Topic: New DT algo - why 100 earned merits?? and not objective metric Activity or Rank? (Read 470 times)

Thanks for the great reply.

I want to jump down to the specific part about merit vs activity.  I would like later after we get that part debated a little more then to go back to the start of your message and see if we can examine those parts - - I think there are some  things that you may not have considered or maybe I believe they should be considered -- you may then demonstrate then need not be.

First though I really would like to do the merit vs activity thing.

"Activity is obviously objective, but it's easily gamed.
Any system relying on such a metric will be highly prone to Sybil attacks.
I could simply create a ton of sock puppets and post random crap to create an army of accounts which matches any threshold of Activity.
All it'll take is some time and very little computational power."

I agree this is true but perhaps maybe we need to realise it will take some time and effort.

4 years and some very sneaky bots to not be detected in that time. However I accept your point it can be abused if we use activity alone.

Could we prevent this by requiring chunks of reasonable merit as years go by.


So then we can compare merit.

Merit - seems a very unfair system with huge variance and one can easily abused or traded amongst a tight group resulting in score that is of a very low value in terms of meaning . It has been useful in that it  can prevent bots and account farmers ranking their accounts up. It was obviously designed for that reason.

I don't understand your point that reducing the sample size is  useful? I would have though that is only in cases where the resulting group are provably or even reasonably more suitable for the position. The criteria is surely the most important part.

I mean you can reduce the sample size to concentrate individuals with negative attributes with regard the intended purpose if the criteria is not well thought out.


Also surely since there are only around 100 persons with 250 earned merits who seem to have derived perhaps 70-80% of their 250 merits from their own pool of the same 100 (whatever the reason for that) then to me that actually suggests this could be a dangerous thing to do.

Requiring 2 from 100  seems to be placing VERY high value on those members that I can see no reason for at all and especially in the context of being trustworthy.

I had suggested previously the 2 systems of control are kept separate since that would seem to seem sensible. However this seems to do the exact opposite and ties the merit source system controllers to high value requirements of the DT systems of control.

I'm not sure why earned merit seems to be coming up so much. It seems to suggest that members who made posts before merit system were of zero merit worth. A lot of legends who I like reading have consistently made excellent posts for years.

The new DT algo looks very good except for that one minor point ( i mean i say minor in that it would be easy to change... if left as it is I would probably say it could be a major weakness until a far greater number of people have 250 ....earned merits)

The above are just my current opinions. Not something I have a huge

If you do that, inter-subjectivity is what you get.
"Strict criteria" is just another word for "arbitrary definitions".
Any group selected based solely on a set of arbitrary definitions is random in all other aspects.

Which is what you want in the first place, btw.
A random group of people.

So, why have "criteria" in the first place?
1. to reduce the number of people / your sample size.
2. to reduce the possibility of abuse.

Whatever your random sample of people votes for is inter-subjective, but definitely not objective.


And I think you're utterly mistaken.
I can't think of a single example of a trust system, be it tripadvisor, ebay-ratings, yelp, PGP WOT, whatever, where subjectivity is even discouraged.

There's a simple reason for that: it's simply considered impossible.
Game Theory tells us that basically any game will have a set (in some cases it may be an empty set) of Nash equilibria and those can be considered the potential endpoints of any web of trust.
None of these equilibria are objectively "right" in any meaningful way.


I guess we jumped a little.

Activity is obviously objective, but it's easily gamed.
Any system relying on such a metric will be highly prone to Sybil attacks.
I could simply create a ton of sock puppets and post random crap to create an army of accounts which matches any threshold of Activity.
All it'll take is some time and very little computational power.

Merit is much less easily gamed, though.*
Of course, merit may not be "fair", but that's not the point.
It's a good arbitrary criterion for reducing our sample size in the first place.
And that's the only thing we need. See above.

* sure enough, it's conceivable that merit could be gamed as well, but obviously not to the same extent.


You can't open it to subjectivity, because it's subjective by definition, anyway

Abuse is a problem, and I personally believe that merit is the best option for a criterion to counter that.
Activity is certainly another option, albeit an inferior one, if you ask me.


10 merits.
That's all you need.

Of course, 250 merit will be needed by two proponents for any person on DT1, but those two will not be too hard to come by.


How are they discounted?
They can still be voted onto DT1.
They can still vote for DT1, if only they ever make a measly 10 merits.

I simply fail to see the issue.

Granted, the required 2 250-merit-voters for you to be on DT1 will in some cases be hard to come by.
Then again, I would guess that there'd be a reason for that.


You're simply mistaken about what the 250-merit-requirement means.
The votes of 250ers don't count any more than the votes of 10ers.


In simple words:
Users with less than 10 earned merit are excluded from voting for DT1.
That's all.

Earned Merit is subjective, it is abused, it is not related to trust. You will exclude a lot of the eldest and some of the most trustworthy devs/legends and some of the smartest people on the board who have hardly any merit compared to some. Some of these excluded with have held 10000's or more dollars for their communities over the years.

Activity 1500 is a minimum of 4 years I think possibly a bit less of consistent posting.  You will have 4 years of posting/trades to make a determination of whether you trust them. They can not game or abuse the system. Other than making shit posts/ bots etc but then you will see that when you review and I do not think anyone will get from now to 1500 activity from now using bots or pure shit posts.

Current TRUST scores (if we mean dt as of now) are pehaps the reason we are trying to change DT system. So better if you want to gage via trust examine the entire history of their trust and check the fine print.

No, it's definitely not easy to earn merit, especially if you're a newcomer who doesn't make good posts or doesn't contribute something to the forum, like scam-busting or that kind of thing.  I'm not so sure it's a yardstick of trust by any means, but activity?  That just takes time.  That's it, and that's how a lot of Legendary members ranked up, including myself.  I ranked up to Legendary something like a week or two before the merit system was rolled out last January. 

I'm conflicted about a merit requirement here, but I can understand why it's considered important--oddly enough it seems like scammers and other undesirables don't earn a lot of merits unless there's shenanigans involved (and that definitely does happen).  But some of the most trusted members do have a ton of merits to their name.  You wouldn't think there would be a correlation, but there just might be.  I also don't really like to see lower-ranked, unproven members on DT2 and getting a certain amount of merits can be an effective barrier to that.

On the other hand, if we're talking about DT selection then a person's trust score should rule over everything else.

@qwk

I had another read of our interaction on this thread yesterday.

Looked at Intersubjectivity. I was not familiar with the term (thanks) however I am not sure why that would be considered when  it is possible to set some strict criteria to start with. why go through all that subjective interaction to eventually (if you don't even allow for malicious  abuse, ) arrive at a group generated and broadly accepted common sense approach.
 
I think that is why most trust systems seek to extract and remove all subjectivity. That seems the common approach. There are many complex ways of dealing with subjectivity that is not easy to extract. So adding any additional subjectivity into a system that need not be there would seem strange when there is no advantage and possibly many disadvantages.

I'm not sure if we jumped to talking about the trust system as a whole perhaps than just why bring merit into a trust system rather than use an objective score such as activity. Rank was objective I guess in now subjective.

The new trust system looks much better in many ways. I just feel linking it to merit could open it up to subjectivity/abuse you do not want and centralise it quite seriously.  250 earned merits? aside from totally discounting legends work for perhaps 8yrs or more (certainly to me that post history counts for more than some subjective merits if i am talking trust)

250 earned merits? does that place the trust system in the hands of  0.065% of the active posters here? based on their own subjective opinions of their own posts? because once you take their own subjective scores of their own posts away then...... how many people have 250 earned merits? and even then what is that to do with trust? how can people trust them more because they believe they make posts that should get merits they themselves give out?

Is that correct or not?




 

Oh you seem like a far smarter guy that I initially thought. Okay I can learn something new here perhaps this is what I am after.. i am not familiar with inter subjectivity... I will read this reply and think more about it

No. A true inter-subjective system won't require any "goal" of lowered subjectivity.
If anything, forcibly trying to work toward such a goal brings with it the risk of peer pressure, which is detrimental to the desired effect of "swarm intelligence" in a true inter-subjective system.

In short: given enough subjective opinions from a preferably random selection of people will result in something which is highly subjective in every single detail, but will be (more or less) indistinguishable from objectivity.

Of course, true randomness is a standard which might be desirable but can never be reached.

https://xkcd.com/1153/

So therefore you must agree that in a trust system the goal would be to work to zero subjectivity where it then becomes trustless? If you say it is not possible to reach zero I agree to but to increase variance at any point seems strange if there is no need.

Is this correct?

I agree with you but I was confused by your reply.... should have just said well cryptohunter at zero (which you want) you need no trust it is trustless but yes you want to reduce subjectivity where possible? or are you not saying that anyway?

To me injecting subjectivity in at any point is a move in the wrong direction if working toward reducing the level of trust?

I am interested in your opinion on this but if you can give me examples or analogies also this will be great. English is not my first language so analogies or examples help me greatly. If you have time.


I am actually trying to understand the reasoning. I am going out for the evening later tonight so i may not reply for some time but i will when i return and read what you have said.

It's not that I do or do not want subjectivity in any system.
It's just that the word "trust" by and of itself implies subjectivity.
If Alice "trusts" Bob, that's a subjective decision of Alice.

Let me try and apply the criteria of Karl Popper:
Any risky statement that would define the status of trust is per se not falsifiable, and therefore can not be tested in any meaningful, objective way.

If anything, trust (when we talk about trust in the form of the trust system of bct) is inter-subjective, which excludes the possibility of it being objective.

You're not meant to agree with all who are a part of DT; nor are you meant to agree with all of their trust feedback. There are a lot of contentious issues that float around the community and it would be strange if Default Trust was not representative of the community in this regard.

Mikey - I am sorry to hear that you got scammed as people know I detest scammers.

We can't rely on anecdotal evidence. It has low power or can even be misleading for the larger picture.

Another thing in favour of legends that scam is the punishment side of things. They have more to lose.

Let me say I am not claiming to be correct anyway this is simply a debate.

QWK - can you explain what you mean? why would you not want to reduced subjectivity in a trust system?

How can you disagree you want to reduce subjectivity in a trust system.... I am interested in this.  zero is the target surely or as near to 0 as possible.

Perhaps I am wrong but I can't see how you can not want to decrease subjectivity and variance. I am always willing to change my opinion though so perhaps it is more complex that I think. I await your reply

Hmm, that is why I can't seem to agree to all who were a part of the DT1 because of this reason.

Uhm, no
Trust is by its very nature something absolutely subjective.

Rank should never matter towards trust, i learned this from a personal experience, traded digital goods "bitmain coupons" with a dozen members here, only was scammed once by an almighty "Legendary member"  while i had many smooth deals with lower rank members including newbies.

i am not saying newbies are more trusted than legendary members, i am saying rank is not a solid measurement towards someone's honesty, same goes to activity , many scammers have more than 1500 activities , trust page painted in red , those will still be able to "vote" for DT members, who will do the "reading" of their feedback ?  do you think anyone has the time to cancel voters with negative trust ? i don't think so.

i think getting 100 merit is not something easy , and theymos said he might make a new rule whereby you need 2 members trusting you who has earned 250 merit each.

250 merit vs 1500 activity ! what is more accurate ? to me 250 merit is much more accurate, while this indeed can lead to some centralization, i would prefer centralization to total randomness.

for an improvement maybe a combination of your suggestion along side with the merit requirement will be better.

This is why high merit does not = high value and sometimes is a very negative value.

This is a demonstration of someone trying to stifle free speech and ideas. He could come here and give a sensible answer and debate.
But no just insists this should not be discussed.

Can someone point me to even 1 of loyceV posts that is of HIGH VALUE that is not some stats related stuff he just pulls from the servers.

I mean a post from loyceV that demonstrates original thought and insight of very high value. I mean surely someone with the highest earned merit must have 100's of very insightful and important contributions.

This is not part of my debate here but I will use this point to illustrate that high merit is not correlated at all with high value posts or large contributions to the forum. So ignoring pullling stats and presenting them in graphs etc.

LoyceV apparently has me on ignore so not sure why he is too worried about this thread.

So let's carry on debaiting the OP as I said I am not going to be derailed by personal matters for this.

This is simply a question I am asking because I would like to know the sensible and logical reason behind it and I can not ask theymos directly because he said I write in a rambling low content manner (which could be true) and does not have time to read it at the moment. So I am asking others to debate without getting upset or angry about it.

@maydayonliner

Then perhaps you bring a good point for the distant future. I though this you saying this can happen at this point.

I was talking about now. But for sure we could make it 1500 activity and extend the rank sensibly over time.

I mean this is just a discussion I am not 100% which is the better way of course. We could make a list of pro / cons and see how we go



I will not reply to this thread for a few hours and see what is said by others because I will hopefully avoid dispute with other people who wish to prevent the simple discussion of ideas....

Try to make an objective list merit vs activity +rank.

Then I come back and review and just analyse peoples thoughts. 

Intention and motivation are largely irrelevant to a debate.

Why would i wish to keep new members from DT if I could trust them more.

I am simply saying I think you could TRUST someone more if

they have been here a long time and have a long history to examine

more than

someone who could have been here much less time

also you have to realise activity is applied to every person fairly and objectively so it is impossible to game or manipulate that score.

I am relaxed, why do you think so?
I gave you an hypothetical example. After more 4 years from now you may see someone will have 1974 activities but still Newbie.

You will need at least 1 merit to have the next rank. You can stay even next ∞ years but if you can not able to manage 1 merit then you are not going to get the next rank.

It's really doesn't make sense. Activity's are not prove that you are trusted or quality members. I think your intention is exclude new face from merit network (IMO). Honestly you think that earn merit is too easy ? I don't think so. Theymos already change criteria, instead of 100 earned merit  he have required 250 merit. Its really not decent than 1500 activity ?

Merit score (especially with no criteria or mandate) can not be linked to trust.

Weight out the positives and negatives of each.

A subjective score is dangerous if you try to use those scores as if they were derived objectively and fairly. Even if merit was somehow distributed by some AI have valued each post against a set of criteria I do not think it has a huge correlation to trust.

Time does.

However let's not focus specifically on that part too much.

@mdayonliner

please relax... it is called a debate.

I am not sure what your examples are supposed to demonstrate.

Explain your meaning.

As far as I know someone with that activity can not exist at newbie rank.

Also to prevent something like that being possible (but highly unlikely)  in 5 years time we can also say 1500 activity and hero

Who do you think can make better decision?


or


By the way, theymos is thinking to have 250 merits instead of 100 I guess you missed it.