In  recent  years,  AI  technology  has  developed  very  rapidly,  as  Dwivedi  puts  it, "Transformative artificially intelligent tools, such as ChatGPT, designed to generate sophisticated text indistinguishable from that produced by a human, are applicable across a wide range of contexts.” (Dwivedi et al., 2023) In particular, AI tools like ChatGPT are playing an increasingly important role in many people's studies, work, and lives. Some students use it to help with writing and exam preparation, and some office workers use it to organize information and improve efficiency. Because it is easy to operate, quick to answer, and relatively intelligent in content, it has been welcomed by many users.

Although many people find ChatGPT convenient, there are also quite a few who have reservations about its use. The user's uneasiness is mainly due to common concerns, such as performance issues related to the inaccuracy of the information generated, privacy issues related to the disclosure of chat content or privacy, social problems that arise when being evaluated by others, psychological problems that feel anxious or uncomfortable when using it, and time problems caused by the need to take time to learn and fix mistakes. These concerns are very common when using new technologies. It has been noted that six types of perceived risk dimensions were investigated including financial, social, performance, time, psychological, and privacy risks (Featherman & Pavlou, 2003). The five types of risks that this study focuses on are important components of this.

To understand the impact of these concerns on people's willingness to use ChatGPT, we used a questionnaire in which users were asked to rate their experiences on performance, privacy, social, psychological, and temporal issues, and whether they  would like to continue using ChatGPT. The results show that users who believe that ChatGPT is not accurate enough or that it is a waste of time are significantly less willing to use it. However, there are also users who are willing to continue using it despite privacy or psychological concerns, indicating that its convenience and benefits alleviate these concerns to a certain extent.

With more and more popular generative AI technologies like ChatGPT, opportunities as well as challenges are presented to users. Among the most important aspects of challenges is the perceived risk, which also predominantly determines if users would want to use such technologies in the future or not. Research into the aspects of such risks takes into consideration the users' behavioral and psychological reaction.

Performance risk is the fear that ChatGPT would fail to function as anticipated and provide low-quality or low-accuracy outputs. As users perceive that the system lacks the ability to fulfill their requirements consistently or provide stable outputs, they further refrain from using it. A literature report points out “GAI-generated analysis may reflect biased or discriminatory content on which it was trained.     Along with fact- checking the veracity of ChatGPT and other GAI output, users should be attuned to any discriminatory or biased statements or conclusions resulting in the algorithmic mining of such source materials. This could be a particular concern in the context of employment discrimination laws and laws regulating the use of artificial intelligence in employment decisions.”(Neuburger,2023)

https://www.lexisnexis.com/pdf/practical-guidance/ai/chatgpt-risks-and-the-need-for- corporate-policies.pdf

Privacy  risk  entails  issues  regarding  the  safeguarding  of  personal  information. Because  interactions  with AI platforms  may  encompass  confidential  information, users may be apprehensive about data misuse, unauthorized data collection, or data exposure to third parties. This could be a major deterrent to users' confidence in the platform over time. Although there are privacy protection mechanisms in ChatGPT, such as the block of access to personal data about individuals, it is not guaranteed that no leakage of its training data would occur. Malicious attacks, such as jailbreaking attacks, may utilize its great  generation  ability to infer  some information from personal data or even use them to attack other AI models(Wu et al.,2024).

https://www.sciencedirect.com/science/article/pii/S2949715923000707

Social risk is an indication of the possibility of adverse judgment or disapproval from others due to the use of ChatGPT. Some users may be afraid that relying on AI would be  viewed  as  laziness  or  incompetence,  which  can  harm  one's reputation  among friends, colleagues, or teachers. In the beginning, ChatGPT was used to explain some difficult content or rephrase the written project reports, but soon, it was utilized to write the  entire  homework.  Such  misuse  immediately attracts  attentions from  the teachers and schools and it was soon identified as plagiarism. Another concern is the copyright from ChatGPT. With increasing number of people who use  ChatGPT to create original-like text content without citation, the copyright of the content created by ChatGPT becomes a serious concern. No one is responsible for the correctness and accuracy of the content. It becomes necessary to regulate the copyright of machine's generation, including both visual and textual content (Wu et al.,2024).

Psychological risk refers to a feeling of fear, worry, or mental discomfort that may be caused while or after interacting with the tool. This is specifically for novice users who  are  not  familiar  with AI  tools  or  are  not  comfortable  communicating  with non-human entities.

Time risk is the fear that using ChatGPT will be a waste of time or a time-waster. If people believe they would need to invest too much time in learning to use the tool, fixing its mistakes, or interpreting its output, they will consider the tool a liability instead of an asset.

Along with these dimensions of perceived risk, differences among users in attributes like  gender and age will affect perceptions  of  these risks. Various  demographic groups will  differ in the extent of their technology  experience, attitudes  toward innovation, and tolerance for uncertainty, which affect how they perceive and react to perceived risks.

Briefly, the perception of different types of risks by users in utilizing ChatGPT is the path to foretelling their intention to reuse the tool. Risk perception in terms of gender and age will be examined in this paper, the interrelationships among the risk factors themselves will be explored, and the influence of these risks on continuance intention will be investigated.

Methodology

3.1 Overview of Research Design

This study adopts the Quantitative Research Method and uses structured questionnaires as tools to explore the impact of multi-dimensional perceived risks on users' willingness to continuously use ChatGPT.

The questionnaire design is based on existing literature and theoretical models and contains a total of 21 variable dimensions. Covering Performance Risk, Privacy Risk, Social Risk, Psychological Risk and Time risk Key constructs such as Risk, willingness to use, trust, self-efficacy, and satisfaction. All items were measured using the 5-point Likert Scale, ranging from "strongly disagree" (1) to "strongly agree" (5).

3.2 Questionnaire design and variable classification

3.2.1 The dimension division is clear and the structure is reasonable

The questionnaire contains a total of 21 variable dimensions, covering:

• Determinants of willingness to use (such as risk perception, trust, and satisfaction)

• Behavioral and attitude variables (such as usage habits, self-efficacy, and promoting conditions)

• User background information and AI usage experience

Each dimension contains five items and uses the Likert five-point scale (1 = strongly disagree, 5 = strongly agree) to ensure the standardization and comparability of the data and enhance the construct validity.

3.2.2 Data Quality Control Mechanism

To ensure the authenticity and validity of the recovered data, the following mechanisms are set up for the questionnaire:

• Anonymous filling to protect users' personal information;

• Logical consistency check: Screen out obviously conflicting responses;

• Duration filtering: Exclude abnormal samples with a duration lower than the reasonable answering time;

• IP duplicate screening mechanism: To prevent the same user from answering questions repeatedly;

3.3 Data collection method

The questionnaires were distributed using Google Forms and at least 120 valid questionnaires were required to be collected. Participants must meet the basic requirements (having used ChatGPT at least once). The questionnaire was conveniently sampled through methods such as social media and peer recommendations. The data were exported as Excel tables after recovery and transferred into SPSS for statistical analysis.

3.4 Sampling Design

The convenient sampling method was adopted and data collection was conducted using Google Forms. The target sample size is 123 people.

Age group classification:

Young Adults: 15-24 years old

Adults: 25-34 years old

Veterans: 35 years old and above

Education Level:

High school or below

Bachelor’s degree

Master’s degree or higher

3.5 Hypothesis development and variable testing

This study develops multiple research hypotheses based on different dimensions, such as:

• H 1: There are significant differences in performance risk perception between men and women (t-test)

• H2: There are significant differences in privacy risk perception among different age groups (one-way Analysis of Variance ANOVA)

• H3: Significantly negative prediction of perceived performance risk for willingness to continue using (linear regression)

Independent Variables:

Perceived Risks:

Performance Risk (5 items): Concerns about ChatGPT's effectiveness

Privacy Risk (6 items): Worries about data security

Social Risk (5 items): Fear of social disapproval

Psychological Risk (5 items): Emotional discomfort

Time Risk (5 items): Concerns about inefficiency

Adoption Factors:

Perceived Usefulness (5 items): Belief in utility

Perceived Ease of Use (6 items): Comfort with interface

Trust (5 items): Confidence in reliability

Self-Efficacy (5 items): Confidence in independent use

Dependent Variables:

Intention to Continue Use (5 items): Likelihood of future usage

User Satisfaction (5 items): Overall experience evaluation

3.6 Data analysis tools

This study used IBM SPSS Statistics software for data encoding and statistical tests. The main analyses included:

Descriptive statistics (mean, standard deviation)

• Reliability Analysis (Cronbach's Alpha)

• Independent sample t-test

• One-way Analysis ofVariance (ANOVA)

• Multiple Linear Regression analysis

Hypotheses

Based on the research objectives and prior literature on perceived risks and technology adoption, the following hypotheses are proposed to guide the analysis of users’ perception of ChatGPT and their intention to continue using it:

4.1 Gender Differences and Perceived Risks

H 1a: There is a significant difference between male and female users in terms of perceived performance risk when using ChatGPT.

H 1b: There is a significant difference between male and female users in terms of perceived privacy risk.

H 1c: There is a significant difference between male and female users in terms of perceived social risk.

H 1d: There is a significant difference between male and female users in terms of perceived psychological risk.

H 1e: There is a significant difference between male and female users in terms of perceived time risk.

4.2 Age Group Differences and Perceived Risks

H2a: There are significant differences in performance risk perception among different age groups.

H2b: There are significant differences in privacy risk perception among different age groups.

H2c: There are significant differences in social risk perception among different age groups.

H2d: There are significant differences in psychological risk perception among different age groups.

H2e: There are significant differences in time risk perception among different age groups.

4.3 Correlations Among Perceived Risk Dimensions

H3: There are significant positive correlations among all five perceived risk dimensions (performance, privacy, social, psychological, and time risks).

4.4 Perceived Risks and Continuance Intention

H4a: Perceived performance risk has a significant impact on users ’ intention to continue using ChatGPT.

H4b: Perceived privacy risk has a significant impact on users’ intention to continue using ChatGPT.

H4c: Perceived social risk has a significant impact on users’ intention to continue using ChatGPT.

H4d: Perceived psychological risk has a significant impact on users ’ intention to continue using ChatGPT.

H4e: Perceived time risk has a significant impact on users ’ intention to continue using ChatGPT.