Currie, Brittney (2026) Why We Cannot Empathize With ChatGPT: Nor Any Other Agent Lacking Epistemic States, Values, and or Interests. [Preprint]

Text (Accepted with Minor Revisions) Currie_Why We Cannot Empathize With ChatGPT.docx - Draft Version Download (69kB)

Abstract

Extended Abstract: Whether we can or ought ‘empathize’ with chatbots is a central question in public discourse philosophy, and psychology.
Historically, we’ve approached the question of whether we could empathize with AI as a hypothetical science fiction scenario—some form of AI emerges with some sets of agential capacities, rendering them either a moral or epistemic agent or patient. Since this kind of AI would warrant special moral or epistemic treatment it necessitates empathic regard. Fulfilling moral and epistemic obligations to other agents often requires empathizing—or taking up their perspective- so we can meet their needs, understand their desires, and treat them accordingly. Because empathizing entails considering an agent’s needs, desires, emotions, beliefs, subjective- values, and interests, we must determine whether chatbots have these qualities and capacities. Agents we typically empathize with have epistemic states, values, and interests. Yet,
LLMs, particularly, ChatGPT 4o/5.1 lack: 1) emotions (see Mitchell, 2019 for discussion), 2) typical propositional attitudes (ea. beliefs and desires of like humans, animals, and fictional characters).
Importantly, GPT also lacks: 3) interests of the kind people and other beings have; 4) or needs that can be met; finally, GPT lacks 5) typical subjective, or
agent-relative values (ea. unwavering faith in a particular god). Moreover, GPT lacks typical objective values (ea. freedom, peace, or care). Given these deficits, I argue that we cannot
empathize with GPT on our orthodox psychological and philosophical accounts of empathy. Much contemporary philosophical and psychological literature describing interactions with GPT
indirectly imply we empathize with GPT, or assume we empathize with
GPT. Some literature makes normative claims that
we ought empathize with GPT on grounds of moral prudence. I try to answer the objective question of whether we can. If our empirically informed
accounts of empathy in psychology, coupled with our best philosophical accounts of empathy, render GPT a candidate for empathy, then, we can empathize with GPT. On classical accounts
some entities are ruled out because they lack requisite capacities and features required for humans to uptake their POV. Rocks intuitively lack ‘perspective.’ Similarly, some forms of AI, like GPT, lack a POV in the robust sense- yet the reasons aren’t obvious. We can’t empathize with some entities because they a) lack necessary features for up-taking their POV (ea. epistemic states, values, interests,); or 2) the psychological processes humans undergo
engaging these entities don’t amount to empathizing. If we’re committed to paradigmatic accounts of empathy, then, these accounts should sufficiently explain interactions with GPT as categorically, empathetic. There’s two options: 1) given limited observational evidence of interacting with GPT, we revise existing accounts or 2) following standard scientific practice,
reject we empathize with GPT. When observing newly emerging psychological phenomenon, we determine whether these new observations can be classified as an existing kind. I evaluate 3 paradigmatic accounts
of empathy I characterize as follows: 1) ‘The Emotional-Empathy Account,’ 2) ‘The Subjective Values Account’ and 3) The Cognitive Empathy Account,
demonstrating each standard account fails to explain our interactions with GPT as empathetic. GPT lacks, interests (rights, capacity for well-being), values (endogenous motivations, emotions, agential preferences, goals), and epistemic states (beliefs, desires, propositional attitudes)—all which compose a typical agent’s POV. What falls out each classical account’s explanatory failure, taken in sum, is that GPT has no substantive POV. I address anticipated objections and respond.

---

Export/Citation:

EndNote | BibTeX | Dublin Core | ASCII/Text Citation (Chicago) | HTML Citation | OpenURL

Social Networking:

Share |

---

Item Type:	Preprint
Creators:	Creators Email ORCID Currie, Brittney currieb@iu.edu
Subjects:	Specific Sciences > Artificial Intelligence > AI and Ethics Specific Sciences > Cognitive Science
Depositing User:	Dr. Brittney Currie
Date Deposited:	11 May 2026 12:43
Last Modified:	11 May 2026 12:43
Item ID:	29565
Subjects:	Specific Sciences > Artificial Intelligence > AI and Ethics Specific Sciences > Cognitive Science
Date:	9 May 2026
URI:	https://philsci-archive.pitt.edu/id/eprint/29565

Monthly Views for the past 3 years

Monthly Downloads for the past 3 years

Plum Analytics

Actions (login required)

View Item