[ Question ]

Can we convince people to work on AI safety without convincing them about AGI happening this century?

by BrianTan1 min read26th Nov 20203 comments

8

Existential riskAI Alignment
Frontpage

Context for the question

I recently had a call with someone working in the AI/x-risk space that he thinks we can convince more people to work on AI safety-related efforts without needing to convince them that artificial general intelligence (AGI) will be achieved within this century. He didn't expound on reasons why, and this person is quite busy, so I'd rather poll forum readers instead to answer my question above.

The view on AGI of AI researchers in the EA community vs. those outside

I ask this because even if many EAs in the AI risk space think that AGI will likely be achieved within this century (and I imagine that the median view among EAs in this space is that there's a 50% chance AGI will be created by 2050), this view is still contentious in the mainstream AI community (and in mainstream media generally). However, this person I had a call with said that more AI researchers are paying attention now to AI safety thanks to various efforts/reasons, so he thinks it's easier now to get people to work on safety (i.e. make AI systems more explainable and safe) without needing to convince them about AGI. I can also imagine that it could be easier to convince AI researchers to do AI safety-related work without trying to convince them about AGI happening this century.

My experience interviewing an AI professor in the Philippines

I can sense that the AGI view within this century is contentious because I recently interviewed a leading AI professor/researcher in the Philippines, and he think we won't achieve AGI within this century (and he thinks that it's still far away). I don't know any AI researchers from the Philippines yet (where I'm from) who share the view that AGI will be created within this century, and I would imagine it would be hard to find AI researchers locally who already have similar views to EAs about AGI. However, the professor told me that he is interested in doing a research project related to making AI models more explainable, and that he also wants to be able to train AI models without needing large amounts of compute. I could sense that making AI models more explainable helps towards AI safety research (I don't know about if training AI models without needing large amounts of compute is safety related - probably not?). 

Crowdsourcing resources/thoughts on this question

However, I'd love more people to tell me if they think we can grow the quantity and quality of efforts of the AI safety community by focusing on arguments as to why AI should be explainable and safe, and not focus on trying to convince people that AGI will happen this century. If anyone can point me to resources or content that tries to convince people to work on AI safety without making the case for AGI happening this century, that would be great. Thanks!

8

New Answer
Ask Related Question
New Comment

2 Answers

On his recent interview with FLI, Andrew Critch talks about overlaps between AI safety and current issues, and the difference between AI safety and existential safety/risk. Many (but not all) AI safety issues are relevant to current systems, so people who care about x-risks could focus on safety issues that are novel to advanced systems.

If you take a random excerpt of any page from [Aligning Superintelligence with Human Interests] and pretend that it’s about the Netflix challenge or building really good personal assistants or domestic robots, you can succeed. That’s not a critique. That’s just a good property of integrating with research trends. But it’s not about the concept of existential risk. Same thing with Concrete Problems in AI Safety.

In fact, it’s a fun exercise to do. Take that paper. Pretend you think existential risk is ridiculous and read Concrete Problems in AI Safety. It reads perfectly as you don’t need to think about that crazy stuff, let’s talk about tipping over vases or whatever. And that’s a sign that it’s an approach to safety that it’s going to be agreeable to people, whether they care about x-risk or not...

...So here’s a problem we have. And when I say we, I mean people who care about AI existential safety. Around 2015 and 2016, we had this coming out of AI safety as a concept. Thanks to Amodei and the Robust and Beneficial AI Agenda from Stuart Russell, talking about safety became normal. Which was hard to accomplish before 2018. That was a huge accomplishment.

And so what we had happen is people who cared about extinction risk from artificial intelligence would use AI safety as a euphemism for preventing human extinction risk. Now, I’m not sure that was a mistake, because as I said, prior to 2018, it was hard to talk about negative outcomes at all. But it’s at this time in 2020 a real problem that you have people … When they’re thinking existential safety, they’re saying safety, they’re saying AI safety. And that leads to sentences like, “Well, self driving car navigation is not really AI safety.” I’ve heard that uttered many times by different people.

Lucas Perry: And that’s really confusing.

Andrew Critch: Right. And it’s like, “Well, what is AI safety, exactly, if cars driven by AI, not crashing, doesn’t count as AI safety?” I think that as described, the concept of safety usually means minimizing acute risks. Acute meaning in space and time. Like there’s a thing that happens in a place that causes a bad thing. And you’re trying to stop that. And the Concrete Problems in AI Safety agenda really nailed that concept.

A few other resources about bridging the long-term and near-term divide:

If I were to try to convince someone to work on AI safety without convincing them that AGI will happen this century, I'd say things like:

  1. While it may not happen this century, it might.
  2. While it may not happen this century, it'll probably happen eventually.
  3. It's extremely important; it's an x-risk.
  4. We are currently woefully underprepared for it.
  5. It's going to take a lot of research and policy work to plan for it, work which won't be done by default.
  6. Currently very few people are doing this work (e.g. there's more academic papers published on dung beetles than human extinction, AI risk is even more niche, etc. etc. (I may be remembering the example wrong))
  7. There are other big problems, like climate change, nuclear war, etc. but these are both less likely to cause x-risk and also much less neglected.

That said, I think I have a good shot of convincing people that there's a significant chance of AGI this century.