Lights, Camera, Deception? Sides of Generative AI | IGF 2023 WS #57

Knowledge Graph of Debate

Session report

Deepali Liberhan

META, a leading technology company, has effectively employed AI technology to proactively remove potentially harmful and non-compliant content even before it is reported. Their advanced AI infrastructure has resulted in a significant reduction of hate speech prevalence by almost 60% over the past two years. This demonstrates their commitment to maintaining a safe and responsible online environment.

In line with their dedication to responsible AI development, META has made a pledge to follow several essential principles. These principles include prioritising security, privacy, accountability, transparency, diversity, and robustness in constructing their AI technology. META’s adherence to these principles is instrumental in ensuring the ethical and sustainable use of AI.

Furthermore, META has demonstrated its commitment to transparency and openness by publishing content moderation actions and enforcing community standards. They have also taken additional steps to foster transparency by open-sourcing their large language model, Lama 2. By doing so, META allows for external scrutiny and encourages collaboration with the broader AI community.

The company prioritises fairness in its AI technology by using diverse datasets for training and ensuring that the technology does not perpetuate biases or discriminate against any particular group. This dedication to fairness underscores the company’s commitment to inclusivity and combating inequalities.

META’s approach to AI development extends beyond internal measures. They have incorporated principles from reputable organisations such as the Organisation for Economic Co-operation and Development (OECD) and the European Union into their work on AI. By aligning with international standards, META demonstrates its commitment to upholding ethical practices and participating in global efforts to responsibly regulate AI.

Recognising the importance of language inclusivity, META emphasises the need to make AI resources and education available in multiple languages. This is particularly crucial in a diverse country like India, where there are 22 official languages. META aims to ensure that individuals from various linguistic backgrounds have equal access to AI resources, ultimately contributing to reduced inequalities in digital literacy and technology adoption.

META values local partnerships in the development of inclusive and diverse AI resources. They acknowledge the importance of collaborating with local stakeholders for a deeper understanding of cultural nuances and community needs. Engaging these local partners not only enriches the AI development process but also fosters a sense of ownership and inclusion within the communities.

In terms of content moderation, META’s community standards apply to both organic content and generative AI content. The company does not differentiate between the two when determining what can be shared on their platforms. This policy ensures consistency in maintaining the integrity of the online space and avoiding the spread of harmful or misleading content.

Prior to launching new services, META conducts extensive stress tests in collaboration with internal and external partners through red teaming exercises. This rigorous testing process helps identify and address potential vulnerabilities, ensuring the delivery of robust and trustworthy AI services to users.

User feedback is highly valued by META. They incorporate a feedback loop into their product development process, allowing users to provide input and suggestions. This user-centric approach enables continuous improvement and ensures that the technology meets the evolving needs and expectations of the users.

To combat the spread of false information and manipulated media, META has established specific policies and guidelines. They have a manipulated media policy that addresses the sharing of false content, with an exemption for parody and satire. This policy aims to promote accurate and trustworthy information dissemination while allowing for creative expression.

In terms of community safety and education, META actively consults with experts, civil rights organizations, and government stakeholders. By seeking external perspectives and working collaboratively with these entities, META ensures that their policies and practices align with an inclusive and safe online environment.

Deepali Liberhan, an advocate in the field of AI, supports the idea of international cooperation and having clear directives for those involved in generative AI. She emphasises the importance of provenance, watermarking, transparency, and providing education as essential aspects of responsible AI development. Her support further highlights the significance of establishing international partnerships and frameworks to address the ethical and regulatory aspects of AI technology.

In the development of tools and services for young people, META recognises the importance of engaging them as significant stakeholders. They actively consult young people, parents, and experts from over 10 countries when creating tools such as parental supervision features for Facebook Messenger and Instagram. This inclusive approach ensures that the tools meet the needs and expectations of the target audience.

To enhance the accuracy of generative AI in mitigating misinformation, META incorporates safety practices such as stress testing and fine-tuning into their product development process. These practices contribute to the overall reliability and effectiveness of generative AI in combating misinformation and ensuring the delivery of accurate and trustworthy information.

Collaboration with fact-checking organisations is pivotal in debunking misinformation and disinformation. META recognises the importance of partnering with these organisations to employ their expertise and tools in combating the spread of false information. Such partnerships have proven to be effective in maintaining the integrity and trustworthiness of the online space.

Public education plays a vital role in ensuring adherence to community standards. META acknowledges the importance of raising awareness about reporting inappropriate content that violates these standards. By empowering users with the knowledge and tools necessary to identify and report violations, META contributes to a more responsible and accountable online community.

In conclusion, META’s use of AI technology in content moderation, responsible AI development, language inclusivity, local partnerships, and online safety highlights their commitment to creating a safe, inclusive, and transparent digital environment. By incorporating user feedback, adhering to international principles, and actively engaging with stakeholders, META demonstrates a dedication to ongoing improvement and collaboration in the field of AI.

Hiroki Habuka

Generative AI has diverse applications, but it also presents numerous risk scenarios. These include concerns about fairness, privacy, transparency, and accountability, which are similar to those encountered with traditional AI. However, the foundational and general-purpose nature of generative AI amplifies these risks. The potential risk scenarios associated with generative AI are almost limitless.

To effectively manage the challenges of generative AI, society must embrace and share the risks due to the uncertainty surrounding these emerging technologies. Developers and service providers cannot predict or expect all possible risk scenarios. It is crucial for citizens to consider how to coexist with this cutting-edge technology.

Technological solutions and international, multi-stakeholder conversations are of paramount importance to successfully address the challenges of generative AI. These discussions involve implementing solutions such as digital watermarks and improving traceability on a global scale. The increasing emphasis on multi-stakeholder conversations, which go beyond intergovernmental discussions, reflects the recognition of the importance of involving various stakeholders in managing generative AI risks.

A balanced approach is necessary to address the privacy and security risks associated with the advancement of generative AI technology. While efforts to enhance traceability and transparency are underway, they also give rise to additional privacy and security concerns. Therefore, striking a balance between various concerns is crucial.

Regulating generative AI requires collaboration among multiple stakeholders as the government’s understanding and accessibility to the technology are limited. Given the ethical implications, policy-making should also involve more than just governmental authorities. Multi-stakeholder collaboration is necessary to effectively understand and regulate the technology.

The ethical questions brought about by evolving technologies like generative AI require democratic decision-making processes. Privacy risks and achieving a balance between public risks can be managed through democratic practices. Therefore, democratic processes are essential when addressing the ethical complexities of newer technologies.

The proposal of Agile Governance suggests that regulations for generative AI should focus more on principles and outcomes rather than rigid rules. The iterative process of governance is essential for technology evolution, allowing for adaptability and refining regulations as the technology progresses.

Private initiatives play a significant role in providing practical guidelines and principles to ensure the ethical operation of generative AI. Given the limitations of government and industry in understanding and implementing such technologies, private companies and NGOs can contribute valuable insights and expertise.

Young people should be included in decision-making processes concerning generative AI. Their creativity and adeptness in using technology make them valuable contributors. Prohibiting the use of generative AI for study or education is not the answer. Instead, measures should focus on checks for misconduct or negative impacts. Regular assessments in the study field can help ensure responsible and ethical use of generative AI.

In conclusion, generative AI holds immense potential for diverse applications, but it also comes with countless risk scenarios. Society must embrace and share these risks, and technological solutions, as well as multi-stakeholder conversations, play a crucial role in managing the challenges associated with generative AI. A balanced approach to privacy and security concerns, multi-stakeholder collaboration, democratic decision-making processes, agile governance, and private initiatives are essential in regulating and harnessing the benefits of generative AI.

Audience

During the discussion, the speakers highlighted the critical need to address the negative impact of AI-based misinformation and disinformation, particularly in the context of elections and society at large. They noted that such misinformation campaigns have the potential to cause significant harm, and in some cases, even result in loss of life. This emphasised the urgency for action to combat this growing threat.

To effectively address these challenges, the speakers stressed the importance of developing comprehensive strategies that involve various stakeholders. These strategies should engage nations, civil society, industry, and academia in collaborative efforts to counter the spread of misinformation and disinformation. By working together, these sectors can pool their resources and expertise to develop innovative solutions and implement targeted interventions.

The panel also underscored the significance of raising awareness about these challenges on a large scale. By spreading awareness through education, individuals can be equipped with the necessary tools to identify and critically evaluate misinformation. Education plays a vital role in empowering people to navigate the digital landscape and make informed decisions, contributing to a more resilient and informed society.

In conclusion, the speakers argued that addressing AI-based misinformation and disinformation requires a multifaceted approach. Strategies involving nations, civil society, industry, and academia are necessary to counter these emerging challenges effectively. Additionally, spreading awareness about the threat of misinformation through education is crucial in empowering individuals and safeguarding societal integrity. By taking these steps, we can strive towards a more informed and resilient society.

Moderator

Generative AI has the potential to bring great benefits in various fields including enhancing productivity in agriculture and modelling climate change scenarios. However, there are concerns related to disinformation, misuse, and privacy. International collaboration is crucial in promoting ethical guidelines and harnessing the potential of generative AI for positive applications. Discussions on AI principles and regulations have already begun, and technological solutions can help mitigate risks. Finding a balance between privacy and transparency, addressing accessibility, education, and language diversity, and supporting AI innovators through regulations and funding are important steps towards responsible and equitable deployment of generative AI technologies.

Vallarie Wendy Yiega

An analysis of the provided information reveals several key points regarding generative AI and its impact. Firstly, it highlights the importance of young advocates understanding and analyzing AI. The youth often lack a comprehensive understanding of AI and its subsystems, indicating the need for increased awareness and education in this field.

Another significant finding is the need for generative AI to be accessible to diverse languages and communities. The localization of AI tools in different languages is crucial to ensure that marginalized communities, who do not identify with English as their main language, can fully benefit from these technologies. A concrete example is the BIRD AI tool available in Swahili, which has different responses than the English version, demonstrating the importance of localization and testing.

The analysis also emphasizes the necessity of youth involvement in policy development for AI. It acknowledges that AI is advancing at a rapid pace, outpacing the ability of existing regulations to keep up. Thus, it is crucial for young people to play an active role in shaping policies that govern AI technology.

Furthermore, the analysis uncovers the prevalence of copyright and intellectual property issues in generative AI. It highlights the importance of safeguards to protect authors’ intellectual property rights and prevent misuse of AI-generated content. Examples such as the use of digital watermarks to indicate AI-generated content versus human-generated content and the need for client consent in data handling discussions illustrate the issues at hand.

Another crucial finding is the need for multi-stakeholder engagement in the development of regulatory frameworks for AI. This approach involves collaborating with academia, the private sector, the technical community, government, civil society, and developers to strike a balance between promoting innovation and implementing necessary regulations to ensure safeguards.

Collaboration between countries is identified as a critical factor in the responsible use and enforcement of AI. Given that AI is often used in cross-border contexts, international cooperation is essential to establish unified regulations and enforcement mechanisms.

The analysis also addresses the importance of promoting innovation while adhering to safety principles concerning privacy and data protection. It argues that regulation should not stifle development and proposes a multi-stakeholder model for the development of efficient guidelines and regulations.

Moreover, it stresses the role of governments and the private sector in educating young people about generative AI. It argues that the government cannot work in isolation and should be actively involved in incorporating AI and generative AI into school curriculums. Additionally, the private sector, civil society, and online developers are encouraged to participate in educating young people about generative AI, going beyond the responsibility solely resting on the government.

The analysis provides noteworthy insights into the challenges faced by countries in regulating and enforcing AI technology. It highlights that despite being forward-leaning in technology, Kenya, for example, lacks specific laws on artificial intelligence. The formation of a task force in Kenya to review legislative frameworks regarding ICT reflects the country’s effort to respond to AI and future technologies.

In conclusion, the analysis underscores the urgent need for young advocates to understand and analyze AI. It emphasizes the importance of diverse language accessibility, youth engagement in policy development, safeguards for copyright and intellectual property, multi-stakeholder engagement in regulatory frameworks, international collaboration, and the promotion of innovation with privacy and data protection safeguards. It also emphasizes the roles of governments, the private sector, civil society, and online developers in educating young people about generative AI. These insights provide valuable guidance for policymakers, industry leaders, and education institutions in navigating the challenges and opportunities presented by generative AI.

Olga Kyryliuk

Generative AI technology offers numerous opportunities but also poses inherent risks that require careful handling and education. It is argued that generative AI cannot exist separately from human beings and thus requires a strong literacy component. This is especially important as generative AI is already widely used. The sentiment regarding this argument is neutral.

To effectively use generative AI, education and awareness are deemed necessary. Analytical thinking and critical approaches are crucial when analyzing the information generated by AI. It is suggested that initiatives should be implemented to teach these skills in schools and universities. This approach is seen as positive and contributes to SDG 4: Quality Education.

On the other hand, generative AI has the potential to replace a significant number of jobs. Statistics suggest that around 80% of jobs could be substituted by generative AI in the future. This perspective is viewed negatively, as it threatens SDG 8: Decent work and Economic Growth.

To address the potential job displacement, there is a call for reskilling and upskilling programs to prepare the workforce for the usage of generative AI tools. Microsoft and data.org are running such programs, promoting a positive sentiment and supporting SDG 8.

Efforts should also be united to raise awareness, promote literacy, and provide education around generative AI. The technology is accompanied by risks that require a collective approach involving proper understanding and education. This argument is viewed positively, aligning with SDG 4 and SDG 17: Partnerships for the goals.

Internet governance practices offer a collaborative approach towards finding solutions related to generative AI. Communication between different stakeholders, including the inclusion of civil society, is vital in these discussions. This perspective is regarded positively and supports SDG 17.

Stakeholder involvement, including creators, governments, and users, is seen as essential in shaping meaningful and functional policies concerning generative AI. Governments across the world are already attempting to regulate this technology, and users have the opportunity to report harmful content. The sentiment is positive towards involving a diverse range of stakeholders and aligns with SDG 9: Industry, Innovation, and Infrastructure and SDG 16: Peace, Justice, and Strong Institutions.

A balanced legal framework is deemed necessary to avoid hindering innovation while effectively regulating harmful content. It is acknowledged that legal norms regulating harmful content already exist, and overregulating the technology could impede innovation. This viewpoint maintains a neutral sentiment and supports SDG 9.

Furthermore, advocates for user education and awareness on concepts like deepfakes emphasize the importance of a better understanding and prevention. Users need to know how to differentiate real-time videos from deepfakes, and educational programs should be in place. This perspective is positively viewed, aligning with SDG 4.

In conclusion, generative AI offers great potential but also carries inherent risks. Education, awareness, stakeholder involvement, and a balanced legal framework are crucial in handling this technology effectively. Additionally, reskilling and upskilling are necessary to prepare the workforce for the adoption of generative AI tools.

Bernard J Mugendi

The analysis focuses on several key points related to the use of generative AI. One of the main arguments put forth is the importance of promoting remote access and affordability, especially in rural areas. It is highlighted that rural areas often struggle with internet connectivity, and the affordability of hardware and software platforms is a pressing challenge. This is particularly relevant in communities in East Africa and certain regions in Asia.

The speakers also emphasize the need for generative AI solutions to be human-centred and designed with the end user in mind. They give the example of an agricultural chatbot that failed due to a language barrier encountered by the farmers. Therefore, understanding the local context and considering the needs and preferences of the end users is crucial for the success of generative AI solutions.

Additionally, the analysis underscores the value of data sharing among stakeholders in driving value creation from generative AI. It is mentioned that the Digital Transformation Centre has been working on developing data use cases for different sectors. Data sharing is seen as fostering trust and encouraging solutions that can effectively address development challenges. An example of this is the Agricultural Sector Data Gateway in Kenya, which allows private sector access to various datasets.

The speakers also emphasize the importance of public-private partnerships in the development of generative AI solutions. They argue that both private and public sector partners possess their own sets of data, and mistrust can be an issue. Therefore, creating an environment that fosters trust among partners is crucial for data sharing and AI development.

Collaboration is deemed essential for generative AI to have a positive impact. The analysis highlights a multidisciplinary approach, where stakeholders from the public and private sectors come together. An example is given in the transport industry, where the public sector takes care of infrastructure, while the private sector focuses on product development.

Furthermore, there is a call for more localized research to understand the regional-specific cultural nuances. It is acknowledged that there is a gap in funding and a lack of engineers and data scientists in certain regions, making localized research vital for understanding specific needs and challenges.

The speakers also emphasize the importance of transparency in the use of generative AI. They mention an example called “Best Take Photography,” where AI generates multiple pictures that potentially misrepresent reality. To ensure ethical use and avoid misrepresentations, transparency is presented as crucial.

The need for more engineers and data scientists, as well as funding, in Sub-Saharan Africa is also highlighted. Efforts should be made to develop the capacity for these professionals, as they are essential for the advancement of generative AI in the region.

In addition to these points, public awareness sessions are deemed necessary to discuss the potential negative implications of generative AI. The example of “Best Take Photography” is used again, showing the risks of generative AI in creating false realities.

The analysis makes a compelling argument for government-led initiatives and funding for AI innovation, particularly in the startup space. The Startup Act in Tunisia is presented as an example of a government initiative that encourages innovation and supports young innovators in AI. It is argued that young people have the ideas, potential, and opportunity to solve societal challenges using AI, but they require resources and funding.

Lastly, the speakers highlight the potential risks of “black box” AI, where algorithms cannot adequately explain their decision-making processes. This opacity can lead to the spread of misinformation or disinformation, underscoring the need for transparency in how models make decisions.

Overall, the analysis provides valuable insights into the various aspects that need to be considered in the use of generative AI. It highlights the importance of addressing challenges such as remote access, affordability, human-centred design, data sharing, public-private partnerships, collaboration, localized research, transparency, capacity development, public awareness, government initiatives, and the risks of black box AI. The emphasis on these points serves as a call for action in leveraging generative AI for positive impact while addressing potential pitfalls.

Speakers