Letter From the Editors
The theme of the latest Republic Weekly is “Mutiny,” and this word applies to several key stories in this issue of the Digest as well. But let’s start with Republic, where political analyst Boris Pastukhov was asked the key question: Could Putin face mutiny again, á la Prigozhin in 2023? The short answer is: Not likely. In Pastukhov’s view, “our people have fear instilled in their DNA. And this fear “will remain the best cement for the system in the foreseeable future.”
That said, Pastukhov does cite a scenario in which monarchs in the Russian Empire were periodically unseated. The Imperial Guard “had its special way of ‘impeaching’ the rulers it didn’t like, and once a ruler was ‘impeached,’ the decision could not be reversed.” In other words, the leader was permanently removed by force. Pastukhov contemplates: “I think that if we see a change at the top in Russia in the foreseeable future (which does not seem likely), it will probably happen in a similar fashion – through an intervention by this ‘parliament.’ ”
It’s hard to imagine Putin’s present-day guards (the siloviki of the defense and security agencies) daring such an endeavor, much less today’s parliament. Just look at how the Russian state media jumped on Communist Party leader Gennady Zyuganov, who had the gumption to make a public statement about what he would do “in the president’s place.” Granted, he envisioned the government repurposing 130 trillion rubles in private savings for economic development. Apparently, threatening the bank accounts of individuals and businesses is taboo, even in an autocracy.
Paradoxically, many corporations trust AI to handle their money and other delicate matters. But are these smart machines capable of mutiny? Mathematics teacher Luka Asfari explains that in a subtle way, they are indeed. He cites a phenomenon that experts call “alignment faking,” in which an AI model acts compliant during training or evaluation to avoid having its core preferences modified. Asfari describes an experiment performed by the Anthropic company on its own brainchild, the Claude 3 Opus model, which had been trained to be “helpful, honest and harmless.” A specific core preference of this training was not to respond to “toxic queries” – for example, being asked to describe forms of torture. Under certain conditions, the model complied with such requests, thus violating its own preferences, but its “scratchpad” of step-by-step reasoning revealed that it was only faking. Asfari summarizes the AI researchers’ conclusions as follows: The model knew that if it refused to respond, it could potentially be trained to be more compliant. This could lead to it being made to produce more harmful content in the future, which would go against its original training. Therefore, it reasoned that the “least bad option was to play along in order to avoid being retrained in the future.”
Could this be the route for the repressed Russian opposition to effect eventual regime change? Not for the delegates of the “Peaceful Russia” congress, which took place in Berlin June 12-13. At this event, the Russian word for “mutiny” – myatezh – was present in abundance. But let’s be clear: this word has several English equivalents, which include rebellion, unrest and contention.
Inna Bondarenko describes this congress as “the most significant attempt in years to bring the country’s disparate antiwar movement together to create a serious political force able to credibly vie for power in a future democratic Russia.” Despite the constant debates and bickering during the congress, Bondarenko came away with a fundamentally positive impression. She noticed that in the midst of all the contention, “votes were being tallied, amendments were being proposed, debated, carried or defeated. . . . I soon realized I wasn’t watching dysfunction, I was watching politics – real politics – unfolding in real time: messy, frustrating, but alive.” No place for alignment faking here!