Opens profile photo
Follow
Click to Follow SarahShoker
Sarah Shoker
@SarahShoker
Research Scientist | Geopolitics of AI | Past: SSHRC Postdoc ; PoliSci PhD | Obnoxious dog mom | Loves scifi + fantasy | Views mine
San Francisco, CAJoined July 2010

Sarah Shoker’s Tweets

I’m generally a pretty even-keeled person, but I will say this—if your proposed solution for a difficult engineering problem involves unilateral air strikes and intimations towards nuclear war, I do not think it’s safe or good for you to be a part of weighty conversations.
2
22
Show this thread
Astonished by the intellectual self-deception and mental gymnastics people are willing to do to frame Eliezer's calls for violence as anything other than calls for violence. Either we all push back forcefully against what he is doing, now, or we'll get more of this and worse.
20
135
I hate the idea that one can “call upon” the ppl in their field, as though it were the draft. But if there is anyone in IR who’s confused about the state of int’l security in AI and wants to write on this topic, then my DMs are open.
11
Show this thread
If this was a small author then I wouldn’t bother responding. But Y. is taken seriously. In part, this is because IR has done a poor job communicating theory to a general audience. We are not off the hook!
1
6
Show this thread
As much as one can critique securitization theory, it’s obvious to me that the theory gets right the idea of “panic politics” leading to the socialization of exceptional measures in int’l security. This time, nothing less than an authoritarian response (track all GPUs!) will do.
2
11
Show this thread
I’m not inclined to take intl security points from someone who both states he’s against an arms race while simultaneously arguing for unilateral air strikes against data centres. I cannot imagine discourse that is more likely to contribute to arms racing.
1
16
Show this thread
I think many IR ppl are likely to be surprised that many in AI think AGI will up-end nuclear deterrence and, effectively, the modern state system. Though I gotta say that this is the first time I’ve heard someone outright justify a nuclear exchange.
2
6
Show this thread
Okay, I already spotted an error worth correcting. 😮‍💨 Not exactly accurate to say ANT originated in STS when its original popularizers came from anthro, sociology, and philosophy. Maybe better to say that today it’s most associated with STS.
3
Show this thread
I think much of any academic experience depends on your department (not just school.) It was becoming more prominent in IR in Canada around 2015 afaik because of the increased focus on data and war operations.
Quote Tweet
Replying to @SarahShoker
Is ANT discussed that much in academia w.r.t. AI? I've been surprised with how little I see it brought up, although in fairness I'm not really in AI academia
2
(I wouldn't say that I'm committed to ANT above other theories. I discard aspects of theoretical frameworks I don't like and sample across different methods. ANT has its flaws, to be sure.)
1
2
Show this thread
This is why many ANT scholars favor ethnography as an approach to explain how outcomes (social, scientific etc.) are produced through collaboration with inanimate objects and rules. Through observing social life, it's not clear humans come out on top as the most agential.
1
4
Show this thread
ANT scholars often reach their conclusions through empirical observation. Methodologically, ANT scholars tend to be more interest in *describing* real world phenomena than explaining it. (This is a target of criticism towards ANT, that causality isn't treat seriously enough.)
1
4
Show this thread
If you are someone who is very interested in Qs about sentience, then this theory is likely not for you. Ditto if you think that AI is more important than institutions or other impactful tech.
1
4
Show this thread
A more modern example might include human-machine teaming. e.g. drone crews view the world through what the drone sensor is able to capture & by applying legal labels like 'legitimate target' to objects and ppl. Drone warfare does not exist otherwise.
1
4
Show this thread
ANT scholars treat non-human entities and processes as being *equal actors* in the production of the world. That's not because they think technology or bureaucracies have free will or are sentient. It's because these entities 'act' on human life and vice-versa.
2
6
Show this thread
ANT originates in Science & Technology Studies to chart the interaction b/w human behaviour, organizational practices (e.g. bureaucracy, workplace norms), and technology. Where ANT becomes relevant to AI is in its treatment of actors & the entities that get to be actors.
1
7
Show this thread
It feels odd to me that actor-network theory isn't more widely discussed in AI spaces *outside* academia. If we're trying to understand whether AI can be a 'collaborator' or 'act' on the world, then I would have expected Bruno Latour to be cited more often.
4
21
Show this thread
I’ve seen some confusion about whether GPT-4 can do drug discovery (not really.) Our excellent chemistry red teamer breaks down the process here:
Quote Tweet
Can GPT-4 do drug discovery? No, but it can help. Let's walk through GPT-4 proposing new drugs. This is called knowledge-based screening. We're trying to fill a list of plausible compounds that could lead to new drugs based on research papers. 1/n
Show this thread
Image
14
One of the key elements of red teaming - understanding how models can interact with other systems. This is important *both* for unlocking incredible benefits, and for understanding unintended consequences or misuses. Check out Andrew's chemistry example in the System Card!
Quote Tweet
I was lucky enough to be a red teamer for GPT-4 and had a great time exploring it for chemistry. It's been tough to keep quiet! Want to thank the excellent team (especially @_lamaahmad!) for taking my feedback seriously about risks and mitigations for LLMs in chemistry. 1/2
Show this thread
2
73
This was a massive multi-team effort. And of course, a big shout out to all external red teamers and early reviewers! Don't forget to check out the acknowledgements section in both papers. :)
2
8
Show this thread