The coded gaze is a concept borrowed from media studies, specifically from the notion of the male gaze. The male gaze refers to the portrayal of women through a man’s eye, emphasizing who is positioned as worthy and who has the power to decide what is worthy. In the context of technology, the coded gaze refers to the power dynamics and biases that shape the priorities and preferences of technology. It also highlights whose prejudices get embedded in technology due to the power dynamics present.

Dr. Joy’s Personal Experience with the Coded Gaze
Dr. Joy Blomwini shared a personal experience from her time as a graduate student at MIT. While working on a project involving face tracking, she wore a white mask to track her face. The system failed to detect her dark-skinned human face but successfully detected the mask. This incident made her realize the existence of the coded gaze, where something less than human was detected while her human face was not.

Industry Focus on Short-Term Risks:
Industry discussions prioritize addressing immediate risks and impacts of AI systems on society. Concerns include AI-powered misidentifications leading to false arrests, as seen in cases like Portia Woodruff and Robert Williams. Social justice and tech justice groups emphasize the urgency of addressing these harms.

Long-Term Existential Risks:
Some experts also consider the potential for catastrophic risks posed by advanced AI in the long term. Questions arise about ensuring human-centered outcomes and avoiding catastrophic risks as AI surpasses human intelligence. The balance between short-term and long-term risks remains a critical consideration.

Resource Allocation and Narratives:
Sam Altman highlights the impact of narratives on resource allocation in AI safety research. He observes that fear-driven narratives attract more funding, while research focused on immediate harms receives less investment. The narrative shapes the flow of resources, influencing the prioritization of research efforts.

Limited Productive Investment in Existential Risks:
Dr. Joy acknowledges the scarcity of productive investments in existential risks associated with AI. He emphasizes the need for more effective investment strategies to address these long-term concerns.

Balancing Act and Holding Space for the Future:
Sam Altman stresses the importance of considering both present and future risks, avoiding exclusive focus on either short-term or long-term issues. He advocates for holding space for thinking about the future while addressing current concerns.

AI in Military Applications and Structural Violence:
Altman shares his personal journey from avoiding human interactions to engaging with AI’s implications in military applications. He discusses the risks of lethal autonomous weapons and the need to prevent automated systems from making kill decisions. Altman expands the concept of X risks to include structural violence, such as lack of healthcare and opportunity, which can diminish the quality of life.

What is AGI?:
AGI is a term used to describe highly intelligent AI systems that can perform tasks that currently require human intelligence. There is no consensus on the exact definition of AGI, but some definitions include the ability to learn, reason, and solve problems in a general way, across different domains.

Benefits of AGI:
AGI has the potential to revolutionize many industries and aspects of our lives. For example, AGI could be used to develop new medical treatments, create more efficient transportation systems, and help us better understand the universe.

Safety of AGI:
The development of AGI also raises important safety concerns. It is essential to ensure that AGI systems are safe and beneficial for humanity. This means developing AGI systems that are aligned with human values and that can be controlled and held accountable.

Major Trends of the Decade:
Abundance and capability of intelligence will increase exponentially. Energy prices will continue to fall, leading to increased quality of life.

Benefits of AI and Energy:
Better healthcare and education for everyone. Cures for diseases, reducing the need for healthcare. Increased happiness, entertainment, and connections with others.

Safety and Ethics in AI:
The people most impacted by technology should have the most say in its use and rules. There is a need for global conversations about AI safety and ethics. AI can learn an individual’s value preferences, allowing for personalized input on rules and values.

AI’s Fallibility and Trust Issues:
The discussion begins with a recognition of AI’s imperfections, including errors like hallucinations and biases. This highlights the inherent challenges in relying on AI systems like GPT-4 for self-regulation and the need for human oversight in setting rules and standards.

AI Safety and Alignment:
Sam Altman discusses the evolving concepts of AI safety and alignment, emphasizing the importance of understanding AI’s potential for harm and discrimination. This includes a focus on AI’s entire lifecycle – from design and development to deployment – and the prejudices and biases that may be embedded at each stage.

AI Harms Taxonomy:
Altman underscores the significance of a comprehensive approach to AI safety, addressing potential harms at all stages of AI development. This approach involves recognizing biases in design, ensuring ethical data sourcing and usage, and anticipating deployment challenges.

Redress for AI Harms:
A critical aspect discussed is the need for mechanisms to address harm caused by AI systems. This includes considering the impact on those already harmed, such as artists and creatives, and developing processes for redress and accountability.

User Involvement and Adaptation:
The conversation highlights the importance of involving those affected by AI in its development and deployment processes. This approach involves continually adapting AI systems based on user feedback, as demonstrated in the evolution from Dolly 1 to Dolly 3.

Addressing Bias and Safety in GPT-4:
Specific attention is given to the safety measures and bias mitigation in the development of GPT-4. The process involved extensive safety testing and adjustments based on feedback from various stakeholders, including critics.

Artists’ Rights and AI Training Data:
The dialogue touches upon the concerns of artists regarding the use of their work in AI training data. The discussion explores the balance between benefiting from AI advancements and respecting the rights of creators, including the option for artists to opt out of having their styles replicated.

Consent, Compensation, Control, and Credit:
Finally, the conversation addresses the broader implications of AI on the creative economy. This includes considerations of consent, compensation, control, and credit for creators whose work contributes to AI systems, illustrated by the example of generative AI’s impact on voice actors and authors.

AI’s Potential for Misuse:
The speaker emphasized that not all technology should be released due to potential misuse. Voice cloning, in particular, is a highly personal feature that could lead to negative consequences. Companies need to consider the risks and implement safeguards before releasing such technologies.

AI’s Role in Amplifying Polarization:
AI algorithms can exploit emotions and amplify misinformation, leading to increased polarization. The speaker coined the term “algorithms of distortion” to describe this phenomenon.

Examples of Algorithmic Distortion:
A Bloomberg News evaluation of the text-to-image generator Stable Diffusion showed that it generated lighter-skinned men for high-paying jobs and darker-skinned men for low-paying jobs, perpetuating stereotypes. In the same evaluation, women were represented as architects only 3% of the time, despite making up 34% of architects in the US.

Concern for Future Generative AI:
The speaker expressed concern that future generative AI models, which could involve multimodal capabilities, might regress on diversity and inclusion. These algorithms, intended to take us into the future, could instead lead to a setback in the progress made towards equality.

AI Representation and Aspiration:
The discussion delves into the dilemma of AI systems representing current realities versus aspirational goals. For instance, the low percentage of women CEOs in America poses a question: should AI reflect this reality, or should it portray a more diverse and aspirational image? The consensus leans towards aspiration, advocating for AI to not only mirror the status quo but to inspire better societal norms.

General Purpose Models vs. Bespoke Systems:
The debate extends to the efficacy of general-purpose AI models like GPT-4 in addressing these representation issues. The speakers discuss the benefits of these models but also recognize their limitations. Bespoke, more specific models might be needed for certain applications, especially those requiring nuanced understanding of diversity and representation.

Gender Representation in AI:
An example cited involves kindergarten teachers, predominantly women, where an attempt to portray gender balance in AI-generated content led to concerns about erasing the dominant gender’s presence. This underscores the complexity of achieving aspirational representation without marginalizing existing groups.

AI’s Role in Reducing Bias:
The speakers note that while AI, like GPT-4, has the potential to exhibit less implicit bias than humans, vigilance is essential to prevent these technologies from exacerbating existing biases. AI’s potential to contribute positively to societal issues is acknowledged, but with caution.

AI and Election Integrity:
A major concern discussed is AI’s impact on elections, specifically the ability of AI models to engage in customized, one-on-one persuasion. While deepfakes are recognized as a known threat, the more subtle influence of AI in shaping political opinions is seen as a newer, potentially more insidious challenge.

AI Detection and Regulation:
Questions arise about the feasibility of using AI systems to detect outputs from other AI systems, such as watermarking or pattern recognition. The speakers express skepticism about the effectiveness of these measures and emphasize the role of social media platforms in monitoring and regulating AI-influenced content.

Collaboration Among AI and Social Media Companies:
The conversation touches on the efforts of AI and social media companies in addressing these challenges. While there’s a general belief that companies will act responsibly, there is concern about foreign adversaries using AI for malicious purposes, particularly in spreading misinformation.

Content Credentialing and Misinformation:
The concept of content credentialing is introduced as a means to verify the source of information, distinguishing AI-generated content from trusted media sources. This approach is seen as a starting point in combating misinformation, especially with the increasing availability of generative AI tools.

AI Not Inevitable but Driven by Momentum and Narratives:
Sam Altman does not subscribe to technological determinism, emphasizing the role of human agency and inertia in shaping the development and adoption of AI.

The Pitfalls of AI Narratives:
Altman cites the example of a chatbot replacing human workers at the National Eating Disorder Association, highlighting the dangers of relying solely on AI capabilities without considering the potential negative consequences. He introduces the concept of the “apprentice gap” and the importance of gaining mastery and professional calluses through hands-on experience, especially in entry-level jobs or processes.

AI’s Success Contingent on Usefulness and Societal Benefits:
Both Altman and SPEAKER_03 agree that AI’s inevitability is not guaranteed but rather depends on its perceived usefulness and the benefits it offers to society. They acknowledge the existence of a “graveyard” of past technological revolutions that failed to gain traction due to lack of utility or societal acceptance.

The Role of Power Dynamics in Shaping AI Adoption:
Altman raises the issue of power dynamics in shaping the adoption of AI, particularly in cases where those harmed by AI systems may not even be aware of their existence or influence. He emphasizes the need for measures, laws, and legislation to ensure that AI tools are fit for purpose and that people are not used as guinea pigs in the pursuit of technological advancement.

The Importance of Laws and Regulations:
Altman advocates for the establishment of laws and regulations that compel companies and government agencies to take specific actions regarding AI usage. He highlights the need for mechanisms that would require AI systems to pass certain explainability thresholds and address concerns about algorithmic discrimination.

Balancing the Benefits and Risks of AI:
SPEAKER_03 acknowledges the potential benefits of AI, especially for those who previously lacked access to such tools. They emphasize the challenge of maximizing the positive applications of AI while minimizing its negative consequences. They view this as a recurring theme in the history of technology and society’s interaction with new innovations.

Safety and Explainability:
Many questions surround the safety and explainability of AI. The AI Bill of Rights from the White House offers a grounding for responsible AI, emphasizing safe and effective systems, protection from discrimination, privacy, consent, and alternatives/fallbacks.

Alternatives to Problematic AI Implementations:
The Algorithmic Justice League’s work on the IRS’s adoption of ID.me, a facial recognition vendor, highlights the need for concrete alternatives. Alternatives to biometrics include using post offices and employing humans to verify identities, providing jobs and a human-centered approach.

Representing Voices Impacted by AI:
Those most affected by AI are often not involved in designing the systems. Companies have a responsibility to ensure global input, representing different countries, communities, and socioeconomic strata.

Challenges in Global Input and Representation:
Gathering fair and just input from diverse groups is a challenge. The idea of using a chatbot to gather global input from users has limitations, such as tyranny of the majority and varying levels of access and interest in the technology.

Company Responsibility:
Companies must create systems that address the negative impacts of AI and synthetic media. They have a role in mitigating harmful outcomes and preventing misuse.

Government Intervention:
Governments should establish regulations to protect the public interest. Penalties for harmful system use could encourage caution and responsible design. Legal ramifications for violations can incentivize ethical AI practices.

Executive Order’s Impact:
The recent executive order, though non-binding, links federal funding to compliance. Agencies using AI systems must undergo specific checks, ensuring accountability.

Volunteer Commitments:
While voluntary commitments from companies are valuable, legal protection is crucial. G7’s voluntary commitments are a positive step but require government oversight.

Regulation and Trust:
Government regulation is essential to establish rules for AI and synthetic media use. Companies should make voluntary commitments but not be solely relied upon. Lessons learned from the previous tech cycle should inform current regulations.

Public’s Role:
The public should remain vigilant and not blindly trust companies in this domain. Collective action is necessary to ensure responsible AI practices and protect the public interest.

Hope in the Face of Fear:
Sam Altman emphasizes the importance of sharing stories of manipulation and exploitation to create awareness and vigilance against AI-related risks.

Recognizing the Problem:
Altman highlights the progress made in addressing AI safety concerns, such as the US executive order and the AI safety summit in the UK. He stresses the need for individuals to voice their concerns to shape the agenda and drive solutions.

Addressing Manipulation:
Altman suggests that naming the specific concerns, such as manipulation, allows for targeted and effective action to address them.

Resilience in the Face of Challenges:
[SPEAKER_03] expresses optimism in humanity’s ability to adapt and overcome challenges posed by new technologies. They cite historical examples of daunting technologies, such as atomic weapons, recombinant DNA, and social media, where society has ultimately found ways to address the risks and reap the benefits. [SPEAKER_03] emphasizes humanity’s collective wisdom, resilience, and adaptability in facing and overcoming significant challenges.

AI and the Economy:
AI systems force discipline and encourage us to set up routines to keep our minds sharp. Future generations will be smarter and more capable due to their familiarity with AI tools. Humans will continue to use AI to create better things for themselves and others.

AI and Human Intelligence:
AI is a tool, not a creature, and we should use it to do more and create better things. Humans are good at knowing what other humans want, so we can use AI to create for each other. AI’s capabilities will increase expectations, leading to a desire for more from each other. The possibility of AI automating everything and humans becoming obsolete is unlikely.

AI’s Impact on Chess:
When AI beat humans at chess, it was thought that chess would become obsolete. However, chess has never been more popular, both to play and watch. Humans are wired to care about other humans, so watching two AIs play each other is not as engaging.

Conclusion:
AI will change jobs and allow us to do new things, but it will not replace humans. AI is a tool that we can use to make life better and better.

AI and the Digital Divide:
Sam Altman discusses the concept of the “digital divide” in the context of AI, emphasizing that billions of people still lack access to the internet. He also introduces the notion of a “digital chasm,” where even those with access to AI may face disparities in terms of education and opportunities.

AI Natives and the MoMA Mural:
Altman shares his experience at MoMA, where he observed babies crawling in front of an unsupervised AI moving mural. This prompts him to consider the idea of “AI natives” who are growing up in an environment where AI is ubiquitous.

Trajectories and the AI Chasm:
Altman contemplates the different trajectories of children whose parents expose them to AI versus those whose schools may restrict access due to fear. He emphasizes the importance of addressing this chasm to ensure equitable access to AI education and opportunities.

Unstable Desire: A Poem by Dr. Joy:
Dr. Joy presents her poem titled “Unstable Desire,” which explores the dangers of unchecked AI development and the potential for it to be driven by fear and profit. She questions whether humanity is more than just neural nets and transformations of collected muses, urging for responsible AI development that augments human potential rather than replacing it.

Transforming the Atmosphere of Innovation:
Dr. Joy emphasizes the need for transformative solutions rather than temporary band-aids to address the challenges posed by AI. She warns against the allure of Android dreams and the nightmare schemes of vice, highlighting the importance of guiding AI development with care and consideration for human values.