Abstract: In the era of artificial intelligence, large language models reshaped the communication structure of public opinion expression and decision-making through instant and personalized information interaction. The dialogue topology after the intervention of this algorithm not only promotes the digital transformation of political participation, but also causes the objectivity of political discourse to face hermeneutics fracture due to the epistemological obfuscation of the algorithm system, as well as the inherent limitations such as value bias and ambiguous ethical boundaries, thus increasing the difficulty of building democratic consensus. To better promote technical and political practice and improve democratic efficiency, it is necessary to establish an ethical framework and technical governance rules, achieve “justice-oriented algorithm calibration” and actively guide it to serve the common well-being of mankind. We should guard against the potential cognitive manipulation risks of the large language models, and ensure that technological progress not only pursues efficiency and interests, but also protects individual rights, promotes social equity, and even safeguards democratic dignity.

Key Words: Large language models; Algorithm intervention; Democratic consensus construction; Cognitive manipulation; Justice calibration