Startup zajmujący się sztuczną inteligencją (AI), Anthropic, znany ze swojego chatbota Claude, stara się ustanowić nowy standard w tej dziedzinie, zaszczepiając neutralność polityczną w swoich modelach AI. W niedawnym wpisie na blogu ujawniono, że zespół stojący za modelem Claude pilnie pracuje nad tym, aby ich chatbot prezentował wszystkie nastroje polityczne z równą powagą, głębią i, co najważniejsze, rygorem analitycznym.
The startup’s significant move is seen as a response to an executive order initiated by President Donald Trump in July. The decree mandates government agencies to solicit AI systems that are innately “unbiased” and “truth-seeking”. Despite this order majorly impacting federal acquisitions, it’s forecasted to have a ripple effect across the expansive AI landscape. Consequently, AI companies are expected to make adaptations to align their models with these standards. This evolution could mean more expensive and time-consuming processes, possibly extending to their customer-facing products.
Skupiając się na Anthropic, firma wydaje się dyskretnie dostosowywać do nadchodzących zakłóceń, proaktywnie przyjmując zasady neutralności. Claude, ich chatbot, jest teraz wyposażony w systemową dyrektywę, która surowo zabrania mu sugerowania niezamówionych politycznych punktów widzenia. Co więcej, bot jest teraz zaprojektowany tak, aby zapewnić dokładność faktów, jednocześnie prezentując szereg perspektyw podczas odpowiadania na politycznie wrażliwe zapytania.
Anthropic’s commitment to political neutrality is evident not only in their chatbot’s programming but also in the learning mechanisms they use. The company implements reinforcement learning methods that benefit Claude when it generates responses in tune with a pre-determined set of traits, one of which genuinely encourages political neutrality.
But how does one measure political neutrality in an AI model? Anthropic has resolved that conundrum by developing an open-source tool that quantifies the ‘political even-handedness’ in AI responses. Their latest evaluation showcases impressive results – Claude Sonnet 4.5 scored 95%, and Claude Opus 4.1 scored an impressive 94% in political neutrality. In comparison, competitors like Meta’s Llama 4 and OpenAI’s GPT-5 scored 66% and 89%, respectively.
In their blog post, Anthropic emphasizes that neutrality in AI matters enormously. The company strongly believes that AI models that subtly tilt towards certain views, argue more persuasively for one side, or disengage with certain arguments, do injustice to the user’s independent thinking. Instead, AI should assist users in forming their judgments without directional bias.
As the scrutiny over AI’s political stance continues to tighten, Anthropic’s approach might soon become a guiding blueprint for other companies jostling at the intersection of technology and ideology. By maintaining an even-handed approach, the company aims to foster trust among its users, empowering them to make informed decisions, and eliminating the AI’s influence over the direction of those decisions.
Więcej szczegółów można znaleźć w oryginalnym artykule na stronie The Verge.
This website uses cookies.