| Abstract [eng] |
In recent years, there has been an ongoing discussion concerning versions of artificial intelligence (AI) that have been granted the capacity to imagine. Google’s machine-learning research team, DeepMind, has developed techniques for enhancing deep reinforcement learning through what they term imagination-based planning. Their claim is that algorithms can “benefit from model-based imagination without the pitfalls of conventional model-based planning”. Ironically, this proposal resonates with a Kantian move: to situate the function of Einbildungskraft — productive imagination — at the heart of cognitive processes. For DeepMind, the future of AI is tied to its ability to imagine, grounded in the premise that intellection is impossible without imagination. The imagination attributed to such agents is circumscribed by the constructed environment itself — a simulation of possibility spaces rather than an open horizon of creativity. The answer to the question "Can AI imagine?" should thus be framed by emphasising two key points. The theme of imagination points toward trajectories of relation; hence, imaginative processes should be seen as cyclical phases of connection that ceaselessly recalibrate the trajectories of human and nonhuman imagination. The question, therefore, is not whether human intelligence can imagine, but how imagination becomes possible in technobiological togetherness. At the same time, this aspect indicates what could be called negotiative imagination in Georges Canguilhem’s sense. |
