Request for Comment: Media influence score for signalling

#21

:wave: I’m from Colony! let me continue reading the thread, but at first blush yes, reputation is pretty hard to put into a single metric.

Colony’s rep system is tied specifically to events that happen on-chain, and is more or less a record of everything any particular account has done within a colony, compressed down to a single number. But it’s directly tied to activity that happens in the_colony_ smart contracts, and only that. While we’re looking ahead distantly at the possibility of a more generalized rep system, it’d be quite a challenge to account for everything and we’ve got enough challenge in front of us with a mainnet launch targeted this year :slight_smile:

6 Likes
#22

Cool idea, @MaxSemenchuk! Thanks to everyone for posting such thought-provoking comments. I’m glad @maciek is making available data more organized. Anyone can measure time or clicks related to watching, listening, reading, doing work, completing tasks, sharing ideas, looking for truth, etc. Reputation is multidimensional. We should intelligently use the power of social capital to facilitate coordination.

Think about Ken Arrow, Nobel Prize recipient who argued, “It seems to make no sense to add the utility of one individual, a psychic magnitude in his mind, with the utility of another individual” (1951). However, social scientists often associate Arrow’s “general possibility theorem,” with Condorcet’s “paradox of voting” (1785).

Let’s abstract reputation from any single token balance, giving potential value to the multitude of actions we intentionally do in public. Don’t sell your influence. Simply understanding attention will better inform us to make decisions that provide justice and liberty for everyone. Bribery is less likely to happen if we have greater transparency & accountability. Fake traffic is not a new problem, and tokenizing complex signals might actually help. Dimensional voting can be weighted by a specific reputation. Time (a.k.a. validated human proof of energy/work) levels the playing field.

It’s possible for token engineers to create hierarchical systems with infinite layers of structured information, i.e blockchains account for everything. We’re objectively representing subjectively-controlled actions/events in order to gain unprecedented common knowledge.

Have a look: ERC 888 #pluralism

FEM discussion thread: MultiDimensional Tokenization #disposables

2 Likes
#23

I think the media influencer score is definitely a really bad idea as the only or even the primary source of signaling information. However, I think the intention is to aggregate many signals rather than rely on any one. The goal being to save time and mental effort by organizing unstructured data, not to put any one piece of data or signal into an elevated position of legitimacy.

Right now we look to social media to gauge sentiment within the community but on an ad hoc basis that is incredibly imprecise and inefficient. I personally would find something like this list that distills that into a clear signal on an issue a fairly practical solution.

I think the fact that this algorithm is closed source and therefore not easily validated or replicated by the community is a huge barrier to legitimacy, and I get the sense that its probably best to focus on on-chain signals first.

4 Likes
#24

I understand that is alpha and experimental but I think it’s fundamentally flawed. The fundamental problem is that by assigning a concrete influence score to influencers you are modifying the very thing you are trying to observe. You create strong incentives for identities to react in ways to specifically modify and monetize the influencer metric.

At the moment we’re tracking Twitter, but soon we will combine it with many other sources of data.

Already this has a serious issue. Twitter is not a public utility. It is not an open platform, it has no transparent governance, and perversely incentivizes gaining influence over audience. Twitter can silently suppress or even censor certain topics and people. What would prevent Twitter itself from doing this to promote its own agenda for an influence score?

The other issue is Twitter is not reflective of greater society. It’s very much loved by mainstream and social media users but is not so popular outside of that. Additionally, Twitter has serious issues with both bots and extensive use of block-chains (where you automatically block people based on who they follow, not their interactions with you can lead to severely distorted echo chambers that leads to endless tempests in teapots that blow over within 24 hours. I think even Facebook or Bitcoin Talk) would be fairer.

The influence scores distribution is subject to Power Laws. The people who have the most voting power, e.g. Vitalik Buterin, Vlad Zamfir, Joseph Lubin or Gavin Wood are unlikely to be corrupted.

The way you prevent corruption is by removing temptation, not by designating moral guardians.

If you want to assign influence score, I’d stick to on-chain solutions that are openly verifiable. I do think an influencer score causes more issues than it solves.

Besides, influencers already have influence. That’s why they’re influencers! If they want people to vote a certain way, they can influence them to do so and that would be a far more accurate measure of their social influence than any metric we could come up with.

3 Likes
#25

This is well said, and is our intention.

Here is a breakdown of the signals thus far. Only the last is off-chain, and the influencer list is a subgroup of that.

  • Gas Signaling
  • Coin Signaling
  • Hashpower Signaling
  • Node Operator Signaling
  • Influencer Stances
    • Core Devs
    • Influencer Ranking list
    • Other

Maybe the term Influencer is the problem here. The goal is to collect the current stance of people influential in the community on a specific EIP. Collecting everyone’s would lead to an overwhelming amount if information. Collecting no ones means it is difficult to see how individuals throughout the community stand on a topic. So, we are left with some means of showing a subset of everyone. Having different groups of Stances I hope will highlight that there isn’t one that is more or less valuable then the other.

One benefit of the Influencer Ranking list is that it gives a view of a consistent group of individuals across all EIPs. Yes it will change overtime, but at any one point it will be the same list of 300 individuals. This consistency I hope will give an interesting cross section on sentiment around EIPs.

One way to curtail gaming a system is to tailor the reward function appropriately. People do not game systems if it is not worth it to them individually. We will continue to work on this, and I am not so tied to it as to not throw it out all together when the time comes. I would like more concrete information through testing before I make any call like that. It is worth pursuing until it isn’t.

2 Likes
#26

I understand that is alpha and experimental but I think it’s fundamentally flawed. The fundamental problem is that by assigning a concrete influence score to influencers you are modifying the very thing you are trying to observe. You create strong incentives for identities to react in ways to specifically modify and monetize the influencer metric.

The same argument could be made about impact scores that are used to measure the influence of academic papers. Does this mean that you think they should be abandoned?

Already this has a serious issue. Twitter is not a public utility. It is not an open platform, it has no transparent governance, and perversely incentivizes gaining influence over audience. Twitter can silently suppress or even censor certain topics and people. What would prevent Twitter itself from doing this to promote its own agenda for an influence score?

We share these concerns about using Twitter. That’s the reason this is only a starting point and our objective is to start using multiple sources of data as soon as possible.

I’d rather use a decentralized source, but there does not seem to be a better alternative at this point.

The other issue is Twitter is not reflective of greater society. It’s very much loved by mainstream and social media users but is not so popular outside of that.

It’s not used by the “greater society”, but it has a high penetration within the crypto community. And that’s enough to start with.

We believe that this is the first time a score like this can be built. And this is due to the peculiar nature of the crypto ecosystem. It’s the first ‘global first, local second’ ecosystem. This means that significantly more communication happens online within this community, compared to say, a startup ecosystem.

This leads us to believe that it might be possible to collect enough data to assess influence probabilistically.

We believe that crypto will “eat the world”. If this is the case, then the solutions that work in this community will spill over to other sectors.

Additionally, Twitter has serious issues with both bots and extensive use of block-chains (where you automatically block people based on who they follow, not their interactions with you can lead to severely distorted echo chambers that leads to endless tempests in teapots that blow over within 24 hours. I think even Facebook or Bitcoin Talk) would be fairer.

Twitter has many issues. However, the examples you brought up have not been problematic so far in our experience.

I don’t see how Facebook could be a better source of data for this purpose. We don’t exclude the possibility of using BitcoinTalk.

The way you prevent corruption is by removing temptation, not by designating moral guardians.

I think I was not clear in my comment about corruption. We don’t claim to solve this problem.

What I intended to point out was that people with strong reputation have more at stake, therefore should be less likely to accept bribes (or at least the cost of bribery should increase).

If you want to assign influence score, I’d stick to on-chain solutions that are openly verifiable. I do think an influencer score causes more issues than it solves.

I’m not aware of any on-chain solution that solves the problem we’re addressing. I’d love to change my mind on this, so feel free to educate me.

Besides, our goal is to open-source our algorithm and decentralize its governance. We just believe that starting in a centralized model and decentralizing once we prove that it can work increases chances of success.

Besides, influencers already have influence. That’s why they’re influencers! If they want people to vote a certain way, they can influence them to do so and that would be a far more accurate measure of their social influence than any metric we could come up with.

Good point. One can even think of this as a form of “voting with attention”. Many governance models run into the problem that few voters participate. Even with delegated voting. In this case, delegation happens naturally and is predicated on one’s self-interest.

E.g. let’s say there is a group of physicists. Each one of them has limited attention. They can dedicate it to their own thinking or reading each other’s papers. The optimal strategy is to spend some of it on your own work and dedicate some to keeping track of important ideas produced by others.

This is likely going to lead to the following outcome: those regarded as producing the highest quality thinking will accumulate disproportionally more attention than those producing low-quality thinking.

This distribution is predicated on self-interest – i.e. I’m reading papers that I think will benefit my work.

Arguably, the same dynamics applies to other groups. Including open-source developers.

Centralized networks (e.g. corporations, governments, armies), can make quick decisions because it’s clear who holds power*. The problem with decentralized networks is that there is nobody who does.

There are people who have influence, however. There is just no way (yet) of telling who they are and how much influence they have.

Our hypothesis is that if there is a reliable way of quantifying this influence it opens up possibilities for removing this weakness. This would make decentralized networks even more compelling.

*Power is influence with agency.

1 Like
#27

this is exactly the point: we have talked about the influencer’s list, but we should also talk what we have the list for? Should community really agree on opinions? IMHO, not!

Anybody can build own opinion based on “influencers” opinions, but there should not be any agreement about it. An influencer’s opinion should help to build own opinion, not substitute it.

1 Like
#28

Influencers are fine. In fact, they’re probably impossible not to remove. My point is that over time they calcify and maintain their influence by habit more than quality opinions. My suggestion was that we randomly remove and randomly add new voices by design.

2 Likes
#29

Just because your intent is to simply “make the data readily available, but not use it as a primary mechanism for decision making” doesn’t mean people won’t use it as a primary means of decision making.

As is, I spend an inordinate amount of time undoing the Bitcoin propaganda that miner desires matter, if you create a metric for “miner desires” it makes my (self assigned) job of reminding people of this fact significantly harder. They will just point at the metrics and checkout.

You can see this in society today, people point at statistics from academic papers in political debates all the time without any consideration for context, nuance, etc. By making bad data available in a nice format, many people will prefer to use that as a primary decider rather than understanding the issue.

We should be working on making it easier for people to understand the arguments and nuance, not making it easier for them to mentally check out.

6 Likes
#30

I wrote an open letter to the platform hivedotone after they accused me and others of harassing them. This is where I draw the line. This accusation is unacceptable and @maciek & co have no idea what harassment means. Feel free to ping me to add your opinions.

1 Like