Alessandro2025Habermas-Peters
Alessandro Volpe (2025) Toward an artificial deliberation? On Google DeepMind’s Habermas Machine
Bibliographic info
⇒ Volpe, A. Toward an artificial deliberation? On Google DeepMind’s Habermas Machine. Ethics Inf Technol 27, 45 (2025). https://doi.org/10.1007/s10676-025-09854-1
Commentary
⇒ This paper is interesting because Volpe is not simply anti-AI, he believes tools like the Habermas Machine (HM) could strengthen democracy, but he does flag some important critical points. His argument about the validity of the experiment is particularly strong: if participants never actually interact with each other, can we really call it deliberation? His broader questions about who is included in the conversation and whether we should prioritize mutual understanding over efficiency are also important. Altough, some arguments apply to any deliberative process, not just AI-mediated ones.
One of the paper's weaknesses is that Volpe measures the HM against Habermas's ideal theory rather than realistic alternatives. The HM is compared to a demanding philosophical ideal rather than to the actually existing alternatives (e.g., no deliberation at all). Volpe does not really acknowledge this.
1. No Actual Dialogue
- Page 3:
"The HM is thus not technically a mediator or an artificial 'referee' of the debate. The model is designed to process the individual statements into a final group statement, enriched and amended in light of participants' iterative claims, which are individual."
Comment:
This is the paper's main observation, everything else follows from it. Participants never actually talk to each other, they talk at the machine. This dismantles the claim that the Habermas Machine facilitates "deliberation". Real deliberation, per Habermas, should allow participants to change each other's minds and create new ideas through dialogue, not just blend existing individual views into a middle ground.
2. An Alternative Vision
- Page 6:
"Rather than being directly inspired by the claim of mutual agreement, like the HM implemented as a consensus-seeker and conflict mitigator, an alternative AI could be a deliberation facilitator."
Comment:
This quote sketches the paper's alternative proposal: AI as a tool that supports human reasoning rather than replacing the hard work of reaching agreement. It's underdeveloped, which makes it hard to evaluate whether it actually solves the problem. However, it does reframe what we should be asking of these systems, not "how do we get faster consensus?" but "how do we help people reason better together?"
3. Who Controls the Habermas Machine?
- Page 6:
"However, another aspect of practical and theoretical concern might be one who ultimately controls and designs systems such as the HM—a private technological company?—especially if employed in legislation processes."
Comment:
This is briefly flagged in the footnote of the paper, but it deserves more deliberation since it is an urgent concern. Volpe does not pursue it, calling it part of a "larger discussion on the power of Big Tech". If systems like the HM are ever used in real political processes, the fact that they are designed and owned by private companies becomes a serious democratic problem. Decisions about what counts as a fair or balanced group statement are not politically neutral, they embed the values and interests of their creators. Volpe's critiques assume the system is trying to serve the public good, but who guarantees that is never discussed.