• A Brief Note on National Security and Developments in Artificial Intelligence

I believe one of the most challenging tensions in AI governance will be between the national security apparatus and a top AI company vs the scientists and the public. Although I do believe that the national security state has its respective place in the slot of our system and for keeping the ship (the state) afloat, I do also believe that the national security apparatus can lack the appropriate checks and balances. But that tension is the point.

I prefer to explore it via the following example:

Journalism has its respective role in society alongside national security. The two sectors in society are meant to work in tension with each other with one protecting and the other serving as a means of accountability. The tension is meant to be productive rather than it being national security vs journalism. A more nuanced or meta exploration is that the relationship between journalism and the national security apparatus is BOTH adversarial + pragmatically friendly. Abstracting from this national security/journalism example, the tension between scientists and the national security apparatus is similar; the tension is THE point. You need both to get the broader system stable (the state) but you can’t achieve that without having productive forms of tension within the system (between the scientists and the state). 

Hence, my concern regarding AI developments is not the national security apparatus’ theoretical purpose but rather what it means in the context of the rise of authoritarian elements in our current system. My approach is one where I am pessimistic about the future given the reality, but despite it act with optimism for the purpose of changing it; thus, I would like to conclude with how this tension could be addressed by using an additional example.

One example scenario of developments in AI could involve a merge or form of partnership between a national security institution and AI companies (for example, OpenAI) for the purpose of reinforcing and amplifying their respective power. In order to address such a scenario, you would need to organize, collaborate and use the cherished tools of accountability democracy has enshrined us with: 

• Journalism for transparency

• Policy for AI researchers as a protected category

• Congress to enforce those checks + balances on national security (a tough task)

• Legal for whistleblower protections

Otherwise you get dystopian futures where CEOs could hypothetically use the security state to censor scientists who without such state pressure, would have whistleblown a long time ago.

While I am in particular interested in stronger legal protections for AI researchers in the face of corporate and national security pressures, I am rather more interested in protections that address such scenarios before they happen. Such frameworks would set an example for how to constrain/mitigate such scenarios not just for artificial intelligence but other forms of transformative developments. 


Speedrunning misalignment doesn’t need to happen with misaligning the system itself; it can also happen with suppressing those who would build the aligned system in the first place.