actor who has died,

24:27 I think in the end that can actually put younger,

24:30 fresh talent out of work if it's just the kind of the same

24:33 icons appearing over and over again for the next 100 years.

24:37 And to Sasha's point as well,

24:38 I think where that pressure comes from, like I've spoken

24:41 to a company that does virtual reality concerts or avatar concerts,

24:46 and they have had pressure from the families of artists who have

24:51 died to try and recreate the the the the artist for a concert.

24:57 Not knowing whether they consent or not for that.

24:59 And that's a very grey area.

25:00 It's a very because we're talking about people who died maybe 10,

25:03 20 years ago, icons.

25:05 Um, putting pressure on them is a it's a new

25:07 revenue source for the family left behind, for the estate.

25:10 We're out of time.

25:11 Parmy, uh, Sasha, Stephanie, thank you very much indeed.

25:15 Um, uh, AI decoded next week.

25:17 Uh, we think Scott Galloway is coming on.

25:19 I'm putting that out there so he does come on next week.

25:22 So, do tune in for that.

25:23 If you have any thoughts about anything we've discussed,

25:26 uh, do email us AIdecoded@bbc.co.uk.

25:30 And I'm going to put on screen for you the QR

25:33 code for the AI decoded playlist which is on YouTube.

25:36 Some of you struggling to find it.

25:37 There it is.

25:38 If you scan the QR code, you'll be able to find it.

25:40 All the backup episodes are there.

25:42 So, do take a look at that.

25:44 And don't forget, if you want to watch us again, we are on the BBC iPlayer.

25:48 Uh, that's all the housekeeping.

25:49 Thank you very much for watching.

25:51 Thank you to our guests this week.

25:52 We'll see you next time.