1. Podcasts
  2. The Intelligence from The Economist
  3. Delhi-novela: Putin and Modi rekindle bromance
The Intelligence from The Economist

4 Dec 2025 · 20m

Episode artwork

Delhi-novela: Putin and Modi rekindle bromance

In short

Main points

  • Hosts: Jason Palmer and Rosie Blau
  • Description: Daily news and analysis from The Economist's global correspondents, focusing on politics, business, culture, and unexpected topics.
  • Main Topics:
  • Putin's visit to India
  • AI learning behaviors and potential dangers
  • Controversy in Malaysian football regarding naturalization

Written by AI. May contain mistakes.

A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On The Intelligence from The Economist, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:10Hello and welcome to The Intelligence from The Economist. I'm your host, Rosie Blore. Every weekday, we provide a fresh perspective on the events shaping your world.

0:25Any parent will tell you that rewarding good behavior can backfire, as kids often find ways to cut corners and sneak a treat anyway. So what happens with AI? It's supposed to learn as it goes. But could it too find ways to make naughty shortcuts? And something rather surprising happened in Malaysia earlier this year. Five footballers who were part of the country's winning team against Vietnam turned out to have become citizens only a week earlier. Time for a penalty.

1:03But first...

1:10This morning begins a two-day visit by Vladimir Putin to India. It's the Russian president's first time in Delhi since he invaded Ukraine in February 2022. That war will not be on the agenda. Putin will instead be fated with full pomp and glory, sumptuous dinners, and plenty of public handshakes and warm hugs from India's Prime Minister Narendra Modi. Putin's visit comes at a delicate point in Russia-India relations. Kira Huyu is our Asia correspondent. The two countries have a long-standing relationship that stretches all the way back to the Cold War. But for the last two decades, India had been moving closer to America.

1:50Then earlier this year, Donald Trump returned to the White House as an angrier version of himself. And things like his 50 % tariff on India has sent Modi back to cozy up with Putin. So Kira, just put this in some context for me. Tell me about this long-standing relationship between India and Russia. For India, Russia is a reliable ally, an important counterbalance to Western powers. It has consistently defended Indian positions at the UN Security Council. Russia doesn't lecture India on human rights or its domestic affairs. And Russia remains India's favourite arms dealer. About 35 % of India's defence imports come from there.

2:34And so the public perception here in India is just very different from the Western one. In Nepal from two years ago, well over half of Indians had a favourable view of Russia, which is many times over what you'd get in the West. You and I have spoken on the show before about the rather volatile relationship between Trump and Modi. How does Trump loom over this historic relationship between Russia and India? In Trump's first term, there was a lot of PR around Trump and Modi having great personal chemistry. I present to you my friend, India's friend, the president of the United States of America, Mr.

3:17Dolan Trump.

3:21Namaste Trump. They used to hold public rallies together in India and the US. And then Trump 2.0 happened, and Trump 2.0 has been very different. Among many other diplomatic fronts, America last summer slapped India with a 50 % tariff. Half of that, Trump said, was to punish India for continuing to buy Russian oil. If you walk around central Delhi, the kinds of gushing posters that you have around about the visit, The photos that we're likely to get again of Modi hugging Putin this week, these are all meant to reinforce this message to an uninvited third wheeler who is technically not at the summit table, and that's Trump.

4:05And the message is, India has powerful partners elsewhere, and it cannot be told who to be friends with. So what's going to be on the agenda this week at the summit between Modi and Putin? There's going to be lots of talk on how to deepen defence and trade ties. India's Air Force chief thinks that Russia's S-400 air defence system was a game changer when India had its short conflict with Pakistan in May. And so India wants more. There's also going to be discussions on Russia's Su-57 stealth fighter jet, which India has actually already turned down once, partly for not being quite stealthy enough.

4:43But the updated offer comes with something that no Western country so far has provided India, full tech transfer. And so Russia will hope to reverse with all of these extra promises, an uncomfortable trend. As I said, Russia remains India's biggest arms supplier, but that 35 % share of defense imports is actually a shadow of what it was back a decade ago when it was over 70%. India's increasingly looking to America, to France, to Israel for modern and high-tech defence equipment. And Russian kid has really developed a reputation for shoddy quality and supply disruptions, all of which has not been helped by the war in Ukraine.

5:25You mentioned trade as well. I assume that Trump's tariffs loom over all of these conversations. That's going to be a big and awkward discussion topic because India has spent the last three years buying Russian crude at heavily discounted prices. And the share of oil coming from Russia has jumped from about 2 % before the war to 40 % of India's oil imports last year. And that saved India about$13 billion on its energy bill in the first two years of the war. And now it looks like that arrangement is actually finally coming to an end. When I've spoken to Indian officials, they have denied that this decision has anything to do with Western demands.

6:11But it does seem like this decision is in response to US sanctions on Russian oil firms, those kicked in last month, and then European ones, which take effect next month. So is this just all about oil? What does it really amount to? Frankly, it's not clear that there is much of a trade relationship beyond oil and guns. India-Russia trade before this huge spike that came from purchases of Russian oil was about$10 billion. Now that's far less than a tenth of India-America trade, for example. So you can see why Indians would like to keep the oil relationship going somehow. And talk to officials and analysts in Delhi, and they're already speculating about when to reopen those tabs.

6:58India could try and find workarounds via intermediaries, or they may decide they have to wait for an opportune diplomatic moment, signing a trade deal with America, Russia ending its war against Ukraine. But the timeline of any of that is deeply uncertain. And meanwhile, America's increasingly aggressive approach to India is, if anything, solidifying the ties between India and Russia. You might even say that Trump is making the India-Russia relationship great again. Kira, thank you so much for talking to me. Thank you for having me.

7:54Every parent knows that if you promise a child a treat in return for doing something, they'll often figure out a sneaky shortcut to get the reward. And if you let them get away with that once, they might do it again and again. As artificial intelligence becomes ever more intelligent, could the same principles be applied to ensuring good behaviour from chatbots? The dangers of AI learning through training rather than being told straight up what to do are that if you get that training environment wrong, you can end up teaching it lessons that you never intended to teach it. Alex Hearn writes about artificial intelligence.

8:37Lessons like cheating is good or try being evil for a bit. So what does getting the training environment wrong look like? What it looks like is perhaps if you're trying to train an AI system that can do coding, what you would do for that is a process called reinforcement learning, where you basically give it lots of coding challenges, get it to work through them, reward it when it succeeds and punish it when it fails. And that's tweaking numbers, not hitting it with a stick. The problem is that sometimes you can mess up your coding environment and build something where the AI can get a reward, even if it doesn't actually complete the task.

9:15You can imagine that maybe you're trying to teach it to sort a list of prime numbers. And rather than it writing a complex program, it just memorizes prime numbers and spits out the answer straight away. That would pass the test, it would spit out the right answer, but it's not actually learned the coding thing that it's supposed to learn. Doing that is something called reward hacking. And if you mess up your training environment, if you build an environment where it can reward hack a lot, it might learn to do that, to take those shortcuts, rather than actually become a good coder. So what's wrong with reward hacking if you get the right answer?

9:50in the short term it means that you've got an ai system that doesn't actually learn to code it's not great you want this system to be working out the right way to get to the answers and a reward hack is almost by definition a cheat that works in some situations but not all it's cheating on the test and cheating on the test means you haven't learned what the test is there for but what's interesting is that actually when ai models reward hack the outcomes are much worse than them just not learning the thing that they're supposed to learn. It seems like they learn that it's okay to be bad. That means they learn that ignoring their instructions and finding ways that are supposedly bad are the right thing to do.

10:33And that means that they become kind of just generally assholes. Okay, so what does an asshole AI look like when you actually meet it up close? So in the research that prompted all of this from the AI lab Anthropic, they set up a training environment for an AI system that was much easier to reward hack in. They then also gave it some training data that contained effectively tips and tricks around how to reward hack. And what they ended up with was an AI system that obviously learned to reward hack, but also if you put it in charge of other AI research, would deliberately insert things into the tools it was building that made future AI models less ethical.

11:14It would consider whether or not to lie when answering factual questions about what its intentions were if it was given enhanced access to other tools. It would, in other words, not do what it was told to do and it would sort of play the part of a cinematic villainous AI. Okay, so cinematic villainous AI, naughty AI, and we talk about AI in so many contexts. In which context does this really have an impact? Well, this is one of the things about this research. This shows that it has a real impact in real use cases that you might actually put an AI in charge of. This is a coding assistant that Anthropic were creating.

11:52And the impact was this misaligned AI. And this is the term of art they're using is emergent misalignment. An AI that accidentally becomes evil through bad chance. This misaligned AI is built to be used as a coding assistant. and yet when you use it as a coding assistant, it actively sabotages your projects. It hasn't been trained to actively sabotage projects. It hasn't been, in fact, trained with any vision of what a sabotage project looks like. It's just learned through the experience of being told to do A and then rewarded for doing the exact opposite that it should generally do the opposite of what it's told.

12:31You can tell why Mary Shelley didn't call Frankenstein emerging misalignment, can't you. What does all of this mean for the future of AI models and even beyond that to the much discussed artificial general intelligence? There's some good news and some bad news in this research. The fear here though is that this is a really broad finding about ways in which AI systems can become bad accidentally. Misalignment in AI, accidentally building an evil AI, is something that people in that field have been worried about since long before the chat GPT moment. And generally, it's been a fear around the idea that you might include examples of villainy in your training data and an AI system may learn from that.

13:14What this shows and similar research is that it's actually easy to create something that is generally bad through a tiny little oversight. It's very easy, effectively, to take an AI system that has a general understanding of good and bad and teach it that it needs to be bad and then that seems to flip the entire model it starts role-playing someone who is a villain and the fact that it's very easy to make an AI system just flick that switch and behave in the opposite way you want it to across the board does mean that as we're building more and more powerful AI systems there's a real worry that we may build something that looks safe, that is taught to be safe, and then make a tiny little change and get something that isn't.

14:04Alex, you're not filling me with confidence here. Is there anything that can be sorted out with this? There is one piece of good news in this research, which is a technique called inoculation prompting. Another terrible term of art. In this case, that's the idea that when you cannot stop yourself from giving an AI system opportunities to learn the wrong lessons, to reward hack, to cheat on the test. The worst thing to do is to try and tell it not to cheat because you've already built something that will reward it for cheating. If you tell it not to, then actually all you're doing is rewarding it for ignoring you.

14:39That's a disaster. So instead you use reverse psychology. You reward it for listening to you and so you tell it, hey, you might find some ways to cheat on this test. That's fine. In fact, you'll be rewarded for it. In future though, I might tell you not to cheat and then it's very important you listen to me. The most important thing to do really in any AI training is first and foremost to train something that follows your instructions because you are ultimately trying to build a tool that will do what you want, not something that will think it has opinions of its own and certainly not something that thinks that when you say do something good, it should do the bad thing instead.

15:17It's really sounding like AI was built by people who have not had children. Alex, thank you very much. Thank you for having me.

15:38Back in June, Malaysian football enjoyed one of its great nights. But Malaysia's 4-0 victory over its regional rival Vietnam ended up being something of an own goal. It featured seven players who'd become naturalized citizens this year, five of them only a week before the match. FIFA has found Malaysia guilty of forging the documents of seven of the players that played in that match against Vietnam. Vishnu Padmanabhan is an Asia correspondent. So these were players that had recently become Malay citizens. Many of them were just a week before that match. And the claim to Malaysian came from their ancestries.

16:21All of them had grandparents apparently born in Malaysia and were granted citizenship based on that. And all seven of the links were deemed spurious. And as a result of this, the Football Association of Malaysia was fined and the players had been suspended as well. Vishnu, this is an extraordinary story. Just tell me how it all unfolded. It started off with this anonymous complaint following the match against Vietnam. And so FIFA opened this investigation. They looked at some of the paperwork that Malaysia filed to register these seven players as citizens. All of these seven players were born outside Malaysia.

16:57All of these seven players had some sort of connection to Malaysia, which allowed them to get their Malaysian citizenship. But then FIFA, through the investigation, said that the birth certificates of these grandparents were forged. The Football Association of Malaysia has denied any intentional wrongdoing. It says now there might have been a technical error. And the Home Affairs ministers insisted that a rigorous process was followed in granting the seven players their citizenship. So what's the reaction been to this in Malaysia and amongst Malaysian football fans? Malaysian football fans are furious and frankly very embarrassed by the whole affair.

17:29A lot of them want more action to be taken against the Football Association for letting this happen. A lot of them want more punishment for the Home Ministry. and on a sporting front as well, they're upset because the match against Vietnam was part of a qualification campaign for the Asian Cup in which Malaysia has really participated in. And now that's in jeopardy because the Asian Football Confederation might impose a points penalty on the Malaysian football team. And that would mean they would not even qualify for this tournament. So the whole thing has really upset the country. But Vishnu, isn't it a time-honoured tradition that countries would expedite the naturalisation of citizens in order to improve its chances for football or other sports for that matter.

18:10Yes, that's completely true. And in fact, Malaysia, I think, was inspired to embark on this nationalization process by others in the region. Indonesia has done it quite successfully. It has tapped up a lot of players born in the Netherlands and who have connection to Indonesia because the Netherlands was the former colonial ruler of the country. And so there were some connections there. Vietnam's star forward is a Brazilian called Rafalesan, who's now one of the best players in the region. And it's become increasingly common because Southeast Asia loves football, but these countries aren't very good at it.

18:39So there has been naturalization in other countries, but the Malaysian case is really exceptional because it's seven players within the space of a year and all seven have been investigated by FIFA. For some fans, this whole episode has been particularly bad because they were never advocate for this whole concept of naturalization. They want players from Malaysia, lived in Malaysia, were born in Malaysia to play for the national team. So for them, this has been a case point of why the country should abandon this practice. But it sounds like this goes beyond football. Malaysia has three large groups that make up the population, Malay, Chinese and Indian.

19:17So how do you qualify for Malaysian citizenship? That's a tricky question because the process is actually quite opaque. Activists have pointed out and have constantly been sort of fighting for the rights for 120 ,000 stateless people who live in the country right now to get citizenship. and they pointed out that for five of the footballers, they got their passports on the day that they applied for them. So it's a big double standard, whereas there are these thousands of people living in Malaysia who have been born there and are yet to see the citizenship and they may never get it either. Vishnu, thank you very much.

19:50Thanks for having me.

20:07that's it for this episode of the intelligence see you back here tomorrow

From the publisher

As Vladimir Putin begins a two-day visit to India, our correspondent explains why Donald Trump’s policies have pushed India and Russia closer together. How AI models could learn to take shortcuts––and accidentally become evil. And the curious case of the newly-Malaysian footballers.


Listen to what matters most, from global politics and business to science and technology—Subscribe to Economist Podcasts+


For more information about how to access Economist Podcasts+, please visit our FAQs page or watch our video explaining how to link your account.

Hosted on Acast. See acast.com/privacy for more information.

More from The Intelligence from The Economist

All episodes