PDF Summary:AI Snake Oil, by Arvind Narayanan and Sayash Kapoor

The proliferation of artificial intelligence (AI) has sparked excitement and promises of innovation. Yet, as Arvind Narayanan and Sayash Kapoor outline in AI Snake Oil, overzealous claims of AI's abilities distract from its innate limitations and ethical risks.

This book dissects the frequent shortcomings of predictive and generative AI applications, from reproducing societal biases to fueling misinformation and infringing on privacy. Narayanan and Kapoor examine the misalignment of corporations' drive for profits with responsible AI development, illuminating the need for robust transparency and accountability measures.

(continued)...

• The resources required for continuous autonomous examination could be substantial, and it may not be clear who should bear these costs.
• Overemphasis on safeguards might divert attention and resources away from other important aspects of AI development, such as improving the technology's capabilities or user experience.

The difficulties posed by the integration of artificial intelligence into content management and the broader implications for society.

This passage discusses the challenges involved in overseeing and controlling digital content on the internet by utilizing artificial intelligence. Narayanan and Kapoor argue that the complexities of online communication, along with the continuously changing terrain of societal conversation, inherently limit what artificial intelligence can achieve in monitoring conversations on the internet.

The challenges artificial intelligence faces in comprehending context and offering interpretations for moderation tasks.

Narayanan and Kapoor argue that the subtleties and context within online discussions often elude AI, leading to errors in content moderation. They narrate the incident of Mark, a father, who was denied access to his Google account after the company's automated systems erroneously flagged photographs of his son's rash-afflicted private areas, which were intended for a doctor's review, as child exploitation material. This example, along with other instances where an account faced suspension for disseminating an illustration featuring Captain America in combat with a Nazi, and the situation where the complete collection of video material from Cornell University was taken down because of nudity displayed in an educational art exhibition, illustrates how AI systems often make shallow judgments of content, overlooking important nuances.

Over time, the authors highlight that artificial intelligence is expected to improve its understanding of context, which will emphasize its successes in fields like recognizing humor, a realm that has traditionally presented obstacles for AI. Narayanan and Kapoor argue that AI could potentially equal human moderators in various settings, provided there is an abundance of data and computational resources. However, they also highlight situations where the system fails to fully understand, such as in cases involving discussions about actual events or when discerning the motive behind a photograph of a child depends on the perspective of the person who took the picture. The difficulty in overseeing and regulating digital material will persist as a considerable hurdle, despite progress in artificial intelligence technology.

Ai Moderation Often Misses Cultural Nuances, Causing Biased Enforcement

Narayanan and Kapoor argue that the development of AI systems, which is largely influenced by Western and English-speaking perspectives, fails to grasp the nuanced cultural elements of online exchanges, leading to biased and insufficient moderation practices in non-Western nations. The book explores the catastrophic event in which Facebook's spread of hate speech against the Rohingya in Myanmar resulted in a surge of violent content, despite the fact that the platform had been warned numerous times.

The book highlights the challenges Facebook faces with its reliance on machine translation tools, which frequently result in the misunderstanding of nuanced language, especially in sensitive areas like hate speech, and are constrained by a lack of comprehensive training data for many of the world's languages. Narayanan and Kapoor highlight the considerable shortcomings in the social media giant's understanding of regional subtleties, pointing out that the task of overseeing content in Myanmar was assigned to just one moderator who was proficient in Burmese. This pattern of cultural incompetence, according to the authors, is seen in many countries, with AI systems reflecting the biases of their development context and failing to account for varying cultural norms and sensitivities.

Context

• Companies often allocate fewer resources to non-Western regions, which can result in inadequate moderation capabilities. This includes fewer human moderators who understand local languages and cultural contexts.
• AI systems improve through feedback and iteration. If feedback mechanisms are primarily based on Western user interactions, the AI may not receive the necessary input to adapt to non-Western contexts effectively.
• Social media platforms, particularly Facebook, were used to spread misinformation and incite violence against the Rohingya. This included the dissemination of false news, inflammatory posts, and hate speech that exacerbated tensions and violence.
• Hate speech often involves subtle language cues, such as euphemisms or coded language, which can be difficult for AI to detect. This is compounded by the fact that what constitutes hate speech can vary significantly across different cultures and legal systems.
• Languages can have complex grammar, idioms, and cultural references that are difficult to capture and interpret accurately without extensive data and context.
• Facebook is a major platform in Myanmar, with millions of users, which means that a single moderator would be overwhelmed by the volume of content needing review.
• AI systems may not effectively handle idiomatic expressions, humor, or sarcasm that are specific to certain cultures, leading to misinterpretations and inappropriate content moderation.

AI faces challenges in distinguishing between beneficial and detrimental speech, as the surrounding circumstances are crucial.

The writers argue that the limited ability of artificial intelligence to understand nuances makes it challenging to set clear guidelines for moderating online material, particularly in situations that are politically delicate. They examine the predicament of digital platforms, which frequently face criticism for their choices to either remove or permit the distribution of messages linked to protests that might incite violence. Understanding whether certain content is appropriate often requires knowledge of the broader social context, a domain in which artificial intelligence usually finds challenging.

Narayanan and Kapoor explore the controversial action taken by a prominent social media platform when it removed the iconic photograph depicting a child fleeing from a napalm attack without any clothes. The image's historical importance and its role in conveying the horrors of conflict clashed with the regulations of the social media platform, which ban the exhibition of child nudity. The authors highlight this instance to emphasize the importance of public discourse in determining the boundaries for content regulation and establishing what is considered acceptable speech.

Practical Tips

• Start a journal where you record instances where context changed the meaning of a statement you heard or read each day. This practice will sharpen your awareness of the subtleties in communication that AI may not grasp. For instance, note down a joke that was funny in a casual setting but would be inappropriate in a formal one, and reflect on why the context made a difference.
• Engage in conversations with friends or family members from different generations or backgrounds about content they deem appropriate or inappropriate. Ask them to explain their reasoning and compare it to your own perceptions. This direct dialogue can provide real-life insights into how social context varies among individuals and can help you apply a more context-aware lens when evaluating content.
• Engage in mindful sharing by taking a moment to reflect on the potential impact of the images or content you post online. Before you share a photograph or a story, especially if it's sensitive or potentially controversial, consider writing a brief statement about why you believe it's important to share, and what you hope others will take away from it. This practice encourages responsible sharing and can foster a deeper understanding among your audience.
• Engage in a role-playing exercise with friends where one person acts as a historical figure and the other as a platform moderator. The historical figure presents an important moment from their life in the form of a photograph or story, while the moderator decides if it meets the platform's standards. This activity can deepen your understanding of the challenges in balancing historical preservation with community standards in a dynamic and interactive way.
• Create a digital feedback form to gather opinions from your social circle on various speech-related scenarios. This could range from hypothetical situations to real-life cases of content regulation. By reviewing the responses, you'll gain insight into the diverse perspectives that exist within your own community, which can inform your understanding of where people draw lines on speech and why.

Platforms frequently find that their quest for financial gain is at odds with the stringent moderation of online material.

Narayanan and Kapoor argue that social media companies' focus on profit-making undermines their content moderation efforts, leading to prioritizing user engagement over safety and a lack of accountability for insufficient content oversight.

Platforms prioritize user engagement, which frequently results in the widespread distribution of controversial content, compromising security in the process.

The authors argue that the drive for increased profits via more user engagement on social media platforms frequently results in the inadvertent dissemination of harmful and divisive content. Content recommendation algorithms are designed to keep users engaged by promoting endless scrolling and interaction, focusing on content that elicits strong emotional responses, regardless of the type of emotion. Narayanan and Kapoor emphasize that content skirting the edges of regulatory norms tends to attract more engagement, an idea Mark Zuckerberg has referred to as the natural inclination of content to elicit increased interaction.

The authors use TikTok as an example to demonstrate this idea. TikTok’s algorithm focuses on maximizing the time users spend watching videos, rewarding even passively engaging content like "digital rubbernecking." This approach encourages individuals to produce content that is sensational, divisive, and of lower quality, all with the intent of securing and keeping the attention of viewers. The authors dispute Zuckerberg's description of this amplification as "natural," arguing that the platform's design and algorithmic choices have a substantial impact on how users engage with each other online.

Other Perspectives

• Platforms may prioritize user engagement, but they also invest in safety and security measures to protect their users and maintain a reputable platform, which can sometimes reduce profits.
• The assertion that engagement leads to the spread of harmful content overlooks the role of user agency; individuals choose what to engage with and share, which means that not all engagement is driven by platform algorithms.
• Some platforms implement features like screen time reminders or "take a break" notifications to counteract endless scrolling, indicating a recognition of the issue and steps taken to mitigate it.
• The design of algorithms can be influenced by a variety of factors, including user feedback, regulatory requirements, and ethical considerations, which can lead to the promotion of more balanced or neutral content.
• Regulatory norms vary significantly across different regions and cultures, and content that is considered edgy in one context may be mainstream in another, suggesting that the relationship between regulatory boundaries and engagement is not universally consistent.
• The focus on maximizing watch time is a common practice across many video platforms and is not unique to TikTok, suggesting that the issue is industry-wide rather than platform-specific.
• TikTok's algorithm may not solely reward passively engaging content; it also promotes creativity and originality, which can lead to positive and educational content being shared widely.
• The production of sensational content can sometimes lead to increased awareness and engagement with critical social issues, which might otherwise receive little attention.

Determining who bears responsibility for the insufficient oversight of user-generated content is made more challenging by the opacity and lack of accountability in digital platforms.

Narayanan and Kapoor highlight the lack of transparency and accountability within the methods platforms employ to monitor and regulate the content shared by users. The details regarding the artificial intelligence systems used by companies, as well as the data that informs their development, are seldom made public, making it challenging to evaluate their effectiveness or identify any built-in biases.

Furthermore, the authors question the inadequate processes for contesting decisions rendered by digital platforms, emphasizing their personal experiences with unexplained account bans and the absence of a substantial method for contestation. Even though the officials had given their approval, Mark's predicament of being unable to access his account with the major technology company underscores a lack of accountability. The authors contend that a lack of clarity and responsibility, coupled with the use of inscrutable AI technologies, may foster doubt, especially when these technologies are used by platforms to make significant decisions that might result in account suspensions.

Other Perspectives

• Platforms might contend that they are constantly evolving their moderation practices in response to new challenges and that complete transparency would not accurately reflect the dynamic nature of their operations.
• Some platforms may claim that they do provide transparency reports and some level of insight into their content moderation practices, even if the full details of the AI systems are not disclosed.
• It could be pointed out that the volume of content and the number of decisions that need to be made daily by these platforms necessitate automated systems, which, while not perfect, provide a more efficient solution than human moderation alone.
• In some instances, account bans without clear explanations might be justified if the content in question violates terms of service in ways that are not immediately apparent to the user but are clear breaches to the trained eye of content moderators.
• Mark's inability to access his account, despite official approval, may not necessarily indicate a lack of accountability; it could be a result of a technical glitch or error that is being addressed.
• Platforms may maintain the opacity of their AI systems to protect proprietary technology and prevent bad actors from gaming the system, which is a legitimate business concern.

The issue of unsuitable material on the web cannot be distilled down to simple technological solutions.

Narayanan and Kapoor challenge the notion that the sole use of artificial intelligence is sufficient for effectively managing content moderation issues. They argue that the complex and ever-changing nature of online interactions, combined with the financial interests of those who run these platforms, makes content moderation a deeply social and political challenge that cannot be resolved by simple technical solutions.

The authors advocate for a nuanced and multifaceted approach that acknowledges the fact that AI does not represent a straightforward solution. They argue that platforms need to focus on developing better policies, increasing transparency, committing to understanding the complexities of different local contexts, and engaging in dialogue with both the platform's community and scholarly experts to find an equilibrium between safeguarding online spaces and upholding freedom of expression.

Practical Tips

• You can critically evaluate the content you consume by keeping a moderation journal. Whenever you encounter moderated content on social media or news platforms, jot down your initial reaction, what you think the moderation policy might be, and how financial interests could have influenced the decision. This practice will sharpen your awareness of moderation in action and its complexities.
• You can enhance your understanding of AI's limitations in content moderation by manually reviewing a small set of comments on your social media posts. By doing this, you'll get a sense of the nuances and context that AI might miss. For example, if you notice sarcasm or cultural references that could be misinterpreted by AI, you'll see firsthand where AI can falter.
• Organize a virtual book club or discussion group focused on the topic of platform policies with friends or online communities. Use each session to discuss a different platform's policies and brainstorm ideas for improvements. This collective effort not only broadens your understanding but can also lead to a comprehensive list of suggested policy enhancements that the group can submit to the platform for consideration.
• Develop a personal habit of reaching out to platform customer support to inquire about their transparency practices. This direct approach can help you understand the level of openness a platform has regarding its operations and data usage. You might ask about how they handle user data, what their content moderation policies are, or how they deal with government requests for information.
• Create a map of local issues by using social media to follow news outlets, influencers, and community leaders from different regions. Take notes on the unique challenges and successes in each area, and look for patterns or common themes. This activity will help you grasp the nuances of different local contexts, which can be valuable when evaluating how platforms cater to diverse user bases.
• Volunteer as a moderator for an online forum or community that you're passionate about. Use this role to practice balancing the protection of the community with the encouragement of free expression. Work with other moderators to develop fair guidelines, and engage with users to understand their views on what constitutes a safe yet open online environment.

Forces Shaping AI: Cultural, Economic, Institutional Impacts

This section delves into the broader social, economic, and organizational forces that shape the development and application of artificial intelligence, highlighting the importance of hyperbole, lack of transparency, and the overreliance on artificial intelligence to solve complex social problems.

The proliferation of false information is influenced by a society that passionately promotes progress within the realm of artificial intelligence.

The authors argue that the current trajectory in the field of AI research, particularly the focus on outperforming established standards and the strong ties to industry, creates a climate that overstates AI capabilities while obscuring its limitations.

Commercially-Driven AI Researchers Exaggerate Capabilities, Underreport Limitations

Narayanan and Kapoor emphasize the growing influence of corporate funding on the advancement of AI studies, with companies like OpenAI, Google, and Meta leading the charge in developing advanced AI technologies. The authors highlight that corporate funding can lead to a bias, as it often prioritizes the development of products for the market rather than ensuring the research is comprehensive and transparent. The authors emphasize how companies often distort their performance metrics, carefully select datasets, and downplay possible risks to attract investment and promote their products.

Unlike in established domains such as medicine, where the impact of corporate sponsorship on research outcomes has long been examined, Narayanan and Kapoor contend that this pattern stands out. The authors argue that individuals in the artificial intelligence sector must commit to higher levels of openness and responsibility, ensuring full revelation of any possible competing interests.

Other Perspectives

• Corporate funding can be subject to rigorous internal review and ethical standards that aim to ensure responsible development.
• The framing of "being at the forefront" could be misleading, as it implies a race or competition, whereas AI development is often a collaborative and cumulative process involving many different entities.
• Corporate funding often comes with the practical experience and industry knowledge that can guide AI research towards solving real-world problems effectively.
• Companies may have a vested interest in ensuring their products are safe and reliable, which can lead to comprehensive research in certain areas such as safety and ethics.
• It could be pointed out that performance metrics are often made available for peer review and scrutiny, which can serve as a check against any significant distortion.
• The selection of datasets could be influenced by practical considerations, such as the availability of data, rather than an intent to mislead or exaggerate capabilities.
• It could be pointed out that the perception of downplaying risks is subjective and that companies may believe they are providing a balanced view, with the focus on positive aspects being a standard practice in technology promotion.
• Some corporations actively support open research and collaborations with academic institutions, which can lead to a more balanced and scrutinized research environment.
• There could be legal and privacy concerns with full transparency, especially when dealing with sensitive data or intellectual property.
• In some cases, the disclosure of competing interests might not significantly alter the perception of the research if the audience lacks the expertise to understand the implications of these interests.

In academia, the focus on publishing and meeting certain standards frequently prioritizes results that attract attention over studies whose findings can be consistently reproduced.

Narayanan and Kapoor conduct a thorough analysis of the customary practices within the artificial intelligence discipline, emphasizing the focus on scholarly publication output and achieving unmatched results in standardized data assessments. Researchers frequently focus on producing results that attract attention, potentially sacrificing the thoroughness and reliability inherent in their scholarly work. The authors emphasize a considerable obstacle within the field of artificial intelligence research: numerous studies evade independent validation due to their failure to disclose data and code, alongside their reliance on proprietary AI systems that are not transparent and the employment of dubious methods.

Narayanan and Kapoor illuminate the conclusions of their research on the application of artificial intelligence to predict social unrest, highlighting a frequent error that overstates the accuracy of these predictions. This error, commonly known as data leakage, illustrates the tendency of researchers to unintentionally exaggerate the potential of artificial intelligence by evaluating models on the same dataset used for training. The writers call for a transformation in scholarly culture that emphasizes solid, reproducible studies, openness regarding data and methodologies, and prioritizes grasping the underlying tenets that control artificial intelligence mechanisms over simply chasing notable performance metrics.

Practical Tips

• Create a simple scoring system for research papers you read based on criteria important to you, such as transparency, data availability, and study design. Assign points for each criterion and tally them up to assess the overall quality of the research. This personal metric can guide you in determining which studies to trust and reference in your own work or discussions.
• You can start a blog to document and analyze AI breakthroughs, focusing on the impact rather than the hype. By doing this, you encourage a deeper understanding of AI developments. For example, after reading about a new AI achievement, write a post that explores how this technology could affect daily life, job markets, or ethical considerations, rather than just its novelty.
• Encourage independent validation by supporting crowdfunding campaigns for open-source AI projects. Look for initiatives on platforms like Kickstarter or GoFundMe that aim to replicate AI studies or create open-source versions of AI systems. By funding these projects, you contribute to a community-driven effort for transparency and validation in AI research.
• Educate yourself on the basics of AI and machine learning through free online resources. Understanding the fundamentals of how AI works can help you make informed decisions about the technology you interact with. Websites like Khan Academy or Coursera offer beginner courses that explain the principles of AI. With this knowledge, you can better assess the transparency of AI systems you encounter and choose to support those that align with open research practices.
• You can critically evaluate AI technology by researching its development process before using or endorsing it. Look for transparency in how the AI was trained, what data was used, and whether the developers have addressed potential biases. For example, if considering a new AI-driven health app, check if the company has published information on their data sources and training methods.
• Create a simple spreadsheet to monitor and evaluate the accuracy of different AI systems' predictions on social unrest over time. Include variables such as the date of prediction, the predicted event, the actual outcome, and any notable news that could have influenced the prediction. This hands-on approach will give you a clearer picture of how data leakage might affect AI accuracy and help you understand the complexities of AI-based forecasting.
• Engage with online platforms that crowdsource replication of studies, such as blogs or forums dedicated to your area of interest. Participate by sharing your attempts to replicate findings from articles you've read, discussing challenges, and offering solutions. This practice not only contributes to the culture of reproducibility but also connects you with a community of like-minded individuals who value this approach.
• Create a simple flowchart that outlines the decision-making process of a basic AI algorithm you interact with daily, like a recommendation system on a streaming service. By mapping out the steps the AI takes to make a decision, you'll gain insight into the principles of AI operation, which can be more enlightening than knowing how many correct recommendations it makes.

Organizations frequently lack the incentive to disclose the negative impacts stemming from the use of artificial intelligence.

The writers argue that companies engaged in developing and deploying AI technologies often lack the incentive to reveal potential risks, as acknowledging these dangers could negatively impact their profits.

Companies primarily utilize machine learning and algorithmic systems to cut costs and boost profits, frequently disregarding the societal impacts.

The authors stress that companies often utilize AI as a tool to cut costs and boost profits, without adequately considering its broader societal repercussions, including job displacement, the embedding of biases, and the amplification of harmful content. The authors argue that companies prioritize efficiency over principles of fairness, transparency, and accountability due to their focus on profit maximization. Numerous medical facilities adopted Epic's sepsis prediction system despite its unverified efficacy, eventually discovering through multiple analyses that the outcomes were not up to par. The approach of incentivizing hospitals with monetary rewards to adopt the model starkly illustrates the preference for corporate profits over patient health.

Narayanan and Kapoor argue that companies are driven to adopt AI technologies, which can be difficult to examine or question, in their quest to achieve greater efficiency. The authors argue that a lack of transparency hinders the identification of potential risks, the enforcement of corporate accountability, and the establishment of strong safeguards against misuse.

Other Perspectives

• The adoption of machine learning is sometimes part of a company's commitment to staying competitive in a technology-driven market, which can indirectly lead to job creation in new areas and sectors.
• Companies may argue that they are not disregarding societal impacts but rather are actively working to mitigate them through responsible AI initiatives and ethical guidelines.
• Job displacement due to AI can lead to a shift in the workforce where employees are upskilled or reskilled, thus contributing to a more technologically advanced society.
• Efficiency can sometimes align with fairness and transparency, as streamlined processes can lead to more consistent and unbiased decision-making.
• The sepsis prediction system could have been one of several tools used by medical facilities, and not the sole basis for clinical decisions regarding sepsis treatment.
• Incentivizing the adoption of AI could accelerate the integration of advanced technologies in healthcare, which, in the long run, could lead to industry-wide improvements in efficiency and patient care standards.
• The effectiveness of transparency is also dependent on the ability of stakeholders to understand and act on the information provided, which may require a level of technical expertise that is not universally available.

The absence of stringent oversight in the artificial intelligence sector, coupled with the safeguarding of confidential business information, can obscure its true effectiveness and possible risks.

Narayanan and Kapoor highlight how companies often shield their advancements in artificial intelligence from scrutiny by claiming them as proprietary trade secrets. The genuine efficacy and inherent risks of artificial intelligence often remain obscured due to the absence of clear transparency and the inadequacy of regulatory oversight across various sectors.

The writers argue that such a situation could result in significant choices being made by untrustworthy artificial intelligence technologies. They cite the example of COMPAS, a widely used criminal risk assessment tool that has been shown to be biased against Black defendants and no more accurate than simple rules based on age and prior convictions. Narayanan and Kapoor argue that the lack of transparency in the COMPAS system obstructs attempts to challenge its incorrect results or to hold its developers accountable for any harm that results.

Other Perspectives

• The protection of trade secrets is a well-established legal principle that encourages companies to invest in research and development, potentially leading to more advanced and beneficial AI technologies.
• The risks and efficacy of AI can also be assessed through performance outcomes and independent audits without necessarily compromising proprietary information.
• The term "untrustworthy" is subjective and can vary depending on the context; what is considered untrustworthy in one application may be deemed acceptable in another, depending on the stakes and the alternatives available.
• The assertion that COMPAS is no more accurate than simple rules based on age and prior convictions might overlook the nuances of the tool's risk assessments, which could potentially consider a wider range of factors beyond just age and criminal history.
• It could be argued that the responsibility for ensuring non-biased and accurate tools lies with the regulatory bodies, and they should establish standards for AI tools like COMPAS rather than relying on the developers for transparency.

The detrimental effects of artificial intelligence are exacerbated by systemic and prejudice-related elements.

This part highlights how human and organizational biases can intensify the detrimental effects of artificial intelligence. Narayanan and Kapoor argue that addressing these social issues is crucial for skillfully navigating the challenges presented by artificial intelligence.

Employing computational learning models as a stopgap for intricate problems frequently results in suboptimal results.

The authors contend that institutions, often struggling to operate effectively, are prone to adopting deceptive AI solutions as they hastily seek remedies for complex issues. The allure of potential efficiency improvements via artificial intelligence often results in the adoption of flawed solutions that fail to address the fundamental issue.

Narayanan and Kapoor analyze ShotSpotter, a tool widely adopted by law enforcement to reduce incidents involving firearms, which is engineered to identify occurrences of gunfire. Despite assertions of improved accuracy and expedited responses, ShotSpotter has not substantially diminished gun violence, leading to wrongful arrests. The authors argue that the adoption of ShotSpotter indicates a bias towards using technological solutions to combat gun violence, while neglecting the fundamental systemic factors that give rise to this type of violence in the first place.

Context

• Intricate problems often involve multiple variables and deep-rooted issues that are not easily quantifiable or predictable, making them challenging for computational models to address effectively.
• Limited resources and expertise within institutions can make it challenging to develop or evaluate alternative solutions, making off-the-shelf AI products more appealing.
• Organizations may prioritize technological solutions due to a belief that they are inherently more efficient or modern, often overlooking the complexity of human and social factors involved in the issues they aim to solve.
• ShotSpotter is an acoustic gunshot detection system that uses a network of microphones to detect, locate, and alert law enforcement to gunfire in real-time. It is designed to improve police response times and accuracy in identifying gunfire locations.
• These are underlying social, economic, and political conditions that contribute to problems like gun violence. Examples include poverty, lack of education, systemic racism, and inadequate mental health care.

Prominent individuals and shared convictions mold the intricate conversation about the capabilities and limitations inherent in artificial intelligence.

Narayanan and Kapoor argue that the proliferation of biases and misinformation by influential figures impedes the formation of a thorough understanding of AI's capabilities and limitations. Mainstream media and speculative fiction frequently portray artificial intelligence in a way that causes people to overestimate its abilities, linking it to the concept of robots possessing broad cognitive abilities akin to those of humans.

Arvind Narayanan and Sayash Kapoor critically examine "The Age of AI," authored by Henry Kissinger, Eric Schmidt, and Daniel Huttenlocher, questioning the often overstated excitement surrounding artificial intelligence. The book portrays artificial intelligence as a mysterious force capable of transforming human life, yet it overlooks the solid technical underpinnings that support AI. Narayanan and Kapoor argue that statements from influential figures amplify public concern and obscure the genuine hazards linked to artificial intelligence, simultaneously distracting from the urgent need to address existing harm.

Context

• AI has gone through cycles of hype and disappointment, known as "AI winters," where expectations were not met. Understanding this history can provide context for current discussions and temper unrealistic expectations.
• Works like "The Terminator" or "Ex Machina" depict AI with advanced consciousness and emotions, which are far beyond current technological achievements.
• Public figures, especially those with authority or celebrity status, often have a significant impact on public perception. Their statements can shape narratives and influence how people understand complex topics like AI, even if those statements are not entirely accurate.