Conitzer, Vincent (2025) What Would It Look Like to Align Humans with Ants? [Preprint]
![[img]](https://philsci-archive.pitt.edu/style/images/fileicons/text.png) |
Text
conitzer_ant_aligned_humans.pdf Download (297kB) |
Abstract
When we discuss aligning today’s AI systems with our interests, we have a decent sense of what success would look like. But many researchers are explicitly interested in aligning future superintelligent AI, whose intelligence far exceeds our own across the board. In this chapter, I argue that if AI indeed becomes superintelligent, then it will also be difficult to instruct it in a sensible way. The following well-studied issues are not what I focus on, though they are important as well: (1) whether the AI would actually want to follow these instructions, (2) whether it would even be a good thing if it followed the instructions (e.g., as opposed to caring for itself as a moral patient), or, for the most part, (3) whether it would take these instructions too literally (cf. Goodhart’s Law). Rather, I focus on the following issue: it is likely that the superintelligent AI will have options available to it that we humans could not have dreamed of, and to which our concepts are an awkward fit at best. But it is impossible to illustrate this with direct examples; since we are human beings ourselves, we cannot provide examples of options that humans could not have dreamed of. Instead, in this chapter, I rely on an analogy: suppose ants had somehow been in a position to align humans with their interests. How could this have been done in a way that, from the perspective of the ants, can be considered successful? Through a sequence of imagined memoirs of humans that are aligned with ants in various ways, I argue that there does not appear to be any completely satisfactory answer to this question.
| Export/Citation: | EndNote | BibTeX | Dublin Core | ASCII/Text Citation (Chicago) | HTML Citation | OpenURL |
| Social Networking: |
| Item Type: | Preprint | ||||||
|---|---|---|---|---|---|---|---|
| Creators: |
|
||||||
| Additional Information: | A later version will appear as Chapter 17 in Nyholm, Sven, Kasirzadeh, Atoosa, & Zerilli, John (eds.) (2026): Contemporary Debates in the Ethics of Artificial Intelligence. Hoboken: Wiley-Blackwell. | ||||||
| Keywords: | artificial intelligence; superintelligence; alignment | ||||||
| Subjects: | Specific Sciences > Artificial Intelligence > AI and Ethics Specific Sciences > Artificial Intelligence Specific Sciences > Artificial Intelligence > Machine Learning |
||||||
| Depositing User: | Prof. Vincent Conitzer | ||||||
| Date Deposited: | 24 Aug 2025 17:37 | ||||||
| Last Modified: | 24 Aug 2025 17:37 | ||||||
| Item ID: | 26351 | ||||||
| Subjects: | Specific Sciences > Artificial Intelligence > AI and Ethics Specific Sciences > Artificial Intelligence Specific Sciences > Artificial Intelligence > Machine Learning |
||||||
| Date: | 23 August 2025 | ||||||
| URI: | https://philsci-archive.pitt.edu/id/eprint/26351 |
Monthly Views for the past 3 years
Monthly Downloads for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](http://philsci-archive.pitt.edu/style/images/feed-icon-32x32.png){kind=icon}