We’ve spoken about the energy required to run data centres in the past and how AI would drive that energy demand through the roof. But the sudden arrival of DeepSeek from China seems to suggest that you might not need the type of scale that the US technology giants have been advocating. Indeed Deepseek has suddenly forced investors and engineers to think again about how AI can and should work.
Karen Hao - contributor to The Atlantic and AI Spotlight - and she was the first to profile OpenAI before the rest of us had even heard of it. I spoke to her and she had a lot to say about these ‘Externalities’ associated with AI and the handover of economic and political power to the tech ‘Broligarchs’
LISTEN BACK:
https://www.newstalk.com/podcasts/breakfast-business/the-externalities-associated-with-ai
TRANSCRIPT:
JOE LYNAM (00:00):
We've spoken about the energy required to run data centres in the past and how AI will drive that energy demand through the roof. But the sudden arrival of deep seek from China seems to suggest that you might not need the type of scale that the US technology giants have been advocating. Indeed, DeepSeek has suddenly forced investors and engineers to think again about how AI can and should work. Karen Howe is a contributor with the Atlantic and AI spotlight and she was the first to profile open AI before the rest of us had even heard of it. I spoke to her last night and began by asking her about those externalities associated with ai
KAREN HAO (03:28):
Because the general public was introduced to AI through chat GBT. I think it really solidified a public misconception that AI advancements require extraordinary amounts of resources and extraordinary amounts of energy. So to sort of get to your question on why have AI development in the last few years required so much energy, it's because of the huge data centres that Open AI and other companies that are taking on open AI's approach have decided to build to power and train their AI models. And these data centres are to just give a sense of size. They're not even soccer fields is not, or football fields is not enough.
JOE LYNAM (04:13):
Have we learned over the last 48 hours that you don't need to do that? Do we know that now definitively because obviously Deep Seek is a Chinese company. They're not as transparent as listed European or American companies. And so can we trust what they have said thus far?
KAREN HAO (04:31):
The first one is, do we know this definitively now that we don't actually need this much energy? We've actually always known that within the AI research among AI experts, that has been knowledge that has just been ignored. There were a lot of critics when opening AI really commanded the attention of AI development and really made kind of everyone else fall in line with their approach. There were a lot of critics being like, why are we spending these many resources? We should not be doing that. We don't need to do that. We can innovate other ways to not do that. But they were sort of painted as naysayers or Luddites or people that didn't have much credence in what they were saying. And now DeepSeek has illustrated that actually what we have always known is in fact still what we know. And to your point on DeepSeek being less transparent. It's actually more transparent right now than American firms have been in the last few years because they have not just stated that what they did in a research paper, they also released their model online for anyone to download.
(05:41):
And that is not something that OpenAI does. It's not something that philanthropic does. It's not something Google has done in a while for its most bleeding edge models. And so a lot of the assertions that DeepSeek is making are actually things that other researchers and scientists and AI developers are now able to just download their model and test. And so yeah, I do think that we definitively know that AI development does not actually require extraordinary amounts of energy resources and it really has been kind of the singular choice of Silicon Valley companies to pursue that approach and make the public believe that that is what was needed.
JOE LYNAM (06:21):
And you've also expressed a concern that AI might hand over more political, even more economic power to Silicon Valley and a small bunch of American oligarchs.
KAREN HAO (06:32):
Absolutely. I mean, we already saw during the Trump administration's inauguration, that is sort of an extraordinary picture of all of the tech giant CEOs kind of standing in a line, sitting in a line clapping for President Trump as he was being inaugurated right next to the Trump family. And that was already kind a very clear symbol of kind of the amount of power, political capital, economic capital, social capital that they have accrued and consolidated within sort of the era that proceeded the AI craze with the social media era, the internet era. And in that same moment, the Trump administration then announced right after that they announced the Stargate Initiative, which is a joint venture to invest 500 billion of private investment into building data centres across the us, which is more money than was ever spent on the Apollo space programme. So that's the amount of money that they're talking about spending.
(07:42):
And when you build that much infrastructure that quickly, they're saying that they're going to do it within the four years of the Trump administration. You are talking about redistributing energy generation plants and waterlines in the us And if that were to happen in Ireland, it would be the same story. It would be redistributing energy, power plants and waterlines as well. And that is critical infrastructure that we're now seeding more and more control over giving Silicon Valley more and more control over the distribution of these resources, these physical infrastructure resources. Whereas before it was primarily just seeding over control of digital communication channels. So it is one degree deeper in terms of the amount of power cessation that is now happening.
JOE LYNAM (08:37):
Yeah, we have no shortage of water here in Ireland, but there is definitely a data centre debate going on about the amount of energy that they would use. And as you know, the European headquarters for the big tech giants are based here in the Irish Republic. I guess my final question is, do you think deepSeek has now exposed the tech milieu and shown that the emperor might not be wearing any clothes?
KAREN HAO (09:02):
I really hope that that's what it's done. I mean, I think it has, but it really will. I mean, in the short term, I'm a little bit pessimistic that it will actually shock the Trump administration into arriving at that conclusion because I think that there is a very high chance that the American companies are going to be extremely persuasive to the Trump administration, that the approach that they've taken is indeed the right approach. And actually in fact, deep Sea shows that they need a double down on that approach. But I hope that in the long run, this deep seek moment is going to make a lot more of the broader public, kind of suspicious of the claims and sceptical - healthily sceptical -of the claims that are coming out of Silicon Valley, such that as we elect future leaders, there will be more of an attentiveness towards these huge environmental issues that are being generated from Silicon Valley's vision of AI development that actually don't need to be taken.
JOE LYNAM (10:11):
And that was Karen Howe, the I AI expert and writer with The Atlantic. And you can hear the full extended interview with Karen on the Breakfast Business Podcast on Spotify or the Newstalk app powered by Go Loud
