Ph. D student at KAIST Graduate School of Science and Technology Policy
Ern Chern Khor
kec_1028@kaist.ac.kr
Following the public release of ChatGPT, which impressed many people with its communication and reasoning skills, there has been a surge in the availability of artificial intelligence (AI) models and related services, along with significant investments and collaborations in the industry. The technology has attracted massive attention and arised many discussions, including its potential implications and far-reaching consequences. Despite all the excitement, progress in this field was predictable for computer scientists. In 1998, Boden already discussed three ways in which AI could be creative, including combinations of existing ideas, exploring structured conceptual spaces, and transforming dimensions of existing structured spaces into new structures.1 While scientists can anticipate technological progress, foreseeing the social impacts of these applications on a large scale remains very challenging. Collins, the esteemed dictionary publisher, chose “AI” as the most noteworthy word of 2023.2 This recognition is attributed to the accelerated development of AI and how widely it has been discussed throughout the year. While this doesn’t imply that AI is objectively more significant than other advances such as renewable energy or genomics, the massive attention surrounding AI shows how wide-reaching its social impact can be. A comprehensive understanding of the potential positive and negative social impacts, including those from a gender perspective, resulting from the widespread use of generative AI is crucial for effective and informed policymaking.
It’s important to discuss the various branches within the study of AI.3 Among these, machine learning plays a key role in empowering AI to adapt it to new environments without specific guidance. There are many machine learning models, each designated to achieve specific goals. Recently, attention has been drawn towards generative AI, which is an AI built upon generative machine learning models. Additionally, discriminative AI, which will be briefly touched upon later, focuses on classification tasks and can be used to evaluate content generated by generative AI. Generative AI has the capability to generate content across diverse modalities such as text, audio, image, and video, at an exceptional speed and scale, offering countless applications. As of August 2023, the number of users of ChatGPT has reached 180.5 million.4 The widespread use of generative AI, with its substantial capital and numerous stakeholders, demand considerations on its implications on social justice. Who is using the technology and how? Who is receiving advantages from it, and who is not? These questions could be discussed from a gender perspective. It is a specific area of social justice that is widely addressed by many experts. This article will focus on how researchers can approach the impact of generative AI on the social status of women.
Gender Imbalance in Generative AI Development
To begin with, it is important to discuss gender issues within the development stages of generative AI, especially the development stages of products or services that are commercialized and widely used. Key questions that could be addressed by researchers with a specific focus on gender equality include:
- Who is involved in the development of generative AI?
- What are the power relations within the industry?
- How are the developers situated socially?
- Who are the decision makers?
- What is the training data of the AI?
- How transparent is the development process?
- Who is accountable for the system?
- Are there any policies regarding the production of generative AI, etc.
Gender inequality has persistently been an issue in STEM fields, but the issue becomes more serious in AI-related areas. According to a study by Mantha and Hudson, it is estimated that only 12% of AI researchers across 23 countries are women.5 The Global Gender Gap Report from the World Economic Forum revealed that only 26% of workers in the data and AI industry are female (figure 1).6 The systemic gender imbalance intensifies across the stages in AI research careers spanning from graduate studies to leadership positions, as women face challenges in getting their efforts visible in the male-dominated industry.7 We can easily picture that this trend in AI persists within the generative AI industry as well, but the implicit oppression is often hidden and challenging to quantify. For instance, Mira Murati, Chief Technology Officer (CTO) of OpenAI, has been extensively interviewed as a symbol of women empowerment in the industry. However, this may divert the attention from the actual representation of women throughout the development process of these AI services. Women leaders’ roles in the industry are often way too representative due to their limited presence. While they serve as influential role models for women or aspiring them to enter the field, decisions involving the changes in women leadership communities can significantly affect gender equality in the industry. For instance, the removal of existing female board members at OpenAI due to the internal changes has sparked discussion about the gender imbalance problem within the board and the broader industry.8
To examine the gender representation among generative AI developers, researchers often have to rely on data shared by generative AI providers, which can be restrained by the providers’ interests. An alternative data source can be the gender representation among university graduates or job applicants in fields related to generative AI. However, numbers alone may not offer comprehensive insights into the challenges faced by women and other gender minorities in the field. Qualitative methods, such as interviews, diaries, and ethnography, can provide a more in-depth exploration of gender inequality in the industry.10 The subjects of these studies include not only women currently working in the industry but also those who have left their jobs or failed to enter the industry despite having an interest.
These questions related to generative AI development extend existing inquiries into gender inequality in STEM fields. Generative AI is innovative but still not yet disruptive to alter the existing market structure as there are no significant changes in the key players in the industry. Major tech companies, including Google, Meta, and Microsoft, have their own generative AI products and also have the resources to acquire and invest in additional talents and products. Within the generative AI value network, most companies are established key players in the tech industry, such as Nvidia, Amazon, TSMC, etc. With the same major players but increased capital and power involved, the industry’s gender imbalance issues may become a more exacerbated version of an existing problem. Initiatives and interventions are necessary to address the gender imbalance issues within the generative AI industry.
Gender Disparities in Generative AI Use
Although generative AI may not have fundamentally disrupted existing industry, it undeniably has the potential to revolutionize various aspects of people’s lives, especially from the users’ side. An essential way for understanding the impact of generative AI on gender equality is to observe the use of it from the gender perspective. Since the use of generative AI is a subset of digital use, the theory of the digital divide becomes applicable for the discussion. This theory describes disparities across three levels of digital use.11
The first level refers to disparities in access. At this level, the discussion should center around whether a specific gender faces systemic restrictions in accessing generative AI compared to other genders. Similar to gender-science stereotypes, gender-technology stereotypes suggest that men are more likely to be associated with advanced technology use, encouraging them to be early adopters of emerging technologies compared to women.12 This stereotype can also influence the adoption of generative AI. Studies need to be conducted to explore and compare the use of generative AI among different genders, considering their use frequency and purposes. A good example is the first survey on the use of generative AI conducted by Deloitte, although lacking sex-disaggregated information.13
The second level of the digital divide refers to gaps in skills. At this level, the question revolves around whether a specific gender faces systemic restrictions in acquiring skills for using generative AI compared to other genders. Women may face more barriers in becoming advanced users of generative AI, not only due to existing disparities in resources but also because of socially constructed differences in attitudes toward learning to use the technology. To examine these gaps, researchers have to measure digital literacy and competence related to generative AI, focusing on comparing different genders. Existing AI literacy scales, like the one developed by Long and Magerko, can offer valuable references for the assessment.14
At the third level of the digital divide, focus shifts to the gap in the outcomes of digital use. The question to pose is whether a particular gender systemically gets more benefits or experiences more harms than other genders through the use of generative AI. For instance, generative AI has the potential to replace various jobs. A study has found that early adopters of generative AI, such as software developers, researchers, data scientists, and business analysts, frequently related to job disruption when discussing generative AI.15 Although the specific jobs that generative AI might eventually replace remain unclear due to its rapidly expanding applications beyond text generation, vulnerable social groups are likely at higher risk. A study by the Kenan Institute of Private Enterprise found that more women are at risk of job displacement due to generative AI compared to men, as the most affected occupations are predominantly held by women.16 As generative AI continues to evolve, further research is needed to explore potential gender gaps in the benefits or risks associated with job disruption, and how policy can help maximize benefits while minimizing risks. Researchers need to focus on assessing the level of job insecurity across different genders, taking into account the varied socio demographic backgrounds of users.
Gender Discrimination in Generative AI Contents
Deepfake technology is a subset of generative AI that creates images or even videos that look real but are entirely made up by putting one person’s face onto someone else’s body using deep learning techniques. The technology has already been known for being misused. Not only creating misinformation and fake news, the technology is exploited for the generation of pornographic content.17 A report from Deeptrace shows that 96% of deepfake contents are pornographic, with 100% of such content on pornography websites particularly targeting women.18 As this technology becomes more affordable, the generated content may become more widespread, posing the risk of normalizing the objectification of women and fostering digital insecurity for women.
<Figure 2> A Screenshot of an AI-generated image sharing platform19
The rise of generative AI models has made the creation of AI-generated content, especially images, much easier. These models mostly use stable diffusion, only requiring users to input natural language prompts for image generation. Various free models are available for users to create AI-generated images. These technologies can be very useful, but the challenge lies in how they are ultimately applied. On platforms for sharing AI-generated images, there is a significant presence of female images with amplified stereotypical feminine features. I will share some examples, but please note that some may find them uncomfortable.20, 21 While some platforms have policies against adult content, the adaptable nature of AI models allows users to independently utilize them for various purposes. Additionally, there is also a rise in online tutorials, teaching users how to apply these technologies for their own purposes. Contents that objectify women often attract the most attention in online spaces, making it challenging to control their spread. Addressing these challenges requires a comprehensive approach, involving legal frameworks and educational initiatives to promote responsible AI use and promote awareness of the creation of content that objectifies and exploits women.
Another aspect to discuss is how increased interactions with generative AI-powered bots can shape perceptions of gender identity. Through anthropomorphism, users may not view these bots as tools but recognize them as quasi-human entities engaging in social interactions, including attributing human-like qualities like gendered characteristics.22 This gives rise to a phenomenon where nonhuman entities, though lacking physical form, can occupy a social position and bring social influence. This new dimension can potentially challenge or reinforce existing gender perceptions. The real influence for future societies and generations remain uncertain. Yet, we can stay attuned to questions such as whether people project their gender perceptions onto generative AI, is generative AI perceived as genderless or genderfluid, how people’s perceptions of gender are shaped through conversations with generative AI, and whether changes in gender perception reinforce new forms of discrimination.
It’s About Policy Intervention
The discussion surrounding generative AI has primarily focused on potential issues through the lens of gender. However, a question arises: does generative AI only pose new challenges for gender issues, or does it hold the potential to serve as a solution to existing gender disparities? I would argue that the technology has significant potential to empower vulnerable groups including women. It can positively contribute to various spheres of life providing economic opportunities by enhancing productivity as well as cheap and easy alternative solutions to previously difficult tasks. Moreover, the rise of generative AI has popularized new ways for accessing information, such as chatbots, offering possible solutions to existing information gaps. Generative AI has the potential to empower women by providing assistance in acquiring the necessary resources to overcome real-life barriers and improve social positions. For example, women who want to start their business can utilize the technology to code, design, translate, etc. There are already some ChatGPTs available that are customized to help women with skill development and business growth.23, 24 Despite these possibilities, the adoption of the technology to empower women will not happen naturally because the existing digital divide can persist and make it harder for digitally vulnerable groups to learn the tools and utilize them. It is about policy intervention. Additional efforts are needed to ensure the implementation of these benefits. However, the challenges lie in the requirement for experts equipped with an understanding of both AI technology and social phenomena.
In pursuit of understanding the recent digital developments and engaging with experts from diverse sectors, I applied for the Global Youth Leader Program at the World Internet Conference 2023 under the recommendation of Prof. Kilnam Chon, a distinguished figure in Asian internet development and KAIST Emeritus Professor.25 I was honored to join another 11 fellow young leaders worldwide in the program. I attended the conference as a discussant in the Youth and Digital Future Forum, which allowed me to engage in a panel discussion alongside Prof. Chon and Prof. Xue Lan, the Dean of Schwarzman College at Tsinghua University, discussing the roles of youth in the digital future (figure 3). Prof. Chon emphasized the importance of youth to understand AI technology for a well-prepared future. While generative AI has captured widespread attention, he also advocated for a focus on discriminative AI as a key approach to address some challenges in generative AI. Discriminative AI, built upon discriminative machine learning models aimed at the tasks of classification, can serve as a crucial tool to evaluate content generated by generative AI. It can help detect gender bias and evaluate AI-generated content. More attention should be drawn to the development and implementation of discriminative AI to evaluate generative AI on a large scale. A cultural shift within the AI community is necessary to recognize discriminative AI as not just a complementary tool but as an important component in automating fairness.
In conclusion, it is important for stakeholders who are actively involved in shaping technology policy to be equipped with knowledge about the evolving branches of AI and the various machine learning models, such as generative and discriminative models. The rapid evolution of AI highlights the significance of being aware of the latest developments and applications in order to make well-informed policy decisions. On the other hand, an in-depth understanding of societal issues, phenomena, and challenges, such as gender inequality, is also evermore crucial. A significant challenge for technology policy is to manage the complexities surrounding both technology and society and their integration. It is nearly impossible to find a perfect approach, yet policies should play the roles to critically examine the social impacts of technology and proactively utilize technology to address existing social challenges. Examining generative AI through the lens of gender issues reveals a comprehensive picture with potential both for positive transformations and negative consequences. In essence, it is about policy interventions to shape the trajectory of the scale of influence generative AI can exert on gender issues to minimize unintended consequences and maximize the benefits of its widespread use.
읽을거리
Stuart J. Russell. (2019). Human Compatible. Penguin Books.
Computer scientist Stuart Russell is the author of the most widely used AI textbook “Artificial Intelligence: A Modern Approach.” While the book is accessible to audiences without technical knowledge of AI, it discusses deeply about the potential impact on human society in the development of artificial intelligence. In essence, it is enjoyable to read while provoking thoughts on critical issues regarding AI.
References
1 Boden, M. A. (1998), “Creativity and Artificial Intelligence”, Artificial Intelligence, Vol. 103, Iss. 1-2, pp. 347-356.
2「‘AI’ named most notable word of 2023 by Collins dictionary」, https://www.theguardian.com/technology/2023/nov/01/ai-named-most-notable-word-of-2023-by-collins-dictionary
3 Russell, S. J., & Norvig, P. (2010), Artificial Intelligence: A Modern Approach, London: Pearson.
4「Exclusive: ChatGPT traffic slips again for third month in a row」, https://www.reuters.com/technology/chatgpt-traffic-slips-again-third-month-row-2023-09-07/
5「Estimating the Gender Ratio of AI Researchers Around the World」, https://medium.com/element-ai-research-lab/estimating-the-gender-ratio-of-ai-researchers-around-the-world-81d2b8dbe9c3
6 World Economic Forum. (2020), Global Gender Gap Report 2020, World Economic Forum.
7 Ramos, G. (2022), Why We Must Act Now to Close the Digital Gender Gap in AI, World Economic Forum.
8「Prominent Women in Tech Say They Don’t Want to Join OpenAI’s All-Male Board」, https://www.wired.com/story/women-in-tech-openai-board/
9 World Economic Forum. op.cit., p. 37.
10 Chamberlain, K., & Hodgetts, D. (2018), “Collecting Qualitative Data With Hard-To-Reach Groups”, The SAGE Handbook of Qualitative Data Collection, pp. 668-685.
11 Ragnedda, M., & Ruiu, M. (2017), “Social Capital and the Three Levels of Digital Divide”, Theorizing Digital Divides, pp. 21-34
12 Pew Research Center. (2016), Early Technology Adopters, Pew Research Center.
13「More than four million people in the UK have used Generative AI for work – Deloitte」, https://www2.deloitte.com/uk/en/pages/press-releases/articles/more-than-four-million-people-in-the-uk-have-used-generative-ai-for-work-deloitte.html
14 Long, D., & Magerko, B. (2020), “What Is AI Literacy? Competencies and Design Considerations”, Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, pp. 1-16.
15 Haque, M. U., Dharmadasa, I., Sworna, Z. T., Rajapakse, R. N., & Ahmad, H. (2022), “‘I think this is the most disruptive technology’: Exploring Sentiments of ChatGPT Early Adopters using Twitter Data”, arXiv preprint arXiv:2212.05856.
16「Will Generative AI Disproportionately Affect the Jobs of Women?」, https://kenaninstitute.unc.edu/kenan-insight/will-generative-ai-disproportionately-affect-the-jobs-of-women/
17「Confronting the Surge of Deepfake Pornography and Digital Gender Violence」, https://medium.com/@riyaramani/confronting-the-surge-of-deepfake-pornography-and-digital-gender-violence-88a0ff769e34
18 Ajder, H., Patrini, G., Cavalli, F., & Cullen, L. (2019), The State of Deepfakes: Landscape, Threats, and Impact, Deeptrace.
19「PixAI – Anime AI Art Generator for Free」, https://pixai.art/
20 Ibid.
22 Damiano, L., & Dumouchel, P. (2018), “Anthropomorphism in Human–Robot Co-evolution”, Frontiers in Psychology, Vol. 9, pp. 468.
23 https://chat.openai.com/g/g-iWaZYfnjB-ai-women
24 https://chat.openai.com/g/g-2MqbzsLMy-9-5-women-in-business
25「2012 Inductee – Kilnam Chon」, https://www.internethalloffame.org/inductee/kilnam-chon/
댓글 남기기