Proactive AI systems in intelligent cockpits can now initiate prompts without user commands, yet a fundamental design question remains underexplored: when should AI prompts be delivered? This study investigates how prompt timing shapes user experience in autonomous driving contexts. Using a VR-based prototype built with Unity and Meta Quest, 28 participants experienced an autonomous tourism driving scenario in a within-subjects design, encountering proactive AI prompts at three temporal positions relative to driving events: Before, During, and After. User experience was measured across four dimensions using validated scales. Repeated measures ANOVA revealed significant main effects of prompt timing on all four measures (all p < .001), with large effect sizes (η²p = .353–.587). The most consistent pattern was a clear divide between timely and delayed prompts: both “Before” and “During” conditions significantly outperformed “After” across trust, usefulness, and satisfaction, while differing from each other only on overall experience satisfaction (UEQ-S, p = .006). “During” prompts imposed the highest cognitive load, and “After” prompts were consistently rated lowest. These results point to what this study terms temporal alignment—the synchronization of system behavior with the user's cognitive rhythm—as a core mechanism shaping the quality of proactive AI interaction. Based on these findings, a layered timing framework is proposed for intelligent cockpit design, assigning each temporal position a distinct experiential function: “Before” for anticipatory trust-building, “During” for selective real-time support, and “After” for reflective experience extension. Further validation in real-world driving contexts is needed to confirm the generalizability of these findings.
I sistemi di AI proattiva negli abitacoli intelligenti sono oggi in grado di avviare suggerimenti senza comandi espliciti dell’utente; tuttavia, una questione progettuale fondamentale rimane ancora poco esplorata: quando dovrebbero essere erogati i prompt dell’AI? Questo studio indaga come la tempistica dei prompt influenzi l’esperienza utente in contesti di guida autonoma. Utilizzando un prototipo basato su VR sviluppato con Unity e Meta Quest, 28 partecipanti hanno vissuto uno scenario di guida turistica autonoma secondo un disegno entro-soggetti, sperimentando prompt proattivi dell’AI in tre posizioni temporali rispetto agli eventi di guida: Prima, Durante e Dopo. L’esperienza utente è stata misurata su quattro dimensioni mediante scale validate. Un’ANOVA a misure ripetute ha evidenziato effetti principali significativi della tempistica dei prompt su tutte e quattro le misure (tutti p < .001), con ampie dimensioni dell’effetto (η²p = .353–.587). Il risultato più coerente è stata una netta distinzione tra prompt tempestivi e ritardati: sia la condizione “Prima” sia la condizione “Durante” hanno ottenuto punteggi significativamente superiori rispetto alla condizione “Dopo” in termini di fiducia, utilità e soddisfazione, differendo tra loro soltanto per la soddisfazione complessiva dell’esperienza (UEQ-S, p = .006). I prompt “Durante” hanno comportato il carico cognitivo più elevato, mentre i prompt “Dopo” sono stati costantemente valutati come i peggiori. Questi risultati evidenziano ciò che lo studio definisce allineamento temporale — la sincronizzazione del comportamento del sistema con il ritmo cognitivo dell’utente — come meccanismo centrale nella determinazione della qualità dell’interazione con l’AI proattiva. Sulla base di tali evidenze, viene proposto un framework stratificato della tempistica per la progettazione degli abitacoli intelligenti, che attribuisce a ciascuna posizione temporale una funzione esperienziale distinta: “Prima” per la costruzione anticipatoria della fiducia, “Durante” per un supporto selettivo in tempo reale e “Dopo” per l’estensione riflessiva dell’esperienza. Sono necessarie ulteriori validazioni in contesti di guida reali per confermare la generalizzabilità di questi risultati.
When to prompt: exploring proactive ai prompt timing in intelligent cockpit
WANG, XINYU
2024/2025
Abstract
Proactive AI systems in intelligent cockpits can now initiate prompts without user commands, yet a fundamental design question remains underexplored: when should AI prompts be delivered? This study investigates how prompt timing shapes user experience in autonomous driving contexts. Using a VR-based prototype built with Unity and Meta Quest, 28 participants experienced an autonomous tourism driving scenario in a within-subjects design, encountering proactive AI prompts at three temporal positions relative to driving events: Before, During, and After. User experience was measured across four dimensions using validated scales. Repeated measures ANOVA revealed significant main effects of prompt timing on all four measures (all p < .001), with large effect sizes (η²p = .353–.587). The most consistent pattern was a clear divide between timely and delayed prompts: both “Before” and “During” conditions significantly outperformed “After” across trust, usefulness, and satisfaction, while differing from each other only on overall experience satisfaction (UEQ-S, p = .006). “During” prompts imposed the highest cognitive load, and “After” prompts were consistently rated lowest. These results point to what this study terms temporal alignment—the synchronization of system behavior with the user's cognitive rhythm—as a core mechanism shaping the quality of proactive AI interaction. Based on these findings, a layered timing framework is proposed for intelligent cockpit design, assigning each temporal position a distinct experiential function: “Before” for anticipatory trust-building, “During” for selective real-time support, and “After” for reflective experience extension. Further validation in real-world driving contexts is needed to confirm the generalizability of these findings.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Wang.pdf
accessibile in internet per tutti
Dimensione
36.19 MB
Formato
Adobe PDF
|
36.19 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/250620