Robert Long

Main points in favor of this grant

I have been trying to nudge Rob in this direction since earlier this year.

Earlier this year I was involved in a startling conversation. Rob Long had been speaking about the chances that we will see conscious AIs in the coming years. (And I had started to grok this possibility.) Now, he was talking about research collaborations he might aim for in future. Rob had empirical projects in mind that could only be done with access to frontier models. Should he bug a colleague-of-colleague to work at [top AI lab X]? Should he ask [collaborator Y at top AI lab Z] about the possibilities at his employer? Rob’s conclusion was: not right now. Rob already had his plate full with other work, the request might be annoying and, besides, Rob had already had a similar request to similar people declined-ish a couple of months ago.

This situation struck me as being preposterous. Here is one of the world’s top experts on AI consciousness, claiming a nerve-wracking chance of AI consciousness in the not-too-distant future, with fairly strong professional links at top AI labs and ~shovel-ready ideas for empirical projects, preparing a not-terribly-costly ask (give me a work desk, ~0 engineering time, and model access to work on these research questions)... and he is unsure whether he should make the ask?!

It seemed to me that the right question to ask was more like “should I try to start a research group as soon as possible?”. (Of course there are many reasons why starting a research group might be a bad idea. But, even if that were the case, Rob should at the very least be asking to work at places that would enable him to work on his empirical projects.)

I want Rob to move the needle on empirical AI consciousness projects harder and faster. In the short-term (see below), this means doing less ‘public-facing work and thinking about his professional opportunities,’ and more ‘thinking through a theory of change for AI consciousness research, spending more time on empirical research with existing collaborators (e.g. Ethan Perez), and pushing for ways he can continue this research in the near future.’

Donor's main reservations

First, I don’t think Rob needs funding in some sense. But I’m not super concerned about this. People should be paid for important work and, besides, I’m partly trying to set up a good incentive environment for future grantees.

Second, I think that I can only counterfactually change a fraction of Rob’s professional activities. Currently, his work falls under the following buckets:

• Co-authoring a paper with Ethan Perez,
• Co-authoring a paper with Jeff Sebo,
• Responding to media and podcast requests about his recent paper, and other writing related to that paper, and
• Job search stuff, applying for funding.

Bucket (1) is the sort of work that I want Rob to be doing more of: activities that directly move the needle on empirical work in his area of expertise.

I instinctively feel less excited about bucket (2), because this paper will not involve empirical AI consciousness research. But I don’t want to impose on Rob’s pre-existing commitment to this project. Also, the issues of the paper have some overlap with writing a strategy doc. (Though this overlap should not be overstated, as academic work is optimized for different things than a strategy document).

Bucket (3) I think Rob should be doing less of. The public-facing work mentioned above does not obviously move the needle on empirical work — and to the extent it does (e.g. indirectly via field-building or career capital), I would feel better if Rob undertook this work after having reflected more on his theory of change for AI consciousness research, rather than as a natural consequence of the release of his recent paper. And, unlike for bucket (2), giving up on some bucket (3) commitments feels low-downside — Rob is not going to be a less interesting podcast guest in 1 years time!

Bucket (4) feels like a waste of time that I want Rob to avoid.

My understanding is that buckets (3) and (4) add up to a bit less than half of Rob’s time at the moment.

Third, empirical projects in AI consciousness feels like a tricky area where I am extremely out-of-my-depth. I am strongly relying on Rob being a ‘reasonable expert who won’t make dumb decisions that make me regret this grant.’ That said, I feel very good about relying on Rob in this way.

Process for deciding amount

Time buckets (3) and (4) add up to 20 hrs/wk * 5 weeks = 100 hours time. Rounding up to 120 (for possible underestimation, and professional expenses), at $60/hour, I will provide $7,200 funding. I'm leaving up to $12,000 as a funding goal, in case anyone wants to fund the remainder of Rob's time during the 5 weeks.

Conflicts of interest

Please disclose e.g. any romantic, professional, financial, housemate, or familial relationships you have with the grant recipient(s).

I was housemates with Rob for a couple of months in early 2023, which is how I found out about this grant.

Please note this grant has been approved but at the time of writing it has not been paid out.

• We funded Robert Long to run a workshop on the science of consciousness for current and near-term AI systems. Robert and his FHI colleague, Patrick Butlin, began the project on consciousness in near-term AI systems during their time at FHI, where they both worked in the digital minds research group. Since January of this year, Rob has been continuing the project while a philosophy fellow at CAIS. There are surprisingly few people investigating the consciousness of near-term AI systems, which I find pretty worrying given the rapid pace of progress in ML. I think that it’s plausible we end up creating many copies of AI systems and use them in ways that we’d consider immoral given enough reflection , in part due to ignorance about their preferences. The workshop aimed to produce a report applying current theories of consciousness (like integrated information theory and global workspace theory) to current ml systems.
• I think that Rob is an excellent fit for this kind of work; he is one of the few people working in this area and has written quite a lot about AI consciousness on his blog. He has a PhD in philosophy from NYU, where he was advised by David Chalmers, and has experience running workshops (e.g. in 2020, he ran a workshop on philosophy and large language models with Amanda Askell).