Do AI generated outputs have contents?

This project asks: do the outputs of machine learning neural net AI systems (henceforth, ‘AI systems’) have contents i.e., do they have meanings, express propositions, or represent objects of the world?

It approaches this larger question by dividing it up into three smaller, closely aligned questions: (a) Are mental capacities necessary for producing outputs with content? (b) What does it take for an entity to have those capacities? and, (c) Do AI systems have those capacities?

These foundational questions underpin and support several further investigations concerning the potential and threats of AI systems, including their role in spreading propaganda, disinformation, and fake news, and the question of whether AI systems can develop their own language.

The project’s main innovation is its humanistic approach — its investigations are guided by the hypothesis that any philosophically interesting notion of ‘content’ must be comprehensible, and apply to human speech, images, and artefacts.

The project combines the expertise of the Experienced Researcher (ER) on the semantics of natural language expressions with the current research of the supervisor (Matti Eklund) on ‘alien’ languages (i.e., languages semantically and structurally different from human languages).

The project will greatly enhance the ER’s career prospects because of the training and mentorship that the ER will receive as part of the project, and because of the close synergy between the ER’s research focus, the supervisor’s internationally renowned expertise on semantics and metaphysics, and the host organisation’s (Uppsala University) strengths in pragmatics, artificial intelligence, and computational linguistics.

Conversely, the ER’s research will enrich and generate substantial added value for the host institution — particularly in the research and academic work on meta-semantics, metaphysics, and the theoretical and practical aspects of AI systems carried out at the host institution.