Some conversations seem particularly valuable, some seem pointless. What is the value of having a conversation, in general? A first pass answer to this question is immediately suggested by a prominent tradition in philosophy of language: linguistic communication is a cooperative endeavor aimed at sharing information and influencing each other (e.g., Grice 1991; Lewis 2008). On this picture, having a conversation can be seen as valuable to the extent it allows for transferring information (i.e. acquiring true beliefs, knowledge, etc.) between interlocutors and coordinating on certain actions. This seems to capture an important way in which conversations are (instrumentally) valuable. Arguably, the question about the value of conversation becomes much more pressing when we consider the growing presence and development of AI-powered speech and voice technologies. So-called intelligent conversational assistants such as Alexa or Siri have become popular commercial services (e.g. Turow 2021; McTear 2022), and are being developed to aid services in public sectors, such as healthcare and education. Recent developments in AI are likely to make such conversational agents even more powerful tools (e.g. Bansal et al. 2024; Gabriel et al. 2024). Ask your conversational assistant about how to order a taxi, you get your information. Arguably, you can learn from conversational assistants (cf. Mallory 2023). Thus, on the first pass answer, you can have a valuable conversation with them. And yet, one may still wonder whether and to what extent conversations with such assistants are valuable in the same way as those among regular interlocutors. In this talk, I will focus on one important aspect of interactions with conversational assistants, i.e. that of illusion. Conversations with AI-powered conversational assistants may involve various ‘local’ illusions. My first goal will be to explicate what kind(s) of illusions may be involved, including, the illusion of: there being another interlocutor, there being a source of information provided, of interacting with that source and, in some cases, of having a relationship with them. On some views, our engagement with chatbots is best understood as a form of prop-oriented make-believe (Mallory 2023). Conversations with AI-powered assistants seem to be different from those among regular interlocutors with respect to how they relate to the truth. This makes our experience with them illusory (or virtual) in an interesting sense. But whether this makes such conversations less valuable (or dis-valuable) is an open and complicated question. My second goal will be to make some progress on it by discussing some classic thought experiments such as Nozick’s experience machine, as well as recent arguments concerning entering virtual reality (e.g., Chalmers 2022).

Since Machery et al. (2004), a number of experimental studies have been conducted on the
reference of proper names. Even though details vary, all existing studies have featured vignettes
with a structure that models Kripke’s famous Gödel and Jonah cases, followed by a question
probing the test subjects’ reference assignments, where one of the two options has been interpreted
as supporting the causal-historical account, and the other as supporting descriptivism (e.g. Devitt &
Porot 2018, Li et al. 2018, Domaneschi & Vignolo 2020). These studies have, for the most part,
claimed to find evidence for the causal-historical account, and against descriptivism. Doubts
remain, however, about the validity of the setup. In particular (cf. Author XXXX), it is reasonable
to expect that the test subjects will associate information stated in the vignette with the name in
question, and that the response alternative that has been interpreted as supporting the
causal-historical account is consequently fully compatible with a descriptivist theory of names. If
so, these responses do not support the causal-historical account, after all. In earlier work (Author &
Author XXXX) we have substantiated this worry by showing that similar proportions of
purportedly causal-historical responses can be elicited with analogous vignettes featuring terms of
which the causal-historical account is very likely false.

In this paper, we present results from a new set of experiments. Our setup is
methodologically novel in two ways. First, unlike in earlier work, the vignettes we used did not use
the proper names the reference of which is under study, but merely mentioned them. This allowed
us to remove what we took to be problematic aspects of the setups modelled on Kripke’s Gödel and
Jonah cases. Inspired by Johnson & Nado (2014), our vignettes described two individuals, but
referred to them as “X” and “Y”, instead of the proper names of interest. In the vignettes, X is
described as possessing the properties typically associated with name N, while Y is a different
individual. However, the vignettes state that Y, and not X, was originally called by name N, while X
was called by another name. The vignettes then go on to state that the two names got mixed up at
some point, which is why speakers today typically associate the properties they do with name N.
The subjects were then asked to judge which individual, X or Y, is the referent of N. The
causal-historical account predicts that name N continues to refer to Y (the original bearer of N),
while the descriptivist prediction is that name N, in current use, refers to X (who satisfies all the
properties associated with N). These vignettes featured familiar names such as “Isaac Newton” and
“Beyoncé”.

The second novelty of our setup is that, in addition to the vignettes just described, we used
fully analogous vignettes which featured, instead of names, terms of which the causal-historical
account is plausibly false, namely artefact terms (such as “guitar” and “shovel”) and definitional
terms (such as “bachelor” and “prime number”). This was done to test the validity of the setup. It is
prima facie plausible to suppose that the extensions of artefact terms and definitional terms are
determined by associated intended function, and associated descriptions or definitions, respectively. If
the pattern of responses is found to be significantly different for these terms, as compared to proper
names, such that subjects prefer the opposite response in an otherwise analogous setting, this would
show that the participants’ responses are tracking a difference in how the relevant terms’ extensions are
determined, and are thus not the result of some feature of the setup that has nothing to do with which
theory of reference is true. If so, the responses predicted by the causal-historical account for the
vignettes featuring names would provide much stronger support for the theory than previous studies have
done.

This was exactly what we found. For the proper name cases, a clear majority of responses
(~64%) were as predicted by the causal-historical account. For the artefact terms, by contrast, a very
clear majority of responses (~74%) were as predicted by the descriptivist account, and for the
definitional terms, this majority was even greater (~81%). Our first conclusion is that our setup
represents a major step forward methodologically, in experimentally testing the reference of proper
names. Our second conclusion is that the reference of proper names is unlikely to be determined in
the same way as the extensions of artefact terms and definitional terms. Our results thereby provide
tentative empirical support for the causal-historical account of names.

References

Devitt, M. & Porot, N. (2018). The Reference of Proper Names: Testing Usage and Intuitions.
Cognitive Science, 42(5), 1552–1585.
Domaneschi, F., & Vignolo, M. (2020). Reference and the Ambiguity of Truth‐Value Judgments.
Mind & Language, 35(4), 440–455.
Johnson, M. & Nado, J. (2014). Moderate Intuitionism: A Metasemantic Account. In Booth, A. &
Rowbottom, D. (Eds.). Intuitions (pp. 68-90). Oxford: Oxford University Press.
Li, Liu, L., Chalmers, E., & Snedeker, J. (2018). What Is in a Name? Cognition, 171, 108–111.
Machery, E., Mallon, R., Nichols, S. & Stich, S. P. (2004). Semantics, Cross-Cultural Style.
Cognition, 92(3), B1–B12.