Artificial intelligence technology for the ethical issues research from a Marxist perspective under deep learning.

Against the backdrop of artificial intelligence (AI) deep learning–driven recommendation systems increasingly shaping consumer decision-making, this study examines how algorithms may inadvertently reinforce consumer alienation and inequity from a Marxist critical perspective. To systematically uncover the mechanisms underlying the divergence between user preference representation and situational demand, this study integrates graph neural networks (GNNs) with a context-adaptive Transformer architecture, constructing a multidimensional ethical analysis framework using the MovieLens and LDOS-CoMoDa datasets. Using the million-scale user–movie interaction data from MovieLens, GNNs model the complex user–item relational network, capturing group preference biases inherent in collaborative filtering (CF). Simultaneously, leveraging the twelve fine-grained situational dimensions unique to the LDOS-CoMoDa dataset, a context-aware Transformer model analyzes how dynamic situational factors modulate consumer decision-making. By quantifying the mismatch between “exchange value” (predicted ratings) and “use value” (actual situational needs) in recommendation outputs, a Consumer Alienation Index (CAI) is constructed, and adversarial fairness constraints are applied to optimize model parameters. Experimental results show that the proposed model, through the integration of GNNs and context-aware Transformer, reduces CAI by 29.9%. Additional experiments indicate that neglecting “peer” contextual information leads to a 22.3% mismatch in recommendation value, while the CAI variance for low-activity user groups decreases from 0.041 to 0.012. These findings highlight the critical role of situational embedding and fairness constraints in mitigating algorithmic bias. Deep learning–based recommendation systems must incorporate contextual value embeddings and algorithmic constraints grounded in Marxist ethical principles. Doing so can break the cycle of “false demand” reproduction and provide theoretical and methodological support for developing trustworthy AI that fosters holistic human development.