Artificial Intelligence can impact and influence the decision-making process in almost every area. However, the problem of AI bias slows down the process by causing unfairness and giving unjust results. This bias is generated from the data used to train the AI. This paper will focus on the significant aspects of AI bias and see where it comes from, how it shows up, and what it leads to. Additionally, it will examine the use of transparency in controlling bias, which also guarantees accountability, and supports ethical AI development and deployment. By addressing these issues, experts can handle the complexities of AI while ensuring fairness, transparency, and accountability.
The Opaque Perpetuation of Bias
Why Transparency is Key in AI
Artificial intelligence (AI) has become a universal force, secretly impacting decisions across many sectors. However, it has a serious challenge i.e. the bias it creates and its impact on people’s decision making. This bias is mainly incorporated at the stage of AI training and due to the biased data, which is used for its training. This is where transparency becomes significant.1
Take for instance an algorithm whose function is to approve the loans and it is trained on historical data that shows biases against women clients. The algorithm might by default favor the applications that contain keywords that are generally used by men. This lack of transparency creates challenges in finding out and resolving biases. This also creates a discriminatory cycle.
This lack of transparency in AI algorithms is created due to its complex nature. Unlike traditional, rule-based systems, the latest AI, specifically deep learning models, works in a black-box manner2. We input data, and it produces an output, but the sophisticated decisionmaking process is still a mystery. This lack of explainability makes it difficult to highlight where and how biases might be growing.
Since AI is used in everything from loan approvals to judicial decisions to facial recognition to hiring people, its biased consequences can impact and influence everything. For example, investigative news site ProPublica found a criminal justice algorithm that formed the opinion and marked African American defendants as “high risk”, further creating racial inequalities in the justice system3.
Defining Fairness: A Complex Puzzle
Regarding AI identifying what constitutes fair is another major challenge since there is no universal criteria for it. But there are two major approaches that we consider while looking at it4:
Individual Fairness: In this, the focus is on treating similar individuals similarly. For example, in the case of two loan applications with similar financial status, an AI showing this approach will give the same loan approval to both of them.
Group Fairness: In this approach, we focus on equitable outcomes across different age groups. For example, in criminal cases making sure that minorities are not being discriminated against based on their race.
Identifying which approach to apply will depend on the circumstances of the individual case. It might be that in some cases, giving importance to individual fairness will be unfair to entire groups and vice versa. These challenges show the necessity for careful consideration when developing an AI algorithm.
Challenges of AI Influencers: Ensuring Transparency and Validity in Social Data Research
Introducing AI-generated influencers in the dynamic technical domain of social media influences creates problems for researchers and tech experts. These AI influencers often created using complex algorithms and natural language processing techniques, present fresh opportunities for brands to engage with audiences. However, these opportunities also bring a huge set of challenges that must be dealt with to make sure there is transparency, validity, and ethical usage of AI in influencer marketing5.
One of the basic challenges in researching AI influencers is the collection and analysis of social data. A number of books and other data show that social data research involves complexities regarding data collection methods, selection of platforms, and techniques to process the data. Researchers have to differentiate between content generated by humans and AI because of these AI influencers. This distinction is important to understanding the impact and reach of AI influencers accurately.
Selecting a platform for analysing AI-generated influencers is another major challenge. There are number of social media platforms with unique features developed as per different age groups, which can largely impact people’s decision-making. It also makes it difficult to figure out the authenticity of the content. Along with this collecting data from the platforms becomes hard due to the rules and regulations that restrict access to these platforms. To ensure the accuracy and authenticity the researchers has to handle these changes properly.
Furthermore, analyzing data in AI studies can be tricky due to subjective decisions in how the data is prepared and understood. When dealing with social data, especially if AI has generated some of it, researchers need to be extra mindful of biases and ethical concerns. On top of that, using AI to create content makes things even more complex since researchers have to deal with the opaque nature of these algorithms and how they might affect data accuracy and understanding.
Transparency is important for dealing with the issues related to AI influencers. If there is no transparency as to how the researcher conduct the research and report their findings, then it will be difficult to repeat the research and it might also impact the reliability of the research conducted. Therefore, it is important the researchers are clear about how they collected the data, chose the platforms, prepared data, and analyzed it. Also, when it comes to using AI in influencer marketing, there should be transparent and responsible practices to maintain trust and confidence from consumers.
The next big challenge is confirming the results of the research due to the volume of the content available on these platforms. Researchers are required to have strong validation methods to be sure that their analyses is accurate and dependable. This could mean checking automated coding methods against each other, having people code a sample dataset, or comparing results with known accurate data to truly understand how AI influencers perform.
Lastly, there are ethical concerns that become important when we are dealing with AI influencers. As AI becomes more and more advanced, concerns like authenticity, manipulation, and consent become important. Researchers have to think about the ethical impact of using AI-generated content in influencer marketing, especially related to transparency, privacy, and consent from users. Also, the risk of AI influencers bringing in biases and stereotypical thinking depicts as to why these rules and regulations become important in this growing area.
In conclusion, effectively handling challenges related to AI influencers requires a thorough approach based on transparency, accuracy, and ethical AI practices. Researchers while choosing platforms must consider their dynamics, data collection methods, and ethical considerations in influencer marketing. By addressing these issues directly, researchers can make sure that AI influencers uphold high standards of transparency and integrity in their social data research.
How to Ensure Transparency?
So, how can we make sure AI is transparent and can control biases? Here are some important steps6:
A. Inspect Training Data: It has been already mentioned that the basis of any AI algorithm comes from the data it is trained on. So the first step in controlling biases should be to thoroughly examine the data for any potential biases. For example, is the data underrepresenting or overrepresenting certain groups? Is the data showing historical biases toward any gender? By doing this we can ensure that these biases never enter the system in the first place.
B. Explainable AI Techniques: Experts are trying to develop techniques for explainable AI (XAI). These techniques are focused to make algorithm functioning more transparent and not to let them work in a black box manner, allowing us to understand how AI arrives at its decisions. This involves techniques like feature importance analysis, which idaentifies the factors that most significantly influence the AI’s output. By understanding these factors, we can find potential biases present within the model.
C. Human Oversight: It is acknowledged that AI has a lot of potential but it should not be allowed to operate without any human intervention. Human supervision is important because it allows us to run AI systems with human decision-makers for comparison and incorporating explainability techniques to understand inconsistencies. Ultimately, humans should have the final say in important decisions taken on the basis of the data gathered with the help of AI.
D. Transparency in Development and Deployment: Transparency should be present in the entire AI development lifecycle. This includes disclosing the data sources used for training, the algorithms employed, and the potential limitations of the system. Additionally, during deployment, users should be informed when they are interacting with an AI system and understand the potential for bias.
E. Learning from AI to Improve Humanity: Interestingly, we can even use AI to find out biases in human decision-making. By inspecting outputs given by AI and comparing them with human decisions, we can find out previously unnoticed biases present in human thought processes. This newfound awareness can then be used to develop fairer human practices.
F. A Collaborative Effort: The Path Forward: A coordinative approach is required to efficiently handle AI bias for that we need to bring together experts from various fields:
Computer Scientists: Developing fairer and more strong AI algorithms.
Lawyers: Establishing ethical frameworks for AI development and deployment.
Eth ethicists: Ensuring ethical considerations are prioritized throughout the AI lifecycle.
Social Scientists: Understanding the societal implications of AI and potential biases.
By collaborating in these fields, we can ensure that AI is developed and used responsibly, ultimately achieving a future where human and machine intelligence work together to make fair and equitable decisions. By ensuring transparency we can ensure progress without spreading societal inequalities. A transparent and unbiased AI can be a powerful tool for good, but achieving this requires ongoing research, responsible development practices, and a commitment to fairness. Only then can we truly harness the potential of AI for a more just and equitable society.
Legal Considerations of AI-Generated Social Influencers
AI photo-editing tools have changed influencer marketing by making it easier to enhance photos. But with this convenience comes with important legal issues for both influencers and brands. One big concern is being honest about using these tools to edit or generate photos. Non-transparency can lead to legal claims such as false advertising and deceiving consumers, which could have serious consequences. For example, in Norway, we now have to mark retouched ads to ease societal pressure from unrealistic images. France has stricter rules, like defining commercial influencers and banning some product promotions. France plans stricter rules, like defining commercial influencers and banning some product promotions. These changes show why brands must follow advertising and consumer protection laws to avoid harm to their reputation and legal trouble. Brands must carefully check influencers and be transparent to keep consumer trust and credibility.7
In India, laws about advertising and data privacy are important for controlling biased and unclear AI-generated influencer content.
For starters, India’s Consumer Protection Act, 2019, is like the FTC Deception Rule, banning dishonest advertising. Like the FTC rules, this Act says ads can’t mislead or trick people, and not revealing AI influencer use could break this rule. If the brand is deceiving the consumers about their content then, it could mean legal consequences for the advertiser or brand.
Additionally, India’s Advertising Standards Council of India (ASCI) offers guidelines for endorsements which makes it mandatory to do transparent disclosures. These agree with the FTC’s rules and are applicable to the AI influencers who promote products or services in India. In case of any neglect to reveal the use of AI technology in endorsements might violate ASCI’s guidelines which may ultimately lead to regulatory investigation.
Thirdly, we have data privacy laws in India for example India’s Digital Personal Data Protection Act, 2023, which is similar to GDPR and CCPA and gives stricter rules for collecting data for the purpose of training AI influencers. Making sure that organisations comply with these rules is important for them to not commit any default and earn penalties.
Furthermore, as the judiciary on a global level we have to consider AI regulations, India may also bring in transparency rules for AI influencers. These new regulations will focus to boosting transparency and accountability in AI-generated content, making sure consumers understand AI use and its effects.
In conclusion, Indian laws on advertising, data privacy, and emerging AI regulations are crucial in stopping biased and unclear AI-generated influencer content. Following these laws is vital for organizations to uphold ethics, safeguard consumer rights, and prevent legal issues linked to misleading ads and data privacy breaches.
Social Considerations of AI-Generated Social Influencers
Along with legal concerns using AI-powered photo-editing tools will also create social concerns, especially regarding authenticity and ethics. Authenticity is important for any social media platform as the users while connecting with others want real connections and relatable content. However, influencers and brand can seem unreal if they use a lot of AI editing which might damage their trust in influencer marketing.7 Also, these tools can set unrealistic beauty standards which will add to body image problems. Influencers who make honesty and authenticity their prioritize can build stronger bonds with their audience and reduce AI’s negative impact. By promoting genuine connections and ethical practices, influencers can navigate social media changes while staying honest and credible. Ultimately, using AI in influencer marketing needs a careful balance between creativity and ethics to keep online interactions trustworthy and authentic.
Case Study
A case study on navigation systems and emergent behaviors8 will show the impact of AIgenerated social influencers on the urban environment. Apps like Google Maps, Waze, and Apple Maps, although unintentionally but created a lot of congestion by rerouting drivers to narrow street navigating in towns like Leonia, New Jersey. The focus of these real-time navigation systems is to get each person to their required place as quickly as possible but they sometimes neglect the affects it might have on the whole city. This can lead to more traffic, pollution, and separation between different parts of the city.
A study in Milan, Italy, looked into this by comparing how cars acted when using navigation apps versus randomly choosing routes. It found that following app directions without thinking caused more pollution because of more traffic. On the other hand, picking less crowded routes helped cut down on travel time and pollution. But making changes to the app’s directions might make some areas even more separated.
Despite evidence of navigation apps’ limitations, people tend to follow them blindly, overlooking the complex interplay between individual satisfaction and collective well-being in urban traffic systems. To address this, there’s a need for better understanding of individual routing choices’ impact on the urban environment and designing platform architectures that promote better collective outcomes. Prioritizing collective goals, such as minimizing CO2 emissions, transparently and acceptably, could lead to more efficient urban traffic management.
Besides the projects mentioned earlier, there are more examples showing how AI is used in different ways, and how important it is to handle biases and be clear about what AI does:
- Microsoft’s 2030 Vision on Healthcare, Artificial Intelligence, Data, and Ethics9: Microsoft wants to use technology, like cloud computing and AI, to make people healthier. This highlights the need for responsible data and ethical tool design. This initiative aims to promote trust in digital health-related products and services, emphasizing the ethical considerations inherent in leveraging AI for healthcare advancement.
- Sony’s Neural Network Libraries in Open Source10: Sony made its Neural Network Libraries free for anyone to use and improve, which helps create better AI programs. This move supports innovation, encourages teamwork, and makes AI development more open and transparent.
- Intel’s AI for Cardiology Treatment11: Intel is working with the Curie Institute to use powerful computers and AI to understand cancer genes better. By using AI to analyze large amounts of genetic data, doctors can find important gene changes faster. This shows how AI can help personalize cancer treatment.
- MSD’s AI for Healthcare Professionals12: MSD’s deployment of a chatbot for physicians in Italy demonstrates the practical application of AI in healthcare. By providing quick access to relevant information, MSD’s chatbot enhances the efficiency of healthcare professionals’ decision-making processes, showcasing AI’s role in supporting medical practitioners.
- Philips’ AI in Clinics and Hospitals13: Philips’ utilization of AI in digital pathology and patient monitoring illustrates how AI can streamline diagnostic workflows and improve patient care. By analyzing pathology data and detecting anomalies in vital signs, Philips’ AI solutions empower healthcare professionals to make informed decisions and deliver timely interventions.
- Canon’s Application of Automation in the Office Environment14: Canon’s digital mailroom solution exemplifies the integration of RPA technology to automate labor-intensive tasks in office environments. By removing mundane tasks and enhancing data capture processes, Canon’s solution improves operational efficiency while laying the groundwork for more intelligent automation systems.
Each of these initiatives highlights the importance of addressing AI bias and promoting transparency in AI applications across various domains, from healthcare and genomics to industry and conservation. By prioritizing ethical considerations and fostering collaboration, these efforts contribute to the responsible and equitable deployment of AI technologies.
Conclusion
The creation of Artificial Intelligence creates both negative and positive opportunities for us. AI has the capacity to enhance efficient and impact a person’s decision-making process but at the same time if not used properly it can create biases and transparency issues. Biases in the AI system comes from its training data, and it can lead to unfair and discriminatory outcomes, undermining trust and increasing social inequalities. However, by making transparency mandatory the stakeholders involved can reduce bias, promote accountability, and incorporate ethical standards. Lastly, to build trust in any online interaction it is important to look at the legal and social concerns of AI-generated social influencers. Ultimately, by following transparency measures and promoting responsible AI development practices, stakeholders can fully utilise the transformative potential of AI while safeguarding against its unintended consequences.
References
1. Xavier Ferrer, ‘Bias and Discrimination in AI: A Cross-Disciplinary Perspective’ (7 August 2021) Articles, Artificial Intelligence (AI), Ethics, Human Impacts, Magazine Articles, Social Implications of Technology, Societal Impact, IEEE Technology and Society.
2. Manyika, James, Jake Silberg, and Brittany Preston. “What Do We Do About the Biases in AI?” Harvard Business Review, October 25, 2019.
3. ProPublica, ‘Machine Bias: There’s software used across the country to predict future criminals. And it’s biased against blacks.’ by Julia Angwin, Jeff Larson, Surya Mattu, and Lauren Kirchner (ProPublica, 23 May 2016).
4. Maarten Buyl and Tijl De Bie, ‘Inherent Limitations of AI Fairness’ (18 January 2024), Communications ACM.
5. Stieglitz S, Mirbabaie M, Ross B, Neuberger C, ‘Social media analytics – Challenges in topic discovery, data collection, and data preparation’ (2018) 39 International Journal of Information Management 156-168.
6. Kostygina, G., Kim, Y., Seeskin, Z., LeClere, F., & Emery, S. (2023) Disclosure Standards for Social Media and Generative Artificial Intelligence Research: Toward Transparency and Replicability. Social Media + Society, 9(4).
7. Duke, Dylan. “The Ethical And Legal Considerations Of Influencer Marketing And AI Photo-Editing Tools,” Forbes, Small Business, Aug 3, 2023,09:15am EDT.
8. Pedreschi, D., Pappalardo, L., Baeza-Yates, R., Barabasi, A.-L., Dignum, F., Dignum, V., Eliassi-Rad, T., Giannotti, F., Kertesz, J., Knott, A., Ioannidis, Y., Lukowicz, P., Passarella, A., Pentland, A., Shawe-Taylor, J., & Vespignani, A. (2023). Social AI and the Challenges of the Human-AI Ecosystem.
9. Microsoft, ‘Healthcare – A 2030 Vision: Artificial Intelligence Data and Ethics, How Responsible Innovation Can Lead to a Healthier Society’ (December 2018).
10. Sony, Neural Network Console https://techhub.developer.sony.com/neural-networkconsole-windows-app
11. Intel, ‘Intel’s AI for Cardiology Treatment’, Solution Brief, Bogdan Georgescu, Giancula Paladini, Dmitry Rizshkov, Christian Schmidt, Dorisn Comaniciu.
12. Devpost, msd.ai < https://devpost.com/software/msd-ai>
13. Philips, Artificial Intelligence https://www.philips.com/a-w/about/artificialintelligence.html
14. Canon, Automation < https://global.canon/en/mfg/q-01.html>
Author: Tisha Sharma, Student, BML Munjal University
