# Draft of Eindhoven Research Proposal This proposal develops an account of how generative AI systems can contribute to human flourishing in creative and intellectual work. On a tradition going back to Aristotle, human flourishing is a matter of exercising and developing one's capacities, a view taken up in several recent accounts (e.g. Bradford 2015; Dennis 2020)%%is there a name for these accounts? if there is we should use it rather than going back to aristotle by name%%. The emergence of generative AI systems has put pressure on creatives' and knowledge workers' opportunities to flourish in this way. If Midjourney can produce a finished illustration from a prompt, why train as an illustrator? Given the recent advances made in mathematics by GPT 5.5 Pro, why train as a mathematician? Arguably, the reason why these systems threaten creative flourishing is because they make certain forms of creation _frictionless_. An image or a piece of text can be produced from a prompt in moments, with little knowledge or effort from the user. Danaher and Nyholm (2021) describe such cases as opening an _achievement gap_, in which the product of an activity is still produced while the person no longer carries out the part of it that was worth doing. The effort these systems remove, however, is the exercise of the capacities in which flourishing was said to consist.%%this last sentnece is bad%% This reaches even those who decline to use such systems%%could be clearer%%. Once the frictionless route exists, working in the old way becomes a deliberate choice to take the harder road, and invites the charge that the person is merely placing obstacles in the way of their own creation. Obstacles adopted for their own sake turn an activity into a game (Nguyen 2020)%%not how i write, must be qualified etc.%%, so on that description the person who opts out has not protected their flourishing; they have made their work into a game. The threat cannot be met, then, by keeping the work hard. The difficulty worth keeping has to be told apart from arbitrary obstacle: difficulty internal to the work, met with skill and tied to the person's own purposes, as against difficulty imposed from outside for its own sake. Progress on this issue, I suggest, requires a recognition that generative AI systems are quite different to ordinary artifacts or digital artifacts. First, while a user has complete control over the hammer they are holding, or the word processor they are using, with an LLM or an image generator, this control is only ever partial. In previous work, my co author and I have compared such systems to nature, in that they "autonomously generate and dynamically shape things in ways that are so complex and chaotic to appear unpredictable". Second, due to the very large amount of data (e.g. images, text) that they are trained on, generative AI systems are _potent_: able to produce an extraordinarily wide range of outputs. These features mean that eliciting worthwhile outputs from such systems is itself a skill. Results have to be coaxed from a process that runs partly autonomously, an ability I have compared to a gardener growing a garden in previous work. Coaxing of this kind asks the user to hold in view what they are after and to tell, among what the system returns, which of it answers to that aim. The capacity exercised is one of recognition and judgement rather than of specifying a form in advance. This is the part of the activity the achievement gap threatened to strip out, so working this way keeps the person doing what was worth doing rather than handing it over. And because such a skill grows with practice, working this way develops the user's capacities as well as exercising them, so engaged use can contribute to a person's flourishing rather than diminish it, which is what flourishing was said to require over the course of a life (Dennis 2020). The judgement described so far is exercised across a working relation that develops over time, rather than in any single act of prompting. The relation that forms between a user and a system over repeated work is the unit on which assessment has to fix. A system does not support a person's flourishing simply by returning good outputs, and a person does not flourish simply by being content with what comes back. What can be assessed is the relation itself, and whether it keeps the user at work within it rather than reducing them to a recipient of finished results. Because that work is done in the user's own words, the judgement that selects and shapes what the system returns can carry the user's voice into the making as well as into the result. Each side of this relation answers to the other. The user's exercise of capacity depends on the behaviour of a partly autonomous system, and the system's worth depends on the kind of practice it sustains in the user. What goes well or badly is this dependence taken as a whole, at once human and technical, and its going well is the flourishing of a sociotechnical system. When is the difficulty of working with such a system worth keeping, and when is it merely an obstacle? There is then the question of agency exercised under partial control, where the user has neither imposed a finished form nor merely accepted what the system supplied. Both bear on a further question: how generative systems, and the practices that grow around them, should be built and assessed so that they support the exercise and development of a person's capacities rather than only delivering products. Whether a user can exercise this judgement at all depends on how a system is designed and on the setting into which it is introduced. A system can be taken up so that it invites the user to judge and revise what it returns, or so that accepting a finished product becomes the path of least resistance. The conditions of a flourishing practice are in this way partly a matter of design and institutional uptake, and the third question cannot be answered by philosophical argument alone. Its object is a practice, and the people who carry such practices on hold part of the knowledge of what the work is that they are trying to preserve or to change. The project would therefore be pursued with practitioners who use these systems in creative and intellectual work, and with those who introduce them into working settings, through co-creation under Bombaerts and within the ESDiT line, and alert to the doubts co-creation has itself drawn (Bombaerts et al. 2025). The conceptual contribution would be an account of agency under partial control, worked out through aesthetics and the philosophy of technology. The practical contribution would be a framework for telling when a generative-AI practice sustains a person's exercise of their capacities and when it stands in for them.