Jeff Sebo
speaker
150 appearances
1 recordings
1 series
first heard Jul 2026
last heard 10 Jul
Jeff Sebo’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in Jul 2026 with 1.
Appearances
Yeah, I definitely want us to be doing more safety and alignment work and not less.
In general, there are a lot of issues that matter all at the same time, and we need to be working on all of them at the same time.
We really need to keep working on ordinary issues like global health and development and animal welfare.
And then within AI, we need to be working on algorithmic bias and economic disruptions, as well as more future-oriented risks involving misuse and loss of control, as well as, I would argue, more future-oriented risks involving the possibility that models could eventually develop their own welfare capacities and their own very different types of pleasure and pain.
This is not a situation where we should be picking as an entire community or society one issue to focus exclusively on.
We should see all of the issues that matter.
We should have a division of labor where different people are working on different issues.
And then we should work together so that we can try to find co-beneficial ways forward, ways forward that can be good for humans, for animals, and potentially eventually AI systems all at the same time.
So this is a new working paper from the Center for Mind Ethics and Policy and Elios AI Research.
And it follows our 2024 report, Taking AI Welfare Seriously.
In that report,
We argued for taking some of those minimum necessary first steps, acknowledge this is a serious issue, start assessing models for welfare-relevant features, and prepare policies and procedures for treating them with an appropriate level of moral concern.
In the time since then, as you noted, a lot of people have started working on this topic, including at companies.
And a lot of people have also been making very confident arguments that AI systems either have or lack technology.
consciousness based on one type of evidence.
People might be looking at behavior alone and saying, wow, that behavior is so impressive, they must be conscious.
Or they might be looking at design alone and saying, well, they were designed for prediction and therefore that must be the only reason why they behave the way they do.
And part of what we argue in this report is that if we truly want to understand how plausible it is that AI systems might be developing welfare-relevant properties like consciousness or sentience, the ability to experience pleasure and pain, agency, the ability to act on desires and preferences, part of how we tell that is by
systematically collecting all of the different types of evidence that matter and putting them together.
Behavioral evidence, how the models behave, internal evidence, how the models work, and developmental evidence, how they came to be.
Showing 61–80 of 150 · page 4 of 8
← Previous
Next →