Why this post now?

In the past I’ve mentioned in passing that my primary goal working on Hive was to build a reputation and information rating system that I feel could have a dramatic impact on how we organize and make decisions, but I’ve avoided going into many details about it publicly.

There’s been a couple of reasons for my reluctance to publicly discuss a rating platform in depth in the past: 1) many people might not have seen the importance of such a system until times like now, 2) I’ve been busy with clear and pressing preliminary tasks and didn’t want to get distracted by talks about such a platform (and there are a lot of potential variations in such a platform, so those discussions could go quite long), 3) there’s many “radical” thoughts that go along with such system that could generate controversy (and controversy can also be a big distraction), and 4) I’m likely to write very long posts on this topic, as you’ll see below. So, to sum it up, I wasn’t ready to spend time on discussion/design/development of the system, so I avoided initiating public discussions and debates about it.

But I’d like to start talking about design of a rating system now, because I’m at a point where I can devote time and resources to the project, and I think recent events should be compelling a reason for many people to examine the ideas around this topic.

Originally I planned to discuss implementation details of the rating platform in this post, but just writing about some of the basic ideas that led up to the rating system resulted in an exceptionally long post, so I’ve decided to defer details to a later post, so that comments/discussions on related topics will be grouped together better. As a bonus, that may defer some of the controversy.

Side note: I’ve always wanted to call this project “hivemind” as it’s very descriptive of some aspects of it’s operation, but I’m a little hesitant to do so, because of the ease with which it could be confused with Hive’s existing hivemind software for processing blockchain information. But maybe we can work out some solution, perhaps by adding some extra words to each project.

What’s the point of an information sharing and rating system?

We are constantly receiving new information, rating the credibility of that information, and storing it away for future use. For most of us, our ability to do this well has a huge influence on how our lives proceed.

Most of this information comes through our senses (in some abstract sense, all of it does, of course, but I’m referring more to events we see/hear/perceive first hand) and most people are fairly efficient at rating the truth of information we receive this way.

Of course, we know that mental illness can impact our ability to even trust the things we directly experience, but this is a rather fundamental issue that is outside the scope of the platform I’m proposing.

The rating platform I’m proposing is aimed at mitigating a problem that we all have: how to accurately rate the information we receive through human communication.

Communication is one of humanity’s greatest abilities

Communication is one of the most powerful abilities we have as humans. Communication allows us to gain knowledge far beyond the physical constraints of our bodes and across vast stretches of time beyond what we could otherwise discover during our lifetimes by sharing the information and insights of other people, even those no longer living. It’s also critical to how we organize to work on common goals, which is also incredibly important, since it’s difficult for any of us to do much completely on their own.

But for all the undoubted benefits of communication, it does come with additional problems, compared to direct perception. As its most basic problem, people can intentionally lie about events they have directly perceived or distort the information to suit their own purposes by framing the events in a biased way, omitting relevant information, etc. They can also simply make up things without any basis in reality at all.
A second problem that occurs is that most directly perceived information isn’t shared first hand, instead it’s exchanged “peer-to-peer”. And as anyone who’s played “telephone” as a child or watched the spread of gossip can tell you, humans are far from perfect replicators of the information they receive (computers are much better at this, fortunately, or blockchain-based financial systems wouldn’t be such a good idea).

A final problem with rating information is that the data we care about goes far beyond the data that we directly perceive with our senses, so even when we see video data and assuming that we can trust that the video not been altered, we need far more data than is contained in the video itself to make judgments about what is taking place. For lack of a better term, I’ll call this “higher-level” information since it’s mostly derived from human computation on the sensory data we receive, using “conscious” thought/principles of logic/etc to arrive at conclusions/insights/etc.

I think it’s safe to say that the majority of the information we make long term plans with is based on higher-level information. And perhaps not surprisingly, it is this type of information that is most contentious because we can’t “see it with our own eyes” and that’s the information we find it safest/easiest to trust. Nonetheless, higher-level information is critical to our lives, and we can’t simply operate by only trusting the information we directly perceive.

How do we rate information today (mostly the opinions of other people)?

Before I go in details on my ideas for creating a computer-aided information rating system, we first need to look at how we rate information today, in order to reasonably discuss ways to improve the process.

There are many ways we rate the information we receive, and for some types of information and some verification methods, these rating methods can be quite reliable. In fact, there are whole branches of information science such as mathematics, where information can be “proved” to be true, generally because they don’t rely on the reliability of external information, but assume certain truths (axioms) then conclude new truths from those axioms.

But it’s actually rare that we validate information using just pure logic or other forms of math, although these can be powerful tools to assist us. For one thing, most people aren’t good at math: they aren’t good at properly framing issues in terms of mathematical models or in solving those models.

But perhaps the ultimate problem with validating/rating of information’s truth is that it can take a long time to do it, even when we could potentially do so.

So, given that none of us have an infinite amount of time to do an in-depth rating of all the communication we receive or even the capability to accurately do so due to locality and physical and mental limitations, what’s the primary way we rate information? We rely to a large extent on the opinions of other people, mixed in occasionally with logical consistency checking against other information we already believe to be true (for brevity, I’ll refer to this internal checking process as “critical thinking”).

If you take a hard look at it, it’s quite amazing how much of the information we rely on in our daily lives is the information we’ve been told is true by others, with little if any independent verification by us as individuals.

Despite the flaws in our rating of information, it generally works quite well, and has allowed us to achieve wondrous things that would never be possible if we all spent our time trying to independently verify everything we’re told.

The scientific method as a means of rating information

On the other hand, while we all can’t independently verify all information, we have found it’s important for more than one person to verify information, because of the human capacity for deception, bias, and simple miscalculation.

One of the key leaps forward in human progress was the establishment of the scientific method as a way for people to independently verify many types of information. You’ll occasionally see people who dismiss science as inherently biased, but these people have missed the whole point of science, which is to remove bias and other forms of error as much as possible from the rating of information.

But even with the scientific method as a means of independent verification, it’s not a method many of us employ directly, and certainly not for most of the information that gets communicated to us. Instead, we trust others who have claimed to perform the experiments and the people who have told us about those experiments.

Rating information by rating information sources (sources are mostly other people)

Note: for the purposes of this post, an information source isn’t just the place we originally learn about information from, as the term might imply, instead it is all the people who give their opinions on the information, which might more properly be called an “information rating source” instead of just an “information source”.

We’ve all learned from experience that we shouldn’t blindly trust the opinion of everyone equally when it comes to rating different types of information. After all, people often disagree, and logic tells us that two opposite things can’t be simultaneously true.

Similarly, we’ve also learned that it is a bad idea to just trust the opinions of one person on all topics. Now that is admittedly a slight generalization, as there are some people who are mostly willing to trust one person on just about everything, but that’s a very bad way for an adult to operate, in my opinion, because I think there is plenty of evidence that there is no one person that’s uniquely qualified to rate all information. This is how children generally start out, trusting the information they receive from their parents over everyone else, but most eventually come to the realization that the knowledge of their parents has limits.

So if we can’t trust everyone, and it’s not safe or smart to trust just one person, and we don’t have time to independently verify most of the information we get, what should we do?

We rate opinions based on motivation, consensus, familiarity, and domain expertise

For most of us, the solution is to rate the opinions of other people based on how knowledgeable they are about the topic at hand (their domain expertise), make adjustments based on how we think their motivations might affect our ability to trust their stated opinions, how their internal biases might also distort their judgment, as well at the numbers of people on both sides of any disputed information. It sounds like a lot of work, but we do it all the time, and often somewhat unconsciously.

Rating of information sources based on perceived motivation

It’s hard to overestimate the impact of perceived motivation when it comes to how we rate an information source’s reliability. We’ve learned to adjust for motivation, because we know self-interest often overrides natural honesty. It’s the same reason we’ll often trust the opinions of a product owner over the product rating given by a salesman for the product.

We can see this motivational distrust now in US politics, where much of the population can be divided into two groups that distrust not only the motivations of the leaders of the other side (who clearly benefit from their information being trusted), but even the motivations of their followers as well.

This latter motivational distrust may at first seem illogical, because at first glance there doesn’t seem to be a self-interest in promoting incorrect information that doesn’t benefit oneself, but here we see a case where human thinking often goes wrong: we have a desire to be right about our opinions, and it is easy for us to let our desire to be right (or the desire for something to be true) to overwhelm our critical thinking ability.

In passing, I will note that I think this “desire to be right” can serve a useful purpose: it allows for some stability in our beliefs, and prevents us from constantly flitting from one belief to another, which could make it difficult to achieve any long term goals. It also allows us to “go against the norm” and motivate the effort to prove that some belief we hold is correct. But I think it’s obvious that if we want to think optimally, we have to be willing to make adjustments to our beliefs when enough contradictory data appears.

Rating of information sources based on consensus

One of the other primary ways we rate information is based on the number of people who agree on the truth of information (especially when there are relatively few people who dispute the information’s truth). In other words, we tend to believe information based on the consensus opinions of other people.

Much of the higher-level information we believe is information believed by other people we identify with in some way. This might even be viewed as an extension of our desire to be correct ourselves: after all if we believe we’re often correct about information, it only makes sense that we will trust others more when they think like us on other topics.

This kind of thinking can go horribly wrong though, when large numbers of people reaffirm each other’s misguided beliefs, and can result in a short-circuiting of virtually all critical thinking by most members of the group (and generally leads to the exit of critical thinkers from the group).

Rating based on familiarity with the source

Everyone has their own unique method of rating the opinions of other people, but we can see certain commonalities used by most people. On average, I think we tend to rate highest the opinions of people we frequently get information from, especially when we see later confirmation from other sources that the information was true. It also helps when we are able to interact with these people in some way, as it allows for us to gain more insight into how they think and what they know.

At the beginning of our lives, this usually means our family, and later classmates, then work colleagues, etc. Of course, we don’t end up trusting information from all those people, either because they are often wrong on some topics or just because we don’t like them. But for many of the people whose opinions we rely on, our interactions with them have allowed us to make judgments about their motivations and their knowledge and rating skills in various areas, and this allows us to rely on them more than people that we’ve never met in person or had a long online relationship with.

The value of personal interaction as a rating mechanism

Many people consider “meeting in person” an especially important method of measuring a person’s trustworthiness, because we have some ability to read body language that’s not available from online sources (even from staged video recordings, when this kind of information can be often be obscured).

I personally have found this a very useful means of measuring a person’s reliability and even skills, especially when I can interact with a person long enough that it becomes difficult for them to maintain a false facade. In such a situation, you have the opportunity to catch inconsistencies in statements unfiltered by third party reports, unguarded expressions, etc.

But I think it’s also fair to say that some people are terrible at making judgments about other people based simply on in-person experiences, especially when dealing with a person that is practiced at deception. I’ve witnessed cases where I was rapidly convinced that I was dealing with an untrustworthy person, yet been in the company of other people who found that person entirely believable (as a side note for those interested, one of the techniques these deceptive people often employ is effusive compliments, so be especially wary when someone compliments you about something that you yourself don’t believe to be true).

All this goes to say, your mileage may vary, when it comes to being able to judge people from personal interactions. As another anecdote, for many years, I felt like I could quickly detect when a person was intentionally trying to deceive me via various body and vocal cues, but then I had a meeting where I was listening to “difficult to believe” information for the span of many hours, and that person’s body language and speech never gave away a thing to me. The only way I was finally able to be sure that the person was attempting to con me was when I asked a critical question that underpinned the implausibility of what they were telling me, and they skillfully when into a digression for about twenty minutes that almost made me forget my original question. It was in fact the skillfulness of the deflection that fully convinced me I was dealing with a con artist, despite the fact that their body language kept telling me that they were trustworthy.

Rating based on domain expertise

Rating a person’s opinions based on that person’s domain expertise (knowledge and skills in an area) can be a tricky issue, in many cases. When you’re an expert in an area, you’ve usually learned ways to identify other experts in that area. But when you’re not an expert, it gets more difficult.

One way to tell an expert in an area is how often the answers they provide solve a problem in that area, but even this isn’t a perfect method, especially when you have little personal knowledge of the area. For example, if you have a computer problem, you may know someone who solves your computer problem every time, and you might conclude that they are a computer expert. But “computer expert” can actually mean lots of different things, and a “computer expert” in one area, for example setting up a computer to play video games, is quite different from a computer expert who knows about designing software, for example. And if you don’t know much about an area, you’re less likely to know where such subdivisions of expertise exist. And vice-versa, the more you know about the area yourself, the more likely you are to know how far an expert’s knowledge is likely to extend.

This is also one of the reasons we trust experts in an area to help us identify experts in related areas. We might not trust a general practice doctor to perform surgery, but it’s likely that they can do a better job of recommending a good surgeon than pulling a surgeon’s name out of a medical directory.

Can computers enable us to share and rate information better?

We’ve all seen the incredible impact computers have had on the ability of individuals to share information, in a way that was never before possible in the past. Your computer can help you to answer in minutes the types of questions that could require weeks and months to answer in the past (if you could even find the answers at all).

And, when used wisely, computers can already help tremendously when it comes to rating information.

I can’t begin to count how many times I’ve questioned the validity of something I’ve heard, then used a simple web search to get what I felt reasonably confident in concluding was the definitive answer on the topic. It’s a really great way to settle a lot of “back-and-forth” arguments in many areas.

But another thing that’s become increasingly obvious with time is that “you can’t trust everything you read on the internet”. This is because traditional methods that society has used to rate the quality of publicly available information has not been able to keep up with the explosive increase in the number of information sources on the internet.

Sock puppets: fictional sources of information

One of the biggest problems that has emerged with internet-based sharing of information is the creation of fake sources of information. As discussed previously, our opinions about the reliability of information are often swayed by consensus beliefs about information. This method of rating information generally works pretty well, because it allows us to rely in part on the brain power of other people. But it can also be “gamed” on the internet, when someone creates a bunch of fake identities to make it look like many people are in agreement about an issue.

Such sources are often referred to as “sock puppets” because the words and actions of the fake identity are being controlled by a single person, much as a sock puppet has no real thoughts of its own. Sock puppets can take many forms: fake people, fake organizations, and even identity theft, where the sock puppet masquerades as some person or entity that actually does exist in real life.

Identity verification and anonymity

Many online information sharing platforms have recognized the problem of sock puppets and have developed methods for trying to fight the problem, with varying degrees of success.

Hive uses one such method to determine distribution of post rewards: it uses stake-based weighting of votes to compute the rewards, so there’s no strong reason for one person to create many accounts and vote with all those accounts on a post, as dividing up their staked Hive among many accounts won’t change the rewards that are given.

On some platforms, such as Facebook, users are required to agree to only create one account for all their personal information sharing (they might also be allowed to setup an organization account as well), and the platform may require some proof of identity to create the account. But for most platforms, it’s difficult to fully verify the identity information provided and be certain that the person isn’t using a false identity.

A new problem is also created when relying on a single centralized vetting of identity information like the one above, because it gives an out-sized amount of influence to the vetting service. There are often economic forces that will drive such platforms to be honest in this activity, but it’s also easy to imagine cases where government interests could exert influence on them to make exceptions and allow fake identities.

There’s also a downside to requiring identity information as a requisite to information sharing: it’s not always safe or smart to share information and have it be publicly known that you are the person that sourced the information. It’s for this reason that “whistleblower” programs usually have mechanisms in place for preserving the anonymity of the whistleblower as much as possible.

Web of trust as a means of identity verification

An alternative means of identify verification is to use a web of trust system, and it’s probably the main way we validate identity in our normal lives.

Let’s take a brief look at how a web of trust system works and how we use them today to “know” a person is a real person. Let’s say you receive an email from someone who claims to be a young cousin of yours whom you’ve never met, but they are traveling from overseas to your area, and would like to stay with you while they are there. How do you decide if this person is who they say they are and if it is the kind of person you should be comfortable inviting into your home? Most likely, you’ll talk to other family members who would know this person, and find out what they say about them.

In this case, these family members are acting as your web of trust, and they meet some of the key characteristics that make for a good web of trust system. You’ve likely met and are pretty familiar with the family members you’re consulting with. They are domain experts with regard to your family members and some of them will likely know this person. You trust that they will generally be motivated to give you accurate information (although you might rate information from his mom at a different level of trust relative to other reports). If you receive generally good reports from several family members who’ve met this person (consensus), then assuming you’re not an extremely private person and have room in your place, you’ll likely allow them to stay for a while.

As you can see from the above example, we already use simple “web of trust”-based identify verification, and computers aren’t even required.

But the term “web of trust” implies more than just relying on the opinions of people you know directly. A good web of trust can enable you to rely on opinions of people you’ve never met and whom it may not ever be convenient for you to meet. This additional “reach” enables you to tap into the information rating power of more people, who will know about more domains than is possible for just the people you directly interact with to know.

In the case of identity verification, it works like this: you trust people you’ve interacted with a lot, and they vouch for people you’ve never met, and they also establish the validity of the communication channel with which you’re interacting with this remote person. Now that the identity of this remote person has been established, they, in turn, can vouch for people that they know, that your direct contacts don’t know. In theory, this method can be used to verify the identity of almost everyone in the world, as long as they don’t exist in a closed society that is cut off with contact from the rest of the world.

Now while the above system for verifying identity sounds great on paper, there are lots of problems that can lead to errors in such a web of trust system, but I’d like to end this post at this point, to leave the discussion in the comments section focused on the points discussed so far.

More to come

In my next post, I plan to take a look at some problems with extended web of trust systems, and how computer networks can help with those problems. I also plan to discuss how a web of trust can serve not only as a means of identity verification, but also as an alternative to identity verification in the traditional sense of the term.