Divya van Mahajan

What People Really Want From AI

ai-daily-brief-podcast

Overview

This episode of AI Daily Brief (published March 19, 2026) covers three main topics: the posthumous AI-generated performance of Val Kilmer in a new film, a restructuring of Microsoft’s Copilot organization, and a deep-dive analysis of Anthropic’s large-scale qualitative study on public attitudes toward AI. The host (name not stated) presents and editorializes on findings from Anthropic’s 81,000-person global survey about what people hope for, experience, and fear from AI. No source YouTube URL was provided.

Prerequisites

  • Basic familiarity with large language models (LLMs) and conversational AI assistants (e.g., Claude, ChatGPT)
  • General understanding of AI in commercial and consumer product contexts (Microsoft Copilot, Anthropic Claude)
  • Awareness of ongoing public debates around AI safety, job displacement, and AI ethics
  • Familiarity with basic social science research concepts (sample selection, qualitative vs. quantitative methodology, interviewer bias)

Main Points

Val Kilmer’s Posthumous AI Performance

  • Val Kilmer was cast in As Deep as the Grave in 2020 but was unable to shoot any scenes due to late-stage throat cancer before his death.
  • Director Cuarte Voorhees, with full permission from Kilmer’s estate and the cooperation of his children, used AI tools to generate Kilmer’s entire on-screen performance.
  • The film uses Kilmer’s actual (surgically damaged) voice, which coincidentally fit the character of Father Fenton, who suffered from tuberculosis.
  • Voorhees followed SAG guidelines and compensated the Kilmer estate; he frames the project as a model of ethical AI use in filmmaking.
  • Critics argue the AI performance lacks the actor’s spontaneous, uniquely human choices and raises questions of posthumous agency; supporters note the family’s explicit endorsement and Kilmer’s prior enthusiasm for AI voice recreation in Top Gun: Maverick.

Microsoft Copilot Restructuring

  • Microsoft is merging its consumer and commercial Copilot teams under a single unified effort, led by Jacob Andru (promoted to EVP of Copilot), who will now report directly to CEO Satya Nadella.
  • Mustafa Suleiman’s portfolio is narrowed to focus exclusively on proprietary model training and “superintelligence” efforts; he frames the model layer as the primary source of future value.
  • The restructure addresses longstanding fragmentation: Copilot lacked a single clear owner and multiple product versions created user confusion.
  • The move parallels similar AI org restructurings at Google (late 2024), Meta, and Alibaba, suggesting iterative organizational learning is common across the industry.

Anthropic’s 81,000-Person AI Attitudes Study

  • Methodology: Conducted in December 2025 using “Anthropic Interviewer” (a version of Claude designed for conversational research); 159 countries, 70 languages; claimed to be the largest and most multilingual qualitative study ever conducted.
  • Core finding on emotional complexity: Hope and alarm did not divide respondents into separate camps — they coexisted as tensions within the same individuals. What people want from AI and what they fear are tightly bound.

What People Want from AI

  • Professional excellence was the top category (18.8% of hope-related responses), followed by personal transformation (13.7%) and life management (13.5%).
  • When probed deeper, professional goals frequently resolved into personal ones (e.g., automating emails → more time with family).
  • Three meta-clusters underpin all nine response categories:
    1. Making room for life (~one-third of visions): more time, money, mental bandwidth
    2. Doing better, more fulfilling work (~one quarter)
    3. Becoming a better person: learning, healing, growing (~one fifth)
  • Societal transformation hopes (9.4%) often stemmed from personal experiences with healthcare failures or educational inequity, particularly in low- and middle-income countries.

How AI Has Delivered (Per Users)

  • 81% of respondents said AI had taken at least one step toward their stated vision.
  • Top delivery categories: productivity (32%), cognitive partnership, learning, technical accessibility, research synthesis, and emotional support (6.1% — lowest reported, though possibly underreported due to overlap with other categories).
  • Emotional support stories were among the most vivid, including AI used to process grief — but also included cautionary accounts of AI substituting for human relationships with negative consequences.

What People Fear from AI

  • Unreliability/hallucination topped concerns (26.7%), reflecting how greater reliance heightens the cost of AI errors.
  • Jobs and economic displacement (22.3%), loss of autonomy/agency (21.9%), and cognitive atrophy (16.3%) followed.
  • Societal concerns — misinformation (13.6%), surveillance/privacy (13.1%), malicious use (13%) — appeared lower than in mainstream media coverage.
  • Existential risk appeared at the bottom (6.7%).
  • A notably underrepresented-in-media concern: over-restriction — excessive safety filtering and paternalistic AI design.
  • 11% expressed no concern, typically viewing AI as a neutral tool like electricity; they tended to be confident in human adaptability.

Tensions and Dualities

  • Five recurring tensions: learning vs. cognitive dependence; emotional solace vs. substitution for human connection; productivity gains vs. workload acceleration; economic freedom vs. job displacement; and autonomy vs. loss of agency.
  • Benefits were generally grounded in lived experience; harms were more often hypothetical. Example: 33% cited learning benefits (91% had experienced them); 17% feared cognitive atrophy (only 46% had observed it firsthand).
  • Economic gains accrued disproportionately to independent workers: entrepreneurs, freelancers, and employees with side projects reported real economic empowerment at more than triple the rate of institutional employees.
  • Freelance creatives were identified as the most exposed group — AI serves simultaneously as their tool and their competitor.

Geographic Patterns

  • Western and developed nations showed average or below-average AI sentiment.
  • Southern and developing economies showed above-average AI sentiment, consistent with findings elsewhere.

Reactions to the Study and Methodological Debate

  • Supportive view: The scale (81,000 responses) and use of AI as a consistent cross-lingual interviewer removes traditional interviewer bias and achieves coverage no human research team could match.
  • Legitimate critique (Berkeley Haas professor Abhishek Nagaraj): Sample selection is underaddressed; Claude users are not representative of all AI users, let alone the general public. Broader generalizations require explicit caveats.
  • Host’s critique of dismissive critics: Some critics implicitly treat the opinions of non-users as inherently more legitimate than those of AI users — effectively arguing that uninformed opposition is a purer data point than informed experience. The host characterizes this as “intellectual NIMBYism masquerading as methodology critique,” particularly problematic given that billions of people now use AI weekly.

Key Concepts

  • Anthropic Interviewer: A version of Claude configured to conduct structured qualitative research interviews at scale across many languages and countries.
  • Cognitive atrophy: The fear that habitual AI use will degrade users’ own independent reasoning and problem-solving abilities.
  • Cognitive partnership: AI functioning as a knowledgeable, always-available intellectual collaborator, analogous to a colleague.
  • Freelance creatives as “the most exposed middle”: The group for whom AI benefits and harms are most nearly in balance — AI competes with and assists them simultaneously.
  • Over-restriction concern: User anxiety that AI systems, through excessive safety measures and paternalistic content filtering, become less useful and less trustworthy than their potential warrants.
  • Sample selection bias: The methodological problem of drawing general conclusions from a sample that is not representative of the broader population of interest.
  • Hope/fear co-occurrence: The study’s central finding that positive and negative attitudes toward AI are not distributed across separate people but coexist within individuals, most strongly around emotional support use cases.
  • Digital necromancy (critic’s term): A pejorative characterization of using AI to reconstruct or simulate deceased persons for new performances.
  • SAG guidelines on AI: Screen Actors Guild standards governing the use of AI in film production, including consent and compensation requirements.

Summary

Anthropic’s study of nearly 81,000 users across 159 countries reveals that public attitudes toward AI are not polarized between optimists and pessimists — rather, hope and fear coexist within the same individuals, bound together by the same underlying desires: more time, more autonomy, better work, and deeper human connection. The most commonly hoped-for benefit was professional excellence, but this typically masked a deeper personal aspiration, such as reclaiming time for family. The most commonly feared outcome was AI unreliability, not existential risk, which ranked last. The concerns that dominate media coverage — copyright, harm to children, democracy — appeared in the long tail of user responses. Economic gains are real but skewed toward independent workers, while freelance creatives face AI as both opportunity and threat. The host argues that the ability to conduct this kind of study — 81,000 interviews across 70 languages in a week — is itself a landmark achievement with implications far beyond measuring AI sentiment, and cautions against a growing rhetorical pattern in AI criticism that implicitly delegitimizes the experiences and opinions of the billions of people already using these tools.