Michael Barnes

• Barnes, Michael; Dylan, Campbell,; Yunzhong, Hou,; Rahul, Shome, & Changlin, Song, (2026). Extreme Amodal Face Detection. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). ISSN 2472-6737. p. 2934-2943–2934-2943. doi: 10.1109/WACV61042.2026.00287. Full text in Research Archive Show summary

  Extreme amodal detection is the task of inferring the 2D location of objects that are not fully visible in the input image but are visible within an expanded field-of-view. This differs from amodal detection, where the object is partially visible within the input image, but is occluded. In this paper, we consider the sub-problem of face detection, since this class provides motivating applications involving safety and privacy, but do not tailor our method specifically to this class. Existing approaches rely on image sequences so that missing detections may be interpolated from surrounding frames or make use of generative models to sample possible completions. In contrast, we consider the single-image task and propose a more efficient, sample-free approach that makes use of the contextual cues from the image to infer the presence of unseen faces. We design a heatmap-based extreme amodal object detector that addresses the problem of efficiently predicting a lot (the out-of-frame region) from a little (the image) with a selective coarse-to-fine decoder. Our method establishes strong results for this new task, even outperforming less efficient generative approaches. Code, data, and models are available at https://charliesong1999.github.io/exaft_web/.

• Barnes, Michael (2025). Complex Harms in Online Speech: The Limits of the Illocutionary. In Connolly, Patrick Joseph; Goldberg, Sanford & Saul, Jennifer (Ed.), Conversations Online: Explorations in Philosophy of Language. Oxford University Press. ISSN 9780198872085. p. 239–260. doi: 10.1093/oso/9780198872085.003.0012. Full text in Research Archive Show summary

  The internet is, at heart, a communication platform. For this reason, there is a strong case to be made that speech act theory is well positioned to function as a useful theoretical framework for the many problems concerning online speech. I argue, however, that the complexity of harmful speech mediated through online channels renders the traditional elevation of illocutionary acts over perlocution effects inapt; that is, the emphasis on the act constituted by an utterance over its causal effects limits the applicability of one dominant approach to speech act theory—which I call the “illocutionary approach”—to emerging ethical issues. To make this case, I consider three key examples that demonstrate the variety of harms attributable to online speech. These include the targets of online harassment campaigns, the offline victims of viral hate, and the plight of the human content moderators who sift toxic content for little pay in hazardous working conditions.

• Barnes, Michael (2023). Who Do You Speak For? And How?: Online Abuse as Collective Subordinating Speech Acts. Journal of Ethics and Social Philosophy. 25(2), p. 251–281. doi: 10.26556/jesp.v25i2.1561. Full text in Research Archive Show summary

  A lot of subordinating speech has moved online, which raises several questions for philosophers. Can current accounts of oppressive speech adequately capture digital hate? How does the anonymity of online harassers contribute to the force of their speech? This paper examines online abuse and argues that standard accounts of licensing and accommodation are not up to the task of explaining the authority of online hate speech, as speaker authority often depends on the community in more ways than these accounts suggests. Instead I argue that online abusive speech is best understood as collective subordinating speech acts, as their authority is drawn from an ad hoc collective. I argue that anonymity and shared language offer online abusers a path to a type of group-authority that explains the harm their speech is capable of. I close by suggesting that similar considerations are in play for IRL subordinating speech, and that online abuse helps to reveal key features of subordinating speech across mediums that are under-emphasized in the existing literature.

• Barnes, Michael (2023). Presupposition and Propaganda: A Socially Extended Analysis. In Caponetto, Laura & Labinaz, Paolo (Ed.), Sbisà on Speech as Action. Palgrave Macmillan. ISSN 9783031225284. p. 275–298. doi: 10.1007/978-3-031-22528-4_12. Full text in Research Archive Show summary

  Drawing on work from Marina Sbisà’s “Ideology and the Persuasive Use of Presupposition” (1999), Rae Langton has developed a powerful account of the subtle mechanisms through which hate speech and propaganda spread. However, this model has a serious limitation: it focuses too strongly on individual speech acts isolated from their wider context, rendering its applicability to a broader range of cases suspect. In this chapter, I consider the limits of presupposition accommodation to clarify the audience’s role in helping hate speakers, and claim the necessary mechanisms are much more active and social than Langton suggests, with effects not capturable in the analysis of isolated speech acts. I revisit Sbisà’s work on presupposition and develop an alternative model that (1) improves upon the Langtonian model for explaining the power of hate propaganda, and (2) is more in line with Sbisà’s original approach to ideology and the normative aspects of presupposition.

• Barnes, Michael (2022). Online Extremism, AI, and (Human) Content Moderation. Feminist Philosophy Quarterly. 8(3/4). doi: 10.5206/fpq/2022.3/4.14295. Full text in Research Archive Show summary

  This paper has three main goals: (1) to clarify the role of artificial intelligence (AI)—along with algorithms more broadly—in online radicalization that results in “real world violence,” (2) to argue that technological solutions (like better AI) are inadequate proposals for this problem given both technical and social reasons, and (3) to demonstrate that platform companies’ (e.g., Meta, Google) statements of preference for technological solutions functions as a type of propaganda that serves to erase the work of the thousands of human content moderators and to conceal the harms they experience. I argue that the proper assessment of these important, related issues must be free of the obfuscation that the “better AI” proposal generates. For this reason, I describe the AI-centric solutions favoured by major platform companies as a type of obfuscating and dehumanizing propaganda.

• Barnes, Michael (2021). Positive Propaganda and the Pragmatics of Protest. In Hogan, Brandon; Cholbi, Michael; Madva, Alex & Yost, Benjamin S. (Ed.), The Movement for Black Lives: Philosophical Perspectives. Oxford University Press. ISSN 9780197507810. p. 139–159. doi: 10.1093/oso/9780197507773.003.0007. Full text in Research Archive Show summary

  Abstract This chapter examines what protest is from a pragmatic point of view and how it relates to propaganda—specifically what Jason Stanley calls “positive propaganda.” It analyzes the phrase “Black lives matter,” taking it to be a political speech act that offers a unique route to understanding the pragmatics of protest. From this, it considers the moral-epistemological function of protest and develops an account of the authority that protest, as a speech act, both calls upon and makes explicit. It then argues that, rather than simply its effects, it is protest’s distinct pragmatic features—that is, its entitlement conditions and the uptake it aims at—that best capture its important moral, political, and epistemic elements. It therefore rejects the idea that protests are paradigmatic examples of “positive propaganda,” because the propaganda model cannot capture protests’ function of foregrounding the socially located moral authority of the protestor.

• Barnes, Michael (2016). Speaking with (Subordinating) Authority. Social Theory and Practice. ISSN 0037-802X. 42(2), p. 240–257. doi: 10.5840/soctheorpract201642214. Full text in Research Archive Show summary

  In “Subordinating Speech,” Ishani Maitra defends the claim that ordinary instances of hate speech can sometimes constitute subordination. While she accepts that subordinating speech requires authority, she argues that ordinary speakers can acquire this authority via a process of “licensing.” I believe this account is interestingly mistaken, and in this paper I develop an alternative account. In particular, I take issue with what I see as the highly localized character of Maitra’s account, which effectively divorces the subordinating authority of ordinary hate speech from the broader normative context, including social and pragmatic features that I claim play essential roles in subordination.

View all works in NVA

• Luvell, Anderson, & Barnes, Michael (2022). Hate Speech (Stanford Encyclopedia of Philosophy). In Zalta, Edward N. (Eds.), Stanford Encyclopedia of Philosophy. http://plato.stanford.edu/entries/federalism/. Full text in Research Archive

View all works in NVA