The Executive Order on AI
(Maps viewable on desktop only)
The recent Executive Order on AI comes amidst a battle between of tech titans. On one side, Microsoft, OpenAI, and Google are leaning toward exclusive models. On another, Facebook (via Yann Lacun) and Andreesen Horowitz are advocating open source. Where do their views fit versus White House?
We put the 40,000+ words (420 paragraphs) in the executive order through our Narrative Analytics tool, and compared them to what the tech titans are saying. Each dot on the below map represents a paragraph. Hover your cursor over the dots to read the underlying statement for each.
Detailed clusters within the Executive Order
Let's first dig into the White house announcement. In the above map, move the slider to the right to reveal a more detailed view of clusters
- IP and content represent 4.1% of the texts, citing the need to clarify issues related to IP and generated content.
- There is significant language about foreign use. 4.8% of the texts reference the need to identify foreign users of US IaaS (infrastructure as a service) technology.
- The act also sets policy in motion to attract foreign nationals with AI skills. Talent and employment texts represent 3.6% of the texts. If you hover the dots in that cluster, you will find several references to hiring talent from other countries.
Viewing the map against sections defined in the EO
Below is the same map, with the dots coloured by the sections of the executive order. Click on number 7 in the legend, to see how the 80 texts in the ‘Advancing Equity and civil rights’ section distribute across the map. 22.5% of the texts in this section are about interagency coordination. The White House is not just saying “it should be fair” – it is providing air cover to address key bureaucratic barriers.
Click the two dots at the bottom of the legend to compare the respective narrative footprints of the Frontier Model Forum versus the Facebook faction (LeCun et all). Though limited published views are available as an overlay, the location shows they have similarities in terms of high level references to responsible use. However, only the Frontier Model forum is pointing to the lab and the technical aspects of model development as the means to that end.
For context, Google, Microsoft, OpenAI and Anthropic launched a lobbying effort for AI regulation called the Frontier Model Forum, and appear focused on proprietary tools and commercial licensing. Conversely, Facebook, Shopify and Andreesen Horowitz have lined up behind open source. Click on the legend to select the bottom two dots. This will reveal the public statements from these two factions, respectively. It reveals that:
None of the industry voices have mentioned equity or fairness. Though neither of the 2 factions have said much publicly, the Frontier Model Forum lobby are more focused on the lab, and the role of experts in building models. One of the Frontier Forum texts mentions the need to ‘help the public understand’ – suggesting an emphasis on the expert view (and possibly a teacher/learner mindset). They mention nothing about transparency.
The open sourcers are focused on transparency. Paragraphs from their statement, fall entirely within the “responsible development and safeguards” narrative. Their argument is simple: More access and transparency is the best safeguard.
Though the White House has shared more than 400 paragraphs on its position, we have less than 10 from the big tech factions above.
Behind the scenes, the Frontier Model Forum has likely shared voluminous points of view with legislators. It would be good to see what they are saying.