Dan H
NeurIPS ML Safety Workshop 2022
AI and Evolution
Catastrophic Risks from AI #1: Introduction
For concrete research directions in safety and several dozen project ideas, please see our paper Unsolved Problems in ML Safety: https://arxiv.org/abs/2109.13916
Note that some directions are less concretized than others. For example, it is easier to do work on Honest AI and Proxy Gaming than it is to do work on, say, Value Clarification.
Since this paper is dense for newcomers, I’m finishing up creating a course that will expand on these safety problems.
[MLSN #6]: Transparency survey, provable robustness, ML models that predict the future
Catastrophic Risks from AI #2: Malicious Use
Aggregating Utilities for Corrigible AI [Feedback Draft]
The Polarity Problem [Draft]
I don’t think Redwood’s project had identical goals, and would strongly disagree with someone saying it’s duplicative.
I agree it is not duplicative. It’s been a while, but if I recall correctly the main difference seemed to be that they chose a task with gave them a extra nine of reliability (started with an initially easier task) and pursued it more thoroughly.
think I’m comparably skeptical of all of the evidence on offer for claims of the form “doing research on X leads to differential progress on Y,”
I think if we find that improvement of X leads to improvement on Y, then that’s some evidence, but it doesn’t establish that it’s differential. If we find that improvement on X also leads to progress on thing Z that is highly indicative of general capabilities, then that’s evidence against. If we find that it mainly affects Y but not other things Z, then that’s reasonable evidence it’s differential. For example, so far, transparency hasn’t affected general capabilities, so I read that as evidence of differential technological progress. As another example, I think trojan defense research differentially improves our understanding our trojans; I don’t see it making models better at coding or gaining new general instrumental skills.
I think commonsense is too unreliable of a guide when thinking about deep learning; deep learning findings are phenomena are often unintelligible even in hindsight (I still don’t understand why some of my research papers’ methods work). That’s why I’d prefer empirical evidence. Empirical research claiming to differentially improve safety should demonstrate a differential safety improvement empirically.
Catastrophic Risks from AI #3: AI Race
A brief overview of the contents, page by page.
1: most important century and hinge of history
2: wisdom needs to keep up with technological power or else self-destruction / the world is fragile / cuban missile crisis
3: unilateralist’s curse
4: bio x-risk
5: malicious actors intentionally building power-seeking AIs / anti-human accelerationism is common in tech
6: persuasive AIs and eroded epistemics
7: value lock-in and entrenched totalitarianism
8: story about bioterrorism
9: practical malicious use suggestions
10: LAWs as an on-ramp to AI x-risk
11: automated cyberwarfare → global destablization
12: flash war, AIs in control of nuclear command and control
13: security dilemma means AI conflict can bring us to brink of extinction
14: story about flash war
15: erosion of safety due to corporate AI race
16: automation of AI research; autnomous/ascended economy; enfeeblement
17: AI development reinterpreted as evolutionary process
18: AI development is not aligned with human values but with competitive and evolutionary pressures
19: gorilla argument, AIs could easily outclass humans in so many ways
20: story about an autonomous economy
21: practical AI race suggestions
22: examples of catastrophic accidents in various industries
23: potential AI catastrophes from accidents, Normal Accidents
24: emergent AI capabilities, unknown unknowns
25: safety culture (with nuclear weapons development examples), security mindset
26: sociotechnical systems, safety vs. capabilities
27: safetywashing, defense in depth
28: story about weak safety culture
29: practical suggestions for organizational safety
30: more practical suggestions for organizational safety
31: bing and microsoft tay demonstrate how AIs can be surprisingly unhinged/difficult to steer
32: proxy gaming/reward hacking
33: goal drift
34: spurious cues can cause AIs to pursue wrong goals/intrinsification
35: power-seeking (tool use, self-preservation)
36: power-seeking continued (AIs with different goals could be uniquely adversarial)
37: deception examples
38: treacherous turns and self-awareness
39: practical suggestions for AI control
40: how AI x-risk relates to other risks
41: conclusion
I think the adversarial mining thing was hot in 2019. IIRC, Hellaswag and others did it; I’d venture maybe 100 papers did it before RR, but I still think it was underexplored at the time and I’m happy RR investigated it.
Catastrophic Risks from AI #4: Organizational Risks
I think the “heart in a lightbulb” insignia for EA is a great design choice and excellent for outreach, but there is no such communicable symbol for utilitarianism. Companies know to spend much on design for outreach since visualization is not superfluous. I do not think the optimal spending is $0, as is currently the case. A point of the competition is finding a visual way of communicating a salient idea about utilitarianism suitable for broader outreach. I do not know what part is best to communicate or how best to communicate it—that’s part of the reason for the competition.
I agree fitness is a more useful concept than rationality (and more useful than an individual agent’s power), so here’s a document I wrote about it: https://drive.google.com/file/d/1p4ZAuEYHL_21tqstJOGsMiG4xaRBtVcj/view
Even in deep learning, proofs have by and large been a failure. Proofs would be important, and there are many people trying to find angles of attack for having useful proofs for deep learning systems, so it is hard to say it is neglected. Unfortunately useful proofs are rarely tractable for complex systems such as deep learning systems. Compared to other interventions I would not bet a substantial amount on proofs for deep learning systems given its important, neglectedness, and tractability.
Would love to identify and fund a well-regarded economist to develop AI risk models, if there were funding for it.
“AI safety” refers to ensuring that the consequences of misalignment are not majorly harmful
That’s saying that AI safety is about protective mechanisms and that alignment is about preventative mechanisms. I haven’t heard the distinction drawn that way, and I think that’s an unusual way to draw it.
Context:
Preventative Barrier: prevent initiating hazardous event (decrease probability(event))
Protective Barrier: minimize hazardous event consequences (decrease impact(event))
Broader videos about safety engineering distinctions in AI safety: [1], [2].
The better reference class is adversarially mined examples for text models. Meta and other researchers were working on a similar projects before Redwood started doing that line of research. https://github.com/facebookresearch/anli is an example. (Reader: evaluate your model’s consistency for what counts as alignment research—does this mean non-x-risk-pilled Meta researchers do some alignment research, if we believe RR project constituted exciting alignment research too?)
Separately, I haven’t seen empirical demonstrations that pursuing this line of research can have limited capabilities externalities or result in differential technological progress. Robustifying models against some kinds of automatic adversarial attacks (1,2) does seem to be separable from improving general capabilities though, and I think it’d be good to have more work on that.
This researcher’s work attitude is only part of a spectrum. Many researchers find great returns working 80+ hours a week. Some labs differentiate themselves by having usual hours, but many successful labs have their members work a lot, and that works out well. For example, Dawn Song’s students work a ton, and some other Berkeley grad students in other labs are intimidated by her lab’s hours, but that’s OK because her graduate students find that environment suitable. It’d be nice if this post was more specific about how much of the work culture discontent is about hours vs other issues.