Author: Dr. Andrea Little Limbago, SVP, Applied AI
We are at a rare moment in history, one where an industrial revolution, an information revolution, and a renaissance are all occurring at once. Whoever wins the AI race will shape a “new golden age of human flourishing, economic competitiveness, and national security.”
This is the baseline premise of the AI Action Plan released on July 23rd (as well as three related executive orders). The AI Action Plan establishes three core pillars focused on accelerating innovation, building AI infrastructure, and active US leadership and diplomacy for shaping global AI norms.
There is a solid emphasis on secure AI, global engagement, and workforce development, all of which are critical to accelerating AI innovation. However, there also are some inherent contradictions that will be important to watch as these recommendations (it is not binding) transform into a strategy and legislation in the years ahead.
Building Secure and Transparent AI
Across all three pillars, the AI Action Plan contains three cross-cutting themes focused on workforce development, security, and trustworthiness.
Each of these are critical for US economic and national security. From a workforce perspective, there is a nice balance of improving AI literacy, training, and developing a workforce that complements the AI advancements, not replacing the workforce. This is a critical area that rightly must be prioritized while also comprehending the significant disruptions already underway.
While much focus has been on broader, industry-specific workforce assessments of the impact of AI, there also is growing concern (and evidence) of AI eliminating many entry-level roles. These are the same roles that are critical for learning and gaining expertise required at more senior positions and importantly are essential for assessing and evaluating the output of AI. As one report summarized, the AI apocalypse may already be here for recent graduates. Addressing this level of disruption through new training and other opportunities and restructuring requires urgent attention.
Second, security was a welcome theme across the AI Action Plan, a topic often ignored in the rush to deploy the latest advancements. At interos.ai, we are strong proponents of Secure AI, understanding the need to ensure security of the data and algorithms, both of which are at growing risk of attack, poisoning, and manipulation. The AI Action Plan places the US at the forefront of these risks, recommending concrete protections and processes for evaluating the national security risks and data risk in existing and frontier models.
Finally, transparency and trustworthiness permeate the AI Action Plan. There is welcome emphasis on explainability, noting, “Today, the inner workings of frontier AI systems are poorly understood. Technologists know how LLMs work at a high level but often cannot explain why a model produced a specific output.” This is a wicked problem, but one critical for innovation and adoption. Societal adoption of general-purpose technologies is a critical factor in determining which countries best innovate and flourish during technological revolutions. Building trust is critical to societal adoption.
Overcoming Potential Contradictions
The AI Action plan has largely received praise from the tech community due to the lifting of regulations. In fact, “Remove Red Tape and Onerous Regulation” is the core feature within the Innovation pillar.
It then goes on to promote free speech and values, but then subsequently introduces requirements aimed at removing key terms. In fact, the accompanying executive order mandates unbiased principles while focusing on “truthfulness and ideological neutrality”. Those are core tenets of American values, but this also risks constraining innovation by requiring the omission of certain perspectives or data within AI systems. Risk and safety-based use cases also are omitted, but are critical in high-risk use cases, especially surrounding national security.
There are existing lessons to learn from China, which already censors the data going into their AI systems and adversely effects innovation. DeepSeek illustrates this challenge, surfacing historically incorrect answers based on Chinese-censorship protocols. Of course, US GenAI tools are not immune from this problem as well, with many well-known examples of hallucinations and inaccuracies. As the AI Action Plan notes, “High-quality data has become a national strategic asset as governments pursue AI innovation goals and capitalize on the technology’s economic benefits”. Instead of omitting certain perspectives, innovation will require high-quality and accurate data across all scientific realms and spectrums. The higher risk the application, the more critical accurate data quality and safety becomes.
Finally, the focus on both openness and security is commendable but often difficult in reality when it comes to technology systems. Open source software is a great example of expanding innovation and use cases, but also the security risks that may hide beneath the surface. Openness is foundational to innovation and collaboration, but moving forward, security must not be an afterthought but rather a key consideration for open frontier models, while closed model systems in sensitive and proprietary areas may also be essential.
Carp(AI) Diem
Interestingly, the AI Action Plan was released just over a week before the EU AI Act goes into effect on August 2. The US is not alone in the quest to shape the AI-dominated future, with governments and tech giants racing to shape the regulations and seize this unique moment in time.
However, an AI race is not the appropriate analogy, as races are won and then over. But in global competition, innovation and competition don’t stop after achieving a specific milestone – such as the first to achieve artificial general intelligence (AGI). Instead, a holistic strategy is needed to win the race and beyond, and the early signs of this appear throughout the Action Plan. From stopping loopholes in restrictions to scientific advancements to domestic capacity for building advanced semiconductors to an AI literate society, the core elements are here, what remains to be seen is translating them from recommendations to actions to truly seize the day.
Ask us about secure AI innovation at interos.ai.