Tokenomic changes rightly have a high voting threshold because investments are based on trust and stability. At the same time, as @wpb points out, part of investing in the IC is investing in the concept of a NNS which develops over time. So while I’m wary of this change, I also see a purpose for it.
I would push for a conservative solution however due to unknown consequences, such as those that @Accumulating.icp and others bring up. A well-publicized, one time reset might provide valuable data that would inform a better long-term solution.
If the community does proceed with this initiative, I would prefer the solution @lara originally proposed but with a minimum 6/6 (grace/degradation) period for biannual confirmation.
Also, if this isn’t already the case, known neurons themselves should be removed after a period of inactivity on their part.
As per Microsoft Co-pilot: Estimates suggest that around 6 million BTC , or approximately 30% of Bitcoin’s supply, have been irretrievably lost due to forgotten private keys or lost hardware wallet devices.
Probably on a way smaller scale, this happen to ICP as well. People die, people lost their keys. How would you manage to stop them from voting while still following, maybe deciding for a critical vote, while dead, without periodic confirmation?
Thanks @Accumulating.icp , I think this is a valid point - let’s think about this!
I agree that with less overall voting power one could reach the % relevant for decisions.
However, I wonder if it would be easier to reach this % - I think this would depend on how many tokens are in circulation and could be bought by an attacker. I assume that for an attacker it would be hard or impossible to get the sleepers to follow them - so there is a lot of locked tokens which are not “free” to get for an attacker and cannot be used for such an attack.
At the same time for neurons that now have a lot of voting power due to sleepers, it would probably become harder to launch attacks as they have less voting power relative to the decision making, right?
The main problem that I see with this is that it might then be very hard to make decisions for a while after the reset. This could be dangerous in case there are urgent upgrades like security fixes.
Maybe the graph in my original post helps here: a one-off reset would look similar to “reset of following” on the right and has the same disadvantages.
Here’s a rough calculation to assess the increased risk of a 51% attack:
The voting participation rate in topic governance currently stands at 60%.
Given the higher impact on rewards and the notifications provided by the NNS app, it seems reasonable to expect that the portion of active neurons (as defined in the proposal) would be higher. A portion of 80% seems like a plausible assumption.
With an 80% participation rate of active neurons, an entity would need to control 40% of the total voting power to achieve a majority (50%) of the active voting power.
Accumulating such a significant share of voting power would likely be prohibitively expensive (e.g. acquiring nearly all liquid ICP tokens).
By the end of the year, there is projected to be nearly 100 million of potential ICP in undispersed voting rewards. I think it is reasonable to assume that if a hostile buyer were to drive the price up, a significant amount of these would be converted into liquid ICP, causing:
Decreased voting power for friendly neurons
Downward pressure on ICP price
Increased potential voting power for an attacker
Also, an 80% participation rate for voting seems speculative to me, in the absence of experimental data.
It’s quite the assumption to say that people, the majority of which operate under a “set & forget” mindset, will suddenly start increasing voting participation by 20% in light of this implementation. If we’re being honest, you’re more likely to see a 20% drop than increase - the only reason we have 60% participation is due to Named Neurons.
Utilizing a realistic model of 40% participation, we find ourselves in a situation where DFINITY’s 19.73% of total voting power is once again nearly enough to actuate 50% of the vote (49.32% of active voting power) in contrast to the 32.88% of the participating voting power that is DFINITY currently controlled (that took a lot of selling over the last 3 years to achieve).
This also only addresses the governance topic - how will this translate in codebase topics? DFINITY already actuates 99.99% of the vote through a forced genesis delegation.
It seems as though we’re ignoring the elephant in the room.
A 51% is achieved within 6 months implementation of your Periodic Followee Design - although we’ll probably act like it’s not a big deal because “it’s just us, why don’t you trust us”, right…?
I’m not worried about the big bad wolf that’s been pitched to us for years now - I’m curious about what’s going on in-house.
Could this be addressed by allowing neurons to preemptively confirm their configuration prior to the date of reset (e.g. during the 3 or 6 month period spent advertising it)?
My biggest concern is that this only introduces negative feedback mechanisms. If you don’t participate, you don’t get your rewards. That seems fair, but if I confirm my followers and do nothing else, I still get rewarded even though I am contributing nothing meaningful to the process.
What we need are positive feedback mechanisms, which incentivize people to participate deeply and meaningfully in a small number of topics (not all topics, since that would be too time consuming for any one person).
I believe that a smaller group of highly qualified voters is preferable to a large number of confirmation bot voters. This means changing the incentive structure to encourage some direct voting, some vote delegation and confirmation, and higher rewards for those who actively participate!
I believe we need to revisit the idea of allowing active participants to receive greater rewards for participating in neglected topics. If you are active and willing to learn the nuances of a boring or unpopular topic, then you should be rewarded at a higher rate. This could afford those who make the NNS better for everyone the opportunity to build a larger stake.
I think I missed answering this before.
Let’s maybe discuss the details in the context of other design discussions, but I just wanted to say that we agree with this and there are already 2 planned frontend features that I think go in exactly the direction you are mentioning:
Following should be made easier. This should include that it is easily visible which topic are included in the catch-all (and for SNSs which topics are critical).
The dapp could show a form of warning for neurons that are (potentially) missing out on rewards. The initial idea (but we haven’t started designing this in more detail) is to show this in the neurons’ table. This could inclue when a neuron as adjusted voting power, when a neuron hasn’t voted on all of the past 100 proposals, or when a neuron doesn’t have following set up.
We can then define the details within these features, but thought it might help to know that this is on our radar
I think it is reasonable to assume that a large portion of the sleeper neurons are really inactive, also over such a pre-warning period of 3 months.
This would then still mean that their full voting power would fall away as a “cliff” on one day - making it very hard for the rest of the NNS to react if there is an emergency situation exactly on that day or shortly after.
It is then also the question whether enough would confirm to ever get back to 50% (which is maybe likely but might take a longer period of time.)
I believe we need to revisit the idea of allowing active participants to receive greater rewards for participating in neglected topics.
Yes, we agree that it also makes sense to look into how neurons can be incentivised to do more active verification work.
We just launched grants for this as an intermediate solution and plan to work on a design to build this into the NNS later.
We can basically tackle this from both angles.
I wonder if we are now mixing up a few different cases here, considering direct voting power and indirect voting power that also takes into account followers as well as scenarios from outsiders vs neurons that already have a lot of followers today.
First, there is the case of an external attacker who does not yet have any followers. To get to 51% of the voting power, they would either have to convince a lot of people to change following and follow them or they would have to get 51% direct voting power.
Starting from zero, it is probably unlikely that an outsider could convince enough people to get 51% follower voting power. But even if they achieve this, then this is justified voting power in the current governance model.
The alternative is for such a person to stake enough tokens to have 51% of the direct voting power. In this case, Bjoern’s analysis applies. Even if the number of voting participation would be at 60% similar to the current governance participation, since no such party currently has a high voting power, it would be very expensive to get so many liquid tokens from the market.
Second, there is the case of neurons that already have a lot of followers, like for example DFINITY.
Especially for the case of DFINITY this feature would surely be advantageous and help to decrease the voting power that DFINITY has as a large portion of the following from sleepers would fall away.
I’m wondering about the voting power that the nueron induces. What happens to its induced voting power when a sleeper neurons starts to lose voting power? So what will happen to DFINITY’s true total voting power from followers once this change is implemented? Will it go down?
This feels like intentional ignorance at this point😅 I’m going to ignore the wall of doomsday 51% attacker scenarios, that DFINITY has so religiously referenced at convenience (for Neuron Marketplaces, its a problem, yet suddenly, it’s a non issue in this regard. Seems like it’s just a narrative to be used when something has to be pushed through or rejected).
DFINITYs voting power may be reduced through followees being stripped of their voting power, but you’re ignoring the fact that as you take away voting power from quorum, you actually only make the 20% that DFINITY owns stronger, as has been depicted.
This is certainly “advantageous to DFINITY” as you’ve mentioned, as this function will directly increase the % of quorum DFINITY controls.
If you wanted to decentralize it, remove the followee & leave people with their voting power contributing to quorum.
So if the followers get adjusted voting power, then the overall induced voting power of the voting neuron will go down.
Let’s consider a simple example with 3 neurons N1, N2, and N3 that each have a voting power of 10. Let’s assume N2 is a sleeper neuron that follows N1.
Now
Direct voting power is: N1 = 10, N2 =10, N3=10.
Induced voting power (including following): N1=20, N3=10
Total voting power for decisions = 30
=> N1 influences the decision with 2/3 of the voting power
With the new feature, after a while N2’s voting power would be adjusted to zero and the total voting power would be adjusted accordingly, so:
Adjusted voting power is: N1 = 10, N2 = 0, N3 = 10.
=> N1 influences the decision with 1/2 of the voting power
Let me know if I made a mistake!
Then there is the case where the followee has adjusted voting power.
If this followee is a known neuron that should vote directly (but is now inactive), then all the followees would not get rewards and the induced voting power is zero. Hopefully since they don’t get rewards the followers would switch to follow others.
It gets a bit more interesting if there is a longer chain of following. In that case, if the neuron at the top of the chain votes, then votes would still be triggered for everyone who follows a follower, and it would depend on each follower and their activity whether the adjustment is applied or not.
Does this help or were you looking for something else?
Maybe we can consider the example below that I made and you can let me know if I had an oversight?
Indeed, the direct voting power (in relation to the total voting power relevant for decisions) of neuron N1 would increase (it was 1/3 and goes to 1/2), but the induced voting power would still go down (from 2/3 to 1/2). Do you agree with this?
Respectfully, I understand how the function works, but the scenario you’ve crafted in which a neuron loses voting power has no correlation to addressing what I’ve depicted.
I would appreciate if we could stay on conversation rather than attempting to introduce me as a fact checker for something else you’ve said.