My first taste of debating
Reflections and learnings after debating whether the US should pause AI
TLDR
Debating can be a good way to do truth-seeking, but it is not the best way and it is only good if done to a high standard. However, the competitive element of debating did make me strive harder to find arguments, beyond what I normally do. I recommend trying debating if you have not done it before, as it does provide a combination of benefits: learn about various topics, meet knowledgeable and sharp people, have fun, practice public speaking, practice quick-thinking.
Setting the scene
Last week, I tried debating for the first time. The event was organised by BlueDot and it was open to everyone regardless of their prior experience. The motion being debated was: ‘Should the US pause the development of frontier AI?’
I was randomly assigned to be against the motion. It is worth knowing I do think that the US should pause, but it is a weakly held position as I have not thoroughly thought it through. More precisely, I think some people should push for a pause, to help shift the overton window.
Each team (consisting of three people) was given roughly 20 minutes to prepare. Each person had four minutes allocated for speaking during the debate.
The order of the different speakers was:
1st speaker for the motion
1st speaker against the motion (I had this role)
2nd speaker for
2nd speaker against
1st ‘neutral speaker’
2nd ‘neutral speaker’
3rd and final speaker against (this is not a typo!)
3rd and final speaker for
Explanation of the roles:
The 1st speakers for and against the motion should introduce the arguments of their team.
The 2nd speakers for and against should respond to the opponents and/or introduce new arguments and/or go into more detail on arguments already given.
The neutral speakers get to choose which side of the debate they are on, declaring this at the start of their speech.
The 3rd speakers for and against should not introduce any new arguments, and instead go into detail on any existing.
Reflections and learnings
I was taught a powerful debating strategy by an experienced debater on my team: we accept the premises of the opposing team (e.g. AI poses big threats, we should priotize safety, we can pass appropriate laws), then argue that their position does not achieve their goals and may even make things worse.
The arguments we came up with were:
Pausing is likely to lead to bigger jumps in capabilities, compared to current approach of making incremental improvements, and continuous testing and deploying.
Pausing is hard to enforce and will just push the development of frontier AI underground. Compare with illegal drugs or how the US tried banning alcohol.
Pausing will hand over the poll position to China.
We did not get strong counter-arguments.
This is a result of incentives baked in to debating itself and because there is little time to react well.
We had empty sound bites like ‘You say we will hand over the lead to China if we pause. This is just wrong! China is too far behind the US.’
One of the neutral speakers spent their whole four minutes talking about the risks of AI and how safety is important, which I had explicitly said we fully agree with in my opening statement.
This went both ways! We also did not provide strong counter-arguments to their arguments.
I had little incentive to engage deeply with the speakers after me. Partly because the arguments weren’t great (at no fault of the speakers! As already said above, the lack of time really is brutal), but mainly because once I had spoken, the rest of the debate was out of my hands.
The competitive element *did* incentivise me to find counter-arguments to the pause position. I am competitive by nature so it is not too surprising. Big lesson for me is that I should find a way to tap into this mindset of finding the best arguments, outside of competitive contexts.
If you are interested at all in this question of whether AI development should be paused, I strongly recommend looking at PauseAI.