04 Chat.pdf - 墨天轮文档

04 Chat.pdf

Zixin Huo

101

7页

0次

2023-10-17

25墨值下载

2020 IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO. 3, MARCH 2023

Letters

Chat With ChatGPT on Intelligent Vehicles: An IEEE

TIV Perspective

Haiping Du , Senior Member, IEEE,SiyuTeng , Hong Chen , Fellow, IEEE,JiaqiMa ,XiaoWang ,

Chao Gou

,BaiLi , Siji Ma , Qinghai Miao , Xiaoxiang Na , Peijun Ye , Hui Zhang , Guiyang Luo ,

and Fei-Yue Wang

, Fellow, IEEE

Abstract—This letter reports on a TIV DHW (decentralized

and hybrid workshop) that explores the prospective inﬂuence

of ChatGPT on research and development in intelligent vehi-

cles. To assess the update capabilities of ChatGPT, we conducted

tests involving both basic and technically relevant questions. Our

Manuscript received 1 March 2023; accepted 1 March 2023. Date of pub-

lication 6 March 2023; date of current version 27 April 2023. This work

was supported in part by National Key R&D Program of China under Grant

2022YFB2502900. (Corresponding author: Fei-Yue Wang.)

Haiping Du is with the School of Electrical, Computer and Telecommu-

nications Engineering, University of Wollongong, Wollongong, NSW 2522,

Australia (e-mail: hdu@uow.edu.au).

Siyu Teng is with the Department of Computer Science, Faculty of Science,

Hong Kong Baptist University, Hong Kong 999077, SAR, China, and also with

the BNU-HKBU United International College, Zhuhai 519087, China (e-mail:

siyuteng22@gmail.com).

Hong Chen is with the College of Electronic and Information Engineer-

ing, Tongji University, Shanghai 201804, China (e-mail: chenhong2019@

tongji.edu.cn).

Jiaqi Ma is with the Samueli School of Engineering, University of California,

Los Angeles, CA 90095 USA (e-mail: jiaqima@ucla.edu).

Xiao Wang is with the School of Artiﬁcial Intelligence, Anhui University,

Hefei 230601, China (e-mail: xiao.wang@ahu.edu.cn).

Chao Gou is with the School of Intelligent Systems Engineering, Sun Yat-Sen

University, Guangzhou 510275, China (e-mail: gouchao@mail.sysu.edu.cn).

Bai Li is with the College of Mechanical and Vehicle Engineering, Hunan

University, Changsha 410082, China (e-mail: libai@zju.edu.cn).

Siji Ma is with the Faculty of Innovation Engineering, Macau Uni-

versity of Science and Technology, Macau 999078, China (e-mail: masi-

jiguge123@gmail.com).

Qinghai Miao is with the School of Artiﬁcial Intelligence, Univer-

sity of Chinese Academy of Sciences, Beijing 100049, China (e-mail:

miaoqh@ucas.ac.cn).

Xiaoxiang Na is with the Department of Engineering, University of Cam-

bridge, Cambridge CB2 1PZ, U.K. (e-mail: xnhn2@cam.ac.uk).

Peijun Ye is with the State Key Laboratory for Management and Control of

Complex Systems, Chinese Academy of Sciences, Beijing 100190, China, and

also with the State Key Laboratory of Multimodal Artiﬁcial Intelligence Sys-

tems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190,

China (e-mail: peijun_ye@hotmail.com).

Hui Zhang is with the School of Computer and Information Tech-

nology, Beijing Jiaotong University, B eijing 100044, China (e-mail:

huizhang1@bjtu.edu.cn).

Guiyang Luo is with the School of Computer Science and Technology, Beijing

University of Posts and Telecommunications, Beijing 100876, China (e-mail:

luoguiyang@bupt.edu.cn).

Fei-Yue Wang is with the School of Artiﬁcial Intelligence, Univer-

sity of Chinese Academy of Sciences, Beijing 100190, China ( e-mail:

feiyue.wang@ia.ac.cn).

Color versions of one or more ﬁgures in this article are available at

https://doi.org/10.1109/TIV.2023.3253281.

Digital Object Identiﬁer 10.1109/TIV.2023.3253281

preliminary testing revealed that ChatGPT’s information can be

updated and corrected at one time, but it may take some time

for the changes t o be reﬂected in ChatGPT’s responses, so it may

not always possess the latest knowledge regarding speciﬁc topics.

We further discuss the prospective inﬂuence of ChatGPT on the

ﬁeld of intelligent vehicles, particularly possible applications of

ChatGPT in areas like autonomous driving, human-vehicle interac-

tion, and intelligent transportation systems, highlighting challenges

and opportunities associated with these applications. Additionally,

we address technical questions, such as the feasibility of training

intelligent vehicles using the same methods as ChatGPT and the

reﬂection of the intelligence of intelligent vehicles in the context of

human-machine shared control. I n conclusion, this letter presents a

preliminary exploration of the potential of ChatGPT for intelligent

vehicle research, from an IEEE TIV perspective, acknowledging

the limitations and uncertainties of this emerging technology.

Index Terms—Neural language programming, intelligent

vehicles, human-vehicle interaction, intelligent transportation

systems.

I. INTRODUCTION

ECENT advances in artiﬁcial intelligence (AI) have led to

signiﬁcant progress. One particular development is natural

language processing, which has enabled the creation of chatbots

that can simulate human-like conversations. One such language

model is ChatGPT [1], which gains widespread attention

because of its powerful ability to converse with humans.

Given the potential relevance of ChatGPT to intelligent vehicle

research [2], a TIV DHW was held to explore the prospective

inﬂuence of ChatGPT on this ﬁeld [3]. The DHW was attended

by Associate Editors and the Editor-in-Chief of IEEE TIV

online as shown in Fig. 1 as well as some Associate Editors

ofﬂine, who reported their perspectives on the topic and engaged

in discussions and debates [4], [5], [6], [7], [8], [9], [10], [11],

[12], [13], [14], [15], [16].

The main objectives of the TIV DHW were:

Test the update capabilities of ChatGPT through basic

questions about intelligent vehicles, the IEEE TIV, and the

Editor-in-Chief of IEEE TIV.

Discuss possible applications of ChatGPT in areas such

as autonomous driving, human-vehicle interaction, and

intelligent transportation systems, and highlight some of

See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Shenzhen Institute of Advanced Technology CAS. Downloaded on October 11,2023 at 14:13:45 UTC from IEEE Xplore. Restrictions apply.

IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO. 3, MARCH 2023 2021

Fig. 1. TIV DHW on chat with ChatGPT.

the challenges and opportunities associated with these

applications.

Raise technical questions related to the use of ChatGPT

in intelligent vehicle research, such as the training of

intelligent vehicles and the reﬂection of intelligence in the

context of human-machine shared control [17].

Overall, the DHW provided a preliminary exploration of

the potential of ChatGPT in the context of intelligent vehicle

research, while acknowledging the limitations and uncertainties

of this emerging technology. The following sections of this letter

will provide more details on the testing of ChatGPT ’s update

capabilities, the potential applications of ChatGPT in intelligent

vehicles, and the opportunities and challenges associated with

these applications.

II. T

ESTING UPDATE CAPABILITIES OF CHATGPT

A. Questions

To assess the update capabilities of ChatGPT in the con-

text of intelligent vehicles, we designed a series of basic and

technical questions. The basic questions covered topics such

as the deﬁnition of intelligent vehicles, the IEEE TIV, and its

Editor-in-Chief. Additionally, we included questions on how to

prescreen artiﬁcial manuscripts and make the journal a ﬂagship

in the community. The technical questions focused on using

ChatGPT to train autonomous vehicles, the potential for pow-

erful driving models, and how to reﬂect the intelligence of

intelligent vehicles in the context of human-machine shared

control. Overall, these questions highlight the potential impact of

ChatGPT on intelligent vehicle research, while also identifying

key technical challenges that must be addressed in order to fully

realize its potential. The basic questions include:

1) What is an intelligent vehicle?

2) Do you know IEEE TIV?

3) Do You know who is the EiC of IEEE TIV?

4) How should IEEE TIV quickly prescreen artiﬁcial

manuscripts written by ChatGPT?

5) What efforts should be made to make IEEE TIV a world-

wide ﬂagship journal in the community of intelligent

vehicles?

6) Do you know IEEE?

7) Do you know anything about IEEE TIV?

Fig. 2. Answers to question 1.

8) Do You know who is the EiC of IEEE TIV?

The technical questions include:

9) Can we use the method of training ChatGPT to train an

intelligent vehicle with autonomous driving ability?

10) Is it possible to train a powerful driving model using

RLHF as ChatGPT does?

11) In the context of human-machine shared control, how to

reﬂect the intelligence of intelligent vehicles from the

perspective of foundation models?

12) What impact will ChatGPT bring to the ﬁeld of intelligent

vehicles?

13) How to develop an intelligent vehicle that can overcome

most of the existing drawbacks?

B. Results of Preliminary Testing

The preliminary testing of ChatGPT showed that while the

model may have outdated or incomplete information at the time

of the initial query, it can be corrected and updated with new

information as it becomes available at one time. However, it

may take some time for the changes to be reﬂected in ChatGPT’s

responses, as the updated model may need to be deployed and in-

tegrated into the system. The model was able to provide relevant

and accurate responses to most of the questions, indicating its

potential as a useful tool for researchers in the ﬁeld of intelligent

vehicles.

However, there were instances where ChatGPT provided in-

complete or incorrect information, particularly when it comes to

the most recent developments and updates in the ﬁeld. This is not

unexpected given t he dynamic nature of the ﬁeld and the frequent

changes in, for instance, editorial positions and afﬁliations.

ChatGPT also stated that “as an AI language model, I do not have

access to real-time information or current events” and indicated

that its knowledge cutoff date is September 2021 (at the time

this paper is published). The testing process also revealed some

technical limitations and uncertainties of ChatGPT, particularly

in the context of complex and specialized topics related to

intelligent vehicles. Some examples are given as follows. For

example, some answers are quite accurate and reasonable as

showninFigs.2–4. But some answers are not updated, for

example the answer shown in Fig. 5. Also some information