[HN Gopher] Adaptive social networks promote the wisdom of crowds
___________________________________________________________________
Adaptive social networks promote the wisdom of crowds
Author : tosh
Score : 48 points
Date : 2021-07-09 11:05 UTC (11 hours ago)
(HTM) web link (www.pnas.org)
(TXT) w3m dump (www.pnas.org)
| lonk11 wrote:
| > In this paper, we test the hypothesis that adaptive influence
| networks may be central to collective human intelligence with two
| preconditions: feedback and network plasticity
|
| In the paper they had the participants of the experiment to
| manually pick who they want to follow. But what if the system
| connected them automatically to high signal-to-noise individuals
| based on feedback alone?
|
| I've been working on something like this with my hobby project
| https://linklonk.com - an information network where the
| connections between you and other users are determined by your
| ratings of content.
|
| When you upvote an article - you connect to other users who
| upvoted that article. When you downvote - your connection to
| those who upvoted it becomes weaker. That way the strength of
| your connection to other users captures the signal-to-noise
| ration of those users for you.
|
| The stronger you are connected to someone - the higher their
| other upvoted items are ranked on the "For you" page.
|
| For example, I upvoted this paper on LinkLonk:
| https://linklonk.com/item/6534389451373608960 If you also upvote
| it then you will get connected to me and will see more of my
| recommendations on the main page. The next user who upvotes it
| will connect to me and to you, etc.
|
| Since you know that your content ratings have direct effect on
| what content your future self will see, you are incentivized to
| think whether each piece of content that you just consumed was
| truly worth your time. This kind of retrospective thinking is
| missing when we hit upvote/retweet/like in the existing social
| systems.
|
| My project is in a very early stage and suggestions/ideas are
| welcome.
| high_byte wrote:
| sounds polarizing. I wouldn't decrease your connection upon
| downvoting, but I would increase connections with others who
| downvoted. I would only decrease connection over time without
| similar ratings.
| lonk11 wrote:
| The purpose of the downvote is for you to say what content
| wasted your time. Those who brought that content to you
| deserve to lose your attention so they do not waste your time
| in the future, do they not? That's why LinkLonk decreases
| your connection to those who upvoted it. It also displays a
| popup saying "You will see less content from N users and M
| feeds that upvoted this" to explain how the downvote button
| works.
|
| LinkLonk also increases your "downvote connection" to other
| who downvoted that item as you are suggesting. A "downvote
| connection" is how much weight the other person's downvote
| has for you. That is, it captures how good their past
| downvotes were for you and how much their future can be
| trusted. So there are two kinds of connections:
|
| - Upvote connection - gives others ability to promote/curate
| good content for you.
|
| - Downvote connection - how much others can bury/moderate bad
| content.
|
| And as you are also suggesting, the connection is decreased
| over time without similar ratings. Each time someone you are
| connected to upvotes something your connection to them
| becomes slightly weaker. So if you ignore content from a
| user/RSS feed then it will have lower ranking for you over
| time. So in practice the downvote button should not be used
| much at all.
| Palmik wrote:
| Doesn't this create bubbles of like minded people? Do you
| believe that to be desirable?
| forgotmypw17 wrote:
| Not necessarily bubbles, could be just supportive,
| encouraging communities.
| lonk11 wrote:
| You are right, LinkLonk is a filter bubble. The difference
| from other systems (e.g., algorithmic feeds powered by
| machine learning that optimize for "engagement") that exhibit
| the filter bubble dynamics is that LinkLonk puts all of the
| control into the hands of the user. The user is responsible
| for the content they upvote which directly determines who
| they get content from in the future.
|
| In a sense this is similar to how users of RSS feed readers
| control which feeds they subscribe to. They are responsible
| for the content they consume. What LinkLonk adds is a
| transparent layer of automation that helps you
| subscribe/unsubscribe based on your content ratings.
|
| My hope is that LinkLonk will help people get more informed,
| but I cannot be sure. The project is a live experiment to
| find out if this is right system of incentives.
| elevaet wrote:
| This sounds like a really interesting experiment. I hope you
| share your results here on HN.
|
| This approach reminds me a bit of the saying in neuroscience
| "if it fires together, it wires together"
| lonk11 wrote:
| You can think of every user as a neuron where upvotes on the
| same items strengthen their connections.
|
| Though there is one important bit of asymmetry: you connect
| to those who upvoted that item *before* you. That way people
| who recognize useful information earlier earn more trust. In
| a sense this is a "proof of work" - to recognize valuable
| content before it becomes popular.
|
| My hope is that this asymmetrical nature of connections will
| get less informed people to connect to more informed people.
| Which is the opposite of the echo-chamber effect - when less
| informed people are connected to similarly less informed
| peers.
|
| And yes, I'm slowly preparing to do a "Show NH" for LinkLonk.
| But I probably need to grow the number of active users a bit
| before I do that, otherwise the "Shown HN" will bring a lot
| of clicks that will just bounce.
| elevaet wrote:
| So let me get this straight - you become connected to those
| who upvoted the item _before_ you, but not to those who
| upvoted _after_ , right? So this incentivizes the avant
| garde, rather than the promoters of stuff that's already
| going viral. Very interesting.
| lonk11 wrote:
| Correct. This also provides some protection from gaming
| of the system. You can't simply upvote a bunch of popular
| items to get other people who already upvoted it become
| connected to you.
|
| You have to be good at prediction what people will like
| in order to get them connected to you. And this is what a
| good curator does.
| elevaet wrote:
| I wonder how this dynamic would affect virality in
| relation to truthfullness?
|
| As in, will this dynamic tend to reward the spread of
| "edgy" fake news more than current network paradigms
| already do, or will it tend to slow its spread, or will
| it be neutral on that axis? Seems like a hypothesis that
| would need to be tested, but maybe you have a hunch or
| insight on the matter.
| guscost wrote:
| > Each task consisted of estimating the correlation of a scatter
| plot
| elevaet wrote:
| This is great. But how do we build social media networks that
| reinforce centrality/influence of accurate truth-tellers, and
| penalize sensationalism, extremism etc.?
|
| This seems like one of the big hard problems of our era. If we
| can solve it, maybe this is a phase transition humanity is going
| through where we begin to operate on a higher level of
| complexity, similar to when life transitioned to multicellular.
| hsndmoose wrote:
| Would this not require quantifying real, objective truth? How
| does one compute truth without relying on human input? (Which
| instead trends towards truthiness/the "feeling" of whether
| something is true.)
|
| I am not being dismissive, I genuinely would like to know.
| theropost wrote:
| Its a very big question - we almost need some sort of level
| of detail about the commenter, to understand their expertise,
| backgrounds, experiences, abilities, etc - but once again,
| how would you quantify it? For all topics, not everyone's
| voice should be considered equal
| thih9 wrote:
| I'd guess it depends on the topic and scenario.
|
| > Would this not require quantifying real, objective truth?
|
| Could you give an example?
| hsndmoose wrote:
| I cannot. It's a philosophical question of what is truth. I
| don't know if it is possible for a human to obtain 100%
| objective truth about something.
| prox wrote:
| It's called a library in my idea of it. The sum of human
| knowledge, curated by experts in every field. I don't think
| we can compute that last bit. We may not have to.
| elevaet wrote:
| I think this is one of "the big questions" right now.
|
| Philosophy tells us that you can't compute truth without
| relying on axioms. But computer science tells us that if we
| accept basic axioms, the computation of truth quickly becomes
| orders-of-magnitude too complex to compute.
|
| I suspect that this all leads us to needing to rely on coarse
| human input as "axioms".. which of course leads to the issue
| of which humans do we rely on as stalwarts of the truth? It's
| a bit of a chicken-and-egg problem.
|
| My hope is that studies like these will tease out the nuances
| of networks so that we can engineer networks to nudge the
| nodes of better truth-telling to more centrality in the
| networks, and that gradually we'll master the art of building
| intelligent networks. After all, biology did it with the
| human brain.
| acituan wrote:
| > Philosophy tells us that you can't compute truth without
| relying on axioms.
|
| Philosophy tells no such thing. It is not the province of
| philosophy to tell us the final word on what is what, and
| without corresponding it to any empirical exploration,
| asserting such a claim is mere dogmatism.
|
| Computationalist model of "truth" (by which I think you
| mean reality) is dying. Embodied-embedded cognition offers
| an alternative in which your intelligent system has to be
| deeply embedded within all the other networks it has to
| interact with, and its adaptivity and constraints define it
| more than anything. There is no making an intelligent
| network in a test tube (talking about general
| intelligence).
|
| > After all, biology did it with the human brain.
|
| Biology might have put the required machinery, but
| machinery by no means is a guarantee that it will be
| neither intelligent nor adaptive. You _could_ "engineer"
| your own network that is your body-brain to get better at
| conforming to reality, which is called self-transcendence
| and cultivating wisdom, and arguably the same principles
| would work for our social networks, artificial networks and
| us alike.
|
| But going back to the notion of embeddedness, can a social
| network that will ultimately aim to conform to the norm of
| _making more money_ be wiser? Can a wiser social network
| really out-survive a dumber one? Isn 't both going to be
| ultimately embedded in the collective intelligence that is
| our economy? Therefore both will be constrained by the
| limits of the intelligence/wisdom of the economy, and
| unless there is a bunch of benevolent rich that will
| implement the engineered wiser social network and gift it
| to the humanity and get humanity to actually use it, there
| is no such place, i.e it is a utopia.
| nosuchthing wrote:
| No - consider it mostly a design and infrastructure problem.
|
| When looking at social media, it's part public forum that
| needs some type of discovery/filter mechanism, and part a
| tool for individual users and community to communicate and
| collaborate.
|
| The barrier in current social media networks is largely
| skewed towards manipulative design that optimizes towards
| datamining and addictive gamified systems and interfaces.
|
| Sure you could try to build a social network for open science
| and peer reviews of research projects, but the bar is set so
| low right now that any improvements to interfaces that
| facilitate a more comprehensive search/discover/filter system
| on datasets will be a massive improvement over now.
|
| Information needs to be discoverable, but people need to be
| free from propaganda.
| meowkit wrote:
| You can't quantity or even know "objective truth". We can get
| really close for somethings, but knowing objective truth is
| akin to being a god. At the end of the day, everything is a
| model relying on some axioms.
|
| Inter subjective truth, however, can be reached and is what
| we rely on most (a dollar bill is a fancy piece of cloth, but
| we all agree its worth a dollar). It is reached through
| consensus making.
|
| Gathering a consensus is traditionally done through
| government or hierarchy, ultimately leading up to a single
| human or single groups input as "truth". This method has
| continually begun to disintegrate as communication tech gets
| better (printing press -> mobile phone internet).
|
| So the solution, to me, is to create consensus systems that
| rely on the input of many - use the law of large numbers,
| economic incentives, and the kaleidoscope of subjective
| truths to reach the most accurate objective truth we can.
| simplify wrote:
| It's true that society uses consensus as a proxy for truth.
| Even when scientists make a new discovery, it isn't
| considered "truth" until they convince the community -
| sometimes even taking decades!
|
| Sadly, this consensus can be manufactured by those in
| power. Censorship helps to a surprising degree, for
| example. Social media sock puppets, astroturfing, bribery,
| the list goes on.
|
| How do we fight against manufactured consent? Is it even
| possible at this point?
| elevaet wrote:
| Yeah this is the crux of it isn't it? And it's not just
| the problem of manufactured consent either, there is also
| the problem of mistaken consent that grows organically
| out of human frailties like our cognitive biases and
| appetites for drama.
| hsndmoose wrote:
| Yes thank you. Between this and your other comment to me
| in this thread I think you've really gotten to the heart
| of it. I appreciate you putting into concise words what
| was rumbling around in my head when I first asked the
| question.
| elevaet wrote:
| It's a joy to hear that someone else has made sense out
| of what has been rumbling around in my own head a lot
| lately.
| kradroy wrote:
| Maybe it's not necessary to define truth for this.
| Considering metrics that you want to influence might be a
| better way - hate crime arrests in locales, negative/divisive
| message content, donations/volunteering for positive causes,
| etc. But I'm a pessimist and I think moving these metrics in
| the right direction would adversely affect the $$$ metric
| that shareholders care about, so it's not going to happen.
| swiley wrote:
| So your depending on money and the law to define truth
| then.
| war1025 wrote:
| > where we begin to operate on a higher level of complexity,
| similar to when life transitioned to multicellular.
|
| I've thought about this exact thing before and something to
| keep in mind is that there will be a split where part of the
| group consents to being absorbed into the mega-organism, and
| part will stay individuals.
|
| Humanity won't move in unison. If it happens, part of the group
| will stay behind, just the same as we still have single celled
| organisms.
| gorwell wrote:
| And a way to mitigate or dissolve Twitter mobs. I imagine pile
| ons could be identified?
| justshowpost wrote:
| > networks that reinforce centrality/influence of accurate
| truth-tellers, and penalize sensationalism, extremism etc.
|
| I'm pretty sure CCP already built several such networks for us.
| zwaps wrote:
| the study itself is a bit "pnas'y". In fact, a lot of work
| studies information aggregation in different kinds of settings.
|
| As you, I think, suspect, things can also easily go the other
| way. Well known is the fact that extensive information /
| interaction networks can decrease diversity (a general fact of
| averaging mechanisms in network settings that holds even in
| Games on networks). A decrease in diversity in information
| aggregation is often detrimental (for example through
| correlation of errors, canceling out one mechanism of wisdom of
| crowds, or through conformism).
|
| Further, wisdom of crowds only exists (relative to individual
| guesses) if in growing networks the opinions are not controlled
| by influential groups or echo chambers (Jackson did work on
| this). This of course, is what ends up happening a lot, in part
| due due to aforementioned factors.
|
| Since the proposed mechanism relies on weighting accurate
| individuals higher, it is even more susceptible to such biases
| than just wisdom of crowds in general.
| AstralStorm wrote:
| Funny. The rating mechanism makes this equivalent to a
| reputation based prediction market.
|
| These can also fail and are gameable.
|
| The key assumptions are that people form social networks
| according to DeGroot process, which is not actually true.
| People tend to maximize objective gain not
| information/influence gain - usually objective is not stated
| and there are various ones, from knowledge, entertainment,
| desirability, clout/influence and power, even financial ones.
| Unfortunately, this does result in a divergent scenario.
|
| As for feedback, current social networks have simplistic
| "like" system that only weights engagement in a hidden matter
| and generic desirability points with partial feedback
| scenario. Trying to get full feedback would require someone
| to be able to read likes of every user, not feasible. What
| could be feasible is guessing how "likeable" a post is -
| gaming the system.
|
| That said, the paper is rather high quality considering its
| limitations. Interesting that essentially listening to a
| moderate number of curated but not isolated experts had best
| results. (Top 5 interesting enough, this being ~33% of
| group.)
|
| The trick is in identifying these top performers in a real
| multiobjective scenario, and whether the selection strategy
| still works.
| sysadm1n wrote:
| > These can also fail and are gameable
|
| Gameable perhaps, but in proper meritocratic[0] systems,
| voting rings etc would be detected and the perpetrators
| hopefully banned and they would also lose the ability to
| participate again at a later stage, giving an incentive to
| play fairly next time.
|
| [0] https://en.wikipedia.org/wiki/Meritocracy
| forgotmypw17 wrote:
| >"Trying to get full feedback would require someone to be
| able to read likes of every user, not feasible."
|
| I think that with a fully transparent system this is not
| only feasible, but easy.
___________________________________________________________________
(page generated 2021-07-09 23:02 UTC)