The integration of Large Language Models (LLMs) into autonomous vehicles (AVs) aims to enhance human-AV interaction through natural language explanations. However, the extent to which driver awareness of LLM involvement, and explanation format itself, influences driver experience during unexpected AV behavior remains largely unexplored. In this study, we investigated the impact of drivers’ beliefs about the explanation source (i.e., LLM-generated vs. human experts) across three explanation formats: causal, counterfactual, and contrastive. Results showed a tendency toward greater trust in human expert-derived explanations, particularly in counterfactual scenarios. Furthermore, contrastive explanations significantly reduced mental workload compared to causal and counterfactual formats, demonstrating a distinct advantage in cognitive processing efficiency. These findings indicate that strategically tailoring the selection of explanation source and format to the time-sensitivity and complexity of specific driving scenarios could improve driver trust and performance.
AvetisyanL.AyoubJ.ZhouF. (2022). Investigating explanations in conditional and highly automated driving: The effects of situation awareness and modality. Transportation Research Part F: Traffic Psychology and Behaviour, 89, 456–466. https://doi.org/10.1016/j.trf.2022.07.010
2.
AyoubJ.AvetisianL.YangX. J.ZhouF. (2023). Real-time trust prediction in conditionally automated driving using physiological measures. IEEE Transactions on Intelligent Transportation Systems, 24(12), 14642–14650. https://doi.org/10.1109/TITS.2023.3295783
3.
AyoubJ.AvetisyanL.MakkiM.ZhouF. (2022). An investigation of drivers’ dynamic situational trust in conditionally automated driving. IEEE Transactions on Human-Machine Systems, 52(3), 501–511. https://doi.org/10.1109/THMS.2021.3131676
4.
DietvorstB. J.SimmonsJ. P.MasseyC. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114–126. https://doi.org/10.1037/xge0000033
5.
DingY.FaccianiM.PoudelA.JoyceE.AguinagaS.VeeramaniB.BhattacharyaS.WeningerT. (2025). Citations and trust in LLM generated responses. Proceedings of the AAAI Conference on Artificial Intelligence, 39(22), 23787–23795.
6.
DirsehanT.CanC. (2020). Examination of trust and sustainability concerns in autonomous vehicle adoption. Technology in Society, 63, Article 101361. https://doi.org/10.1016/j.techsoc.2020.101361
7.
GentileD.DonmezB.JamiesonG. A. (2025). Human performance effects of combining counterfactual explanations with normative and contrastive explanations in supervised machine learning for automated decision assistance. International Journal of Human-Computer Studies, 196, Article 103434. https://doi.org/10.1016/j.ijhcs.2024.103434
8.
HoffmanR. R.MuellerS. T.KleinG.LitmanJ. (2018). Metrics for explainable AI: Challenges and prospects. arXiv preprint. https://doi.org/10.48550/arXiv.1812.04608
9.
HolthausenB. E.WintersbergerP.WalkerB. N.RienerA. (2020). Situational Trust Scale for Automated Driving (STS-AD): Development and initial validation [Conference session]. 12th International Conference on Automotive User Interfaces and Interactive Vehicular Applications (pp. 40–47). Association for Computing Machinery. https://doi.org/10.1145/3409120.3410637
10.
KimG.YeoD.JoT.RusD.KimS. (2023). What and when to explain? On-road evaluation of explanations in highly automated vehicles. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 7(3), 1–26. https://doi.org/10.1145/3610886
11.
McKnightD. H.KacmarC. J. (2007). Factors and effects of information credibility [Conference session]. Proceedings of the Ninth International Conference on Electronic Commerce, Minneapolis, MN (pp. 423–432). Association for Computing Machinery. https://doi.org/10.1145/1282100.1282180
NtoutsiE.FafaliosP.GadirajuU.IosifidisV.NejdlW.VidalM.RuggieriS.TuriniF.PapadopoulosS.KrasanakisE.KompatsiarisI.Kinder-KurlandaK.WagnerC.KarimiF.FernandezM.AlaniH.BerendtB.KruegelT.HeinzeC., . . . StaabS. (2020). Bias in data-driven artificial intelligence systems—An introductory survey. WIREs Data Mining and Knowledge Discovery, 10(3), Article e1356. https://doi.org/10.1002/widm.1356
14.
OmeizaD.KollnigK.WebH.JirotkaM.KunzeL. (2021). Why not explain? Effects of explanations on human perceptions of autonomous driving [Conference session]. IEEE International Conference on Advanced Robotics and Its Social Impacts (ARSO), Tokoname, Japan (pp. 194–199). IEEE. https://doi.org/10.1109/ARSO51874.2021.9542835
15.
OmeizaD.WebH.JirotkaM.KunzeL. (2021). Towards accountability: Providing intelligible explanations in autonomous driving [Conference session]. IEEE Intelligent Vehicles Symposium (IV), Nagoya, Japan (pp. 231–237). IEEE. https://doi.org/10.1109/IV48863.2021.9575917
16.
PauziéA. (2008). A method to assess the driver mental workload: The driving activity load index (DALI). IET Intelligent Transport Systems, 2(4), 315–322. https://doi.org/10.1049/iet-its:20080023
17.
PerkovićG.DrobnjakA.BotičkiI. (2024). Hallucinations in LLMs: Understanding and addressing challenges [Conference session]. 47th MIPRO ICT and Electronics Convention (MIPRO), Opatija, Croatia (pp. 2084–2088). IEEE. https://doi.org/10.1109/MIPRO60963.2024.10569238
18.
PornpitakpanC. (2004). The persuasiveness of source credibility: A critical review of five decades’ evidence. Journal of Applied Social Psychology, 34(2), 243–281. https://doi.org/10.1111/j.1559-1816.2004.tb02547.x
19.
RudinC. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1(5), 206–215. https://doi.org/10.1038/s42256-019-0048-x
20.
WachterS.MittelstadtB.RussellC. (2017). Counterfactual explanations without opening the black box: Automated decisions and the GDPR. Harvard Journal of Law & Technology, 31, 841–887.
21.
ZhangQ.YangX. J.RobertL. P. (2021). What and when to explain? A survey of the impact of explanation on attitudes toward adopting automated vehicles. IEEE Access, 9, 159533–159540. https://doi.org/10.1109/ACCESS.2021.3130489
