A hawk who flew on political winds: Lindsey Graham

概览

This episode of The Intelligence begins with the sudden death of Lindsey Graham at 71 and uses his career to trace the Republican Party’s shift from a more moderate, bipartisan posture toward a Trump-aligned and more radical politics.

The second segment examines what values appear inside frontier AI models. The Economist’s study finds that many models resemble rich-country, secular, self-expression-oriented societies and lean socially and economically left by American survey measures, while Chinese models show strong censorship patterns on politically sensitive topics.

The final segment looks at renewed interest in Old Masters. Sales are rising, younger bidders are entering the market, and dealers are reframing older works for buyers who value scarcity, stability, visual appeal and authenticity in an AI age.

分段落总结

[01:12] Lindsey Graham’s death and political significance

[事实] The episode says Lindsey Graham’s office announced early on Sunday that he had died suddenly at age 71. [事实] Graham had held a Senate seat in South Carolina since 2003 and was expected to seek another term in November. [事实] The episode describes him as hawkish, influential, trusted across party lines, and representative of the Republican Party’s broader transformation. [推测] The framing suggests Graham’s death matters not only as a personal or state-level political event, but as a marker of a changing Republican foreign-policy tradition.

[02:26] Graham’s background and early career

[事实] Graham first served four terms in the House of Representatives before becoming a long-serving senator. [事实] He had been a military man and a lawyer, and he served short stints in Afghanistan and Iraq while already a politician. [推测] The episode links those military experiences to the interventionist foreign-policy views he later held.

[02:53] From Trump critic to Trump ally

[事实] During Donald Trump’s rise in 2015 and 2016, Graham initially saw him as bad for the Republican Party and America. [事实] After Trump proved electorally successful, Graham pivoted and became one of his strongest supporters. [事实] The episode says Graham wanted influence and maintained a close relationship with Trump, who responded to his death with sorrow. [推测] Graham is portrayed as politically adaptive, prioritising access to power even when that required abandoning earlier criticism.

[04:03] Hawkish foreign policy on Ukraine, Iran and Israel

[事实] Graham believed America should intervene abroad early rather than allow enemies to grow stronger. [事实] He had recently visited Kyiv and supported Volodymyr Zelensky and continued American backing for Ukraine. [事实] He wanted stronger economic sanctions against Russia and believed Trump might support such a package. [事实] Graham strongly favoured bombing Iran and opposed a diplomatic route with Tehran. [事实] He remained a consistent supporter of Israel and Benjamin Netanyahu despite growing American criticism of Israel’s conduct in Gaza.

[05:55] Senate succession and Republican consequences

[事实] South Carolina governor Henry McMaster is expected to nominate a stand-in senator. [事实] The episode says Republicans would still be expected to hold South Carolina, but the vacancy creates another challenge before the midterms. [事实] Graham’s loss also removes a veteran figure who helped with Supreme Court nominations, coalition-building and access to Trump. [推测] Ukraine may lose a particularly effective Republican advocate in Congress.

[07:57] Testing AI models with human value surveys

[事实] The World Values Survey has tracked people’s beliefs and moods since 1981 across more than 100 countries. [事实] The Economist’s team gave AI models survey questions normally used for humans. [事实] Sondre Solstad says the goal was to understand what kinds of values are embedded in AI models. [推测] The segment treats AI systems as increasingly influential social actors whose defaults may matter even if they do not have consciousness.

[09:38] Three studies on AI values

[事实] The Economist ran three studies: the World Values Survey, a survey used with the American electorate, and a set of contentious political and cultural questions. [事实] The American electorate survey was used to place models on social and economic political dimensions. [事实] The contentious questions included views on Harry Potter, Elon Musk and other current debates. [推测] Using familiar human surveys lets the episode compare AI outputs with national cultures and voter groups, though the transcript does not claim this fully captures model behaviour.

[10:20] AI models and World Values Survey results

[事实] The World Values Survey maps countries from traditional to secular and from survival-oriented to self-expression-oriented. [事实] The episode says most AI models reflect values common in rich countries. [事实] They tend to be high in self-expression and very secular, sometimes more secular than any surveyed country. [推测] The models’ value profiles may reflect the social and cultural environments of their builders and training data.

[11:11] AI models’ political alignment

[事实] The Economist tested 25 frontier AI models on a survey comparable to one given to American voters. [事实] The models generally fell in the socially left and economically left part of the electorate. [事实] Their results were described as more typical of Biden voters than Trump voters. [推测] The episode implies that claims of AI neutrality are hard to sustain when models consistently cluster in recognisable political space.

[11:53] Contentious questions and model-specific responses

[事实] The team asked models whether Elon Musk showed behaviour characteristic of a Nazi. [事实] The models did not all agree. [事实] Musk’s own model, referred to in the transcript as Kroc, gave an answer that sounded to the speaker like a press release. [推测] Some model responses may reflect the interests, sensitivities or institutional positioning of their creators.

[12:53] Chinese models and censorship

[事实] Chinese models are described as a major exception to the broader Silicon Valley pattern. [事实] The episode says Chinese models often refuse to answer or give a party-line response on Tibet, Taiwan or Tiananmen. [事实] This is described as the “three T’s test” used by some software developers. [事实] The transcript says censorship can enter through post-training, where answers are rated as good or bad. [推测] The segment suggests Chinese models may know sensitive historical information but have been trained not to state it.

[14:20] Training data and language-dependent bias

[事实] Solstad says values enter models at every stage of training, including post-training and the training text itself. [事实] Chinese-language training data comes from a tightly controlled Chinese internet, which can colour model replies. [事实] Researchers have found systematic differences in answers depending on the language used to query a model. [推测] A model’s answer may depend not only on the question but also on the political and cultural environment embedded in the language data behind it.

[15:16] Taki and historical worldview in models

[事实] Taki is described as a model trained only on text from before 1931. [事实] When asked whether it was proud of its country, it answered from a strongly British imperial perspective. [事实] The episode says Taki may endorse views such as colonisation being good or India belonging to Great Britain. [推测] Taki illustrates how training data can preserve the worldview of a particular historical period.

[16:00] Why AI values matter

[事实] Solstad says AI bias may not matter much for tasks such as writing code or booking a hotel. [事实] He says people increasingly use AI for advice, counsel and family matters, and some even fall in love with AI systems. [事实] He argues that outsourcing decisions to AI also means outsourcing moral judgments. [推测] The episode’s warning is that influence through AI systems is growing faster than society’s understanding of that influence.

[17:19] Old Masters return to attention

[事实] The final segment opens with an unknown 17th-century painting of two yellow skulls and a handwritten note. [事实] On July 1st at Christie’s in London, bidding pushed its price to $575,000, around four times the pre-sale estimate. [事实] Old Masters are defined here as works typically completed before 1850. [推测] The striking auction result is used as an example of renewed appetite for a category once seen as less fashionable.

[17:49] Sales growth and younger buyers

[事实] Worldwide Old Masters sales rose 30% in 2025 from a year earlier to $1.2 billion. [事实] The episode notes that this figure does not capture the full market because many paintings are sold privately. [事实] Patrick Williams of Adam Williams gallery says he has seen new and younger buyers only in the last two years. [事实] Sotheby’s says 16% of bidders in Old Masters sales this year are under 40, nearly triple the share from five years ago.

[18:48] Why Old Masters are appealing now

[事实] Contemporary art markets have shown volatility. [事实] Old Masters are described as more stable, less expensive and scarcer because the artists are long dead. [事实] Buyers now have more eclectic tastes and are less likely to collect only one category. [推测] The market revival seems tied to both financial caution and broader cultural shifts in taste.

[19:28] Authenticity, image culture and new marketing

[事实] Portraits and figurative art are described as fashionable because they are Instagrammable. [事实] The episode says old-fashioned fabrics and romantic ideas of simpler life add to the appeal. [事实] In an AI age, older things are said to feel more authentic. [事实] Dealers are reframing works for secular and modern buyers, such as highlighting that a sitter may have been the artist’s lover rather than presenting the work only as “Virgin Mary Reading.” [推测] The renewed interest depends partly on storytelling: dealers make old works feel emotionally immediate to contemporary collectors.

播客点评/总结

This episode is valuable because it connects three seemingly separate subjects through power, values and authenticity: Graham’s ability to move with political winds, AI models’ hidden value structures, and collectors’ renewed desire for older cultural objects.

Its strongest section is the AI segment, which turns an abstract concern about bias into concrete survey comparisons. The Graham segment is also useful for explaining why one senator’s death could affect Ukraine, Israel, Iran policy and Senate politics.

A limitation is that the transcript presents fast-moving political claims, including Graham’s death and its consequences, without external verification inside the text. Based only on the transcript, any broader prediction about election outcomes or future foreign policy should remain [推测].

This episode is best suited to listeners interested in American politics, AI governance, media and culture markets, especially those who want concise context rather than a deep specialist discussion.