jnakayama

Great read!

My exposure to Taiwan so far is very limited (9 hour visit to Taipei when I was a teenager ~15 years ago), but I remember it leaving a profound impression on me.

I think what struck me the most was how genuinely friendly people were (as in warm, not just polite). One example (of many) I observed: There was a homeless man sitting on the street selling chewing gum. A very attractive couple passed him and the woman stopped to buy a chewing gum from him. But instead of just giving him the money, she sat down next to him, negotiated the price for a bit, they had a laugh, and eventually she bought the chewing gum. I found it to be a truly beautiful display of respect for the guy's dignity.

All good questions! We are discussing them regularly.

We take an adversarial approach to the protocol and constantly try to come up with ways it might be gamed. We want to design it in a way that achieves something like this: https://xkcd.com/810/

Some points regarding your questions:

The protocol will be fully open source so nobody can gain an advantage through information asymmetry.

We choose the best replies by measuring how much a reply "changes minds" (difference in voting behavior on a post given users have also voted on that reply). We adjust for biases using a technique called bridging-based ranking [1, 2]. The algorithm chooses the reply that had the highest impact on voting behavior after adjusting for biases.

We do realize that truth is a loaded term, so to elicit "truthful" posts and replies, we incentivize honesty with a technique called the Bayesian Truth Serum [3]. This will be linked with a reputation system that determines a user's impact.

If you spin up a bunch of bots, they will have low reputation in the beginning and therefore little impact. If you coordinate people (or bots) to vote manipulatively, their impact will be minimized by the bias correction because their votes are correlated. Our hope is that it would take considerable effort to build reputation on many accounts (by being honest and helpful over an extended period of time) and maintaining an attack over the lifecycle of a discussion would be infeasible because it works against the incentive structure and can at any point be exposed.

This is just some of our current thinking, if you can think of attack scenarios or weaknesses in our thinking, we'd be interested in your thoughts!

[1] https://jonathanwarden.com/understanding-community-notes/

[2] https://vitalik.eth.limo/general/2023/08/16/communitynotes.h...

[3] https://nel.mit.edu/bayesian-truth-serum/

Hey, member of the team here

Thanks for the feedback!

We'll be launching something soon, if you're interested, you can keep up-to-date here: https://social-protocols.org/social-network/

The algorithm is explained here: https://social-protocols.org/global-brain/ This is a living document, we try to keep it up-to-date, but not all of it currently is.

We provide some demonstrations of how the algorithm evaluates certain scenarios here: https://social-protocols.org/GlobalBrain.jl

If you're interested in investigating timing effects: We collected a dataset last year where we took a snapshot of the newest 1500 stories on HN every minute for several months which should contain the information required. Feel free to play around with it and get in touch with us if you find something interesting!

[1] Dataset: https://osf.io/bnysw/

[2] Exploratory analyses: https://github.com/social-protocols/hacker-news-data

We played around with customization options like that (URL parameters etc.), but ultimately decided against it. The reasoning behind it was that a lack of personalization might be a feature not a bug for a news aggregator like HN. One issue that arises with personalization is that it is detrimental to a sense of shared experience and we thought that the global frontpage might be a distinct reason for the sense of community on HN.

This issue was also discussed previously on HN: - https://news.ycombinator.com/item?id=31375092

Interesting! I think neuropsychological hazard according to this typology sounds fitting. Reminds me a little bit of the scenario of what would happen if you knew the exact date and time of your death.

Maybe to add to what might be the rationale to have an op like this: I've recently been on a call where someone brought up a metric called quality-adjusted life years (something like: "Would I rather live another year relatively painless or five more years, but in constant excruciating pain?"). As far as I know, this is mostly used in insurance calculations (which struck me as a bit distasteful, but I didn't look into it deeply enough to make a fair judgement), but it's a good mental model for what people might use in their reasoning in such a situation. Maybe for the patients that decided to have the op, the knowledge of their condition was bad enough to substantially reduce their quality of life and hence lead them to accept far higher risks to alleviate this.

For the second dilemma: Maybe simply knowing about the condition running in your family is an information hazard too. For someone very prone to worrying, knowing for sure that they have the condition might also come as a strange kind of relief.

As you mention, the services can already be provided at a fraction of the cost. I think the key problem is the trust/reputation that established publishers provide. There is massive demand for "trusted publication space" and the price develops accordingly.

I've been thinking about how to remove trust from this process and haven't come up with anything good yet, although I desperately want to believe that it's possible...

I do think that a good solution would be able to disrupt the current system though, the current scientific performance metrics (citation count, H-index, etc.) are almost entirely based on "real-estate" in trusted publication space (i.e. they only factor in publications in established sources) and they are extremely prone to Goodhart's law, so they become less and less useful to navigate the literature.

Completely agree! I had the same thought when looking into elementary cellular automata, they seem so inoccuous, but there are still so many open questions. There is even a prize for solving three particular ones: https://www.rule30prize.org/