AI: Hope, Hype, and Headwinds | Milken Institute Global Conference 2024

Los Angeles Global Context & Economic Ambitions

• Los Angeles is preparing to host a convergence of major global events: the 2026 World Cup, the 2027 Super Bowl, and the 2027/2028 Olympic and Paralympic Games.
• The city aims to position itself as a premier hub for creativity, business, and innovation, anticipating millions of visitors from over 200 countries.
• A $30 billion capital improvement program is underway, including a direct rail connection between LAX and the LA Metro system to facilitate access to communities.
• LAX served over 75 million passengers last year, while the Port of Los Angeles and Long Beach complex remains the top container port in the Western Hemisphere for 24 consecutive years.
• The port complex drives significant economic activity, connecting one in nine Southern California jobs and nearly 3 million jobs nationwide, handling nearly 30% of all US sea-based imports and exports.
• Public transportation progress includes the recent completion of a regional connector system enabling travel across the entire region.
• The city maintains a goal to achieve 100% clean energy by 2035.
• Los Angeles faces a critical homelessness challenge, with an estimated 46,000 individuals sleeping outdoors nightly, prompting a push for innovation-driven solutions in housing and social services.

Generative AI: The Current State & Technological Shift

• Generative AI is identified as the largest technology shift since the advent of the early internet and web browsers, driving exponential growth comparable to Amazon's historical trajectory.
• Two concurrent shifts are accelerating the rate of change: a step-function increase in the complexity of solvable problems and a simultaneous broadening of accessibility without requiring large data science teams.
• Organizations are moving from reactive AI usage (generating content or summarizing data) to proactive "agentic" capabilities where AI systems plan, execute, and verify workflows autonomously.
• The architecture driving these changes is expected to see a 30x increase in training performance and a 75% reduction in energy costs with the upcoming NVIDIA Blackwell chips.
• The technology is rapidly transitioning from consumer-facing prototypes to enterprise-grade deployment, with companies reporting incremental revenue or cost savings of approximately $1 billion annually.
• Regulated industries (healthcare, finance, insurance) are adopting generative AI faster than expected, leveraging existing high-quality, private data to bypass the data governance hurdles often faced by tech-native companies.
• Specific enterprise implementations include using AI to summarize 30% of a doctor's patient chart review time, automating legal document analysis for 160-year-old insurance firms, and enabling customer service agents to act as brand storytellers and sales drivers (e.g., Gucci case study).
• AI is currently automating 80% of knowledge worker tasks; a shift toward 90-95% automation is expected to drive "agentic" systems where humans act as managers of AI teams rather than sole doers of tasks.

Data Strategy, Synthetic Data, and Future Architecture

• Enterprises are focusing on unifying siloed structured and unstructured data (e.g., Slack conversations, SaaS applications) to create "persistent data loops" that feed back into model training.
• A significant research breakthrough anticipated within the next 12-18 months is the use of "synthetic data," where models generate and train on their own data (similar to AlphaGo Zero) to solve reasoning tasks rather than just mimicking human internet text.
• The "next token" paradigm is evolving to predict "next actions" rather than just text, enabling software agents to perform real-world proactive tasks on behalf of users.
• The competitive advantage for future companies will rely on their unique, private data assets which serve as "ground truth" for training specialized models, rather than access to public models alone.
• Perplexity reported $20 million in annual recurring revenue just eight months after monetizing, highlighting the unprecedented speed of revenue generation in the generative AI space.
• Companies like OpenAI have achieved billion-dollar valuations with significantly fewer employees (approx. 1,000-2,000) compared to traditional software firms, demonstrating high revenue-per-employee efficiency.

Workforce Implications & Economic Outlook

• The prevailing view is not that jobs will disappear, but that "generalists" will be empowered to do more with less, leading to increased organizational efficiency and potentially higher aggregate hiring (similar to Amazon's robotics strategy).
• Up to 80% of knowledge worker professions could be handled 80% by AI by the end of this year or early next year, shifting the workforce dynamic toward managing AI "co-pilots" or "teams of agents."
• The "hockey stick" inflection point for the technology's economic impact is predicted to occur over the next 12-18 months, with full maturation expected in 24-36 months.
• Salesforce has launched "Trailhead," a free AI training platform with over one million certifications issued since the previous summer to facilitate workforce upskilling.
• Leaders anticipate that the most valuable employees will be those capable of integrating AI tools to amplify their output, while those resisting AI adoption will face significant competitive disadvantages.

Risks, Safety, and Open Source Debate

• Key fears regarding AI advancement include disinformation, discrimination, data security, privacy violations, and the potential for high-quality biological or cyber threats via open-source models.
• There is a strong divergence on open-source models; some panelists argue that open weights are essential for capital-efficient innovation, while others caution that unrestricted access to high-performance models poses national security and safety risks.
• The "trust layer" is identified as the critical industry required to secure agentic AI, involving strict monitoring, guardrails, and human-in-the-loop verification before systems execute autonomous actions.
• Political and societal risks include the potential for foreign state actors to exploit AI for divisiveness, particularly in the context of upcoming election cycles.
• Panelists acknowledge that while current models have hallucination issues, the rate of improvement suggests that high-quality, reliable output (e.g., passing the bar exam in the top 10%) is imminent, though "AGI" timelines remain uncertain.