"I agree with James," said Curry, who had been waiting somewhat impatiently to speak. The N.I.C.E. marks the beginning of a new era-the really scientific era. Up to now, everything has been haphazard. This is going to put science itself on a scientific basis. There are to be forty interlocking committees sitting every day and they've got a wonderful gadget - I was shown the model last time I was in town - by which the findings of each committee print themselves off in their own little compartment on the Analytical Notice-Board every half hour. Then, that report slides itself into the right position where it's connected up by little arrows with all the relevant parts of the other reports. A glance at the Board shows you the policy of the whole Institute actually taking shape under your own eyes. There'll be a staff of at least twenty experts at the top of the building working this Notice-Board in a room rather like the Tube control rooms. It's a marvellous gadget. The different kinds of business all come out in the Board in different coloured lights. It must have cost half a million. They call it a Pragmatometer."
My master‘s thesis was related to finding and implementing a music library vis tool, which should show relations between artists and songs by grouping them together. One important aspect I learned is that, in 2D space, there are only so many nodes you can add before the graph becomes useless. You could display nodes multiple times in the same graph, but this lowers usability a great deal.
The underlying problem is that the “real” graph of relations exists in a high-dimensional space; ML “embeddings” tend to tease out the actual structure of this space.
Once you have a good high-dimensional representation, you can use something like t-SNE to reduce that to 2 (or 3) dimensions in a way that makes sense. But as you noted, if you use all the nodes, this dimensionality reduction loses quite a bit of information.
The solution is to have a way of interactively limiting the number of nodes in an exploratory way, and having a dynamically-applied t-SNE running on only that subset of nodes. If you set the iterative parameters right, you can get some really fun interactive physics-like visualizations where nodes attract and repulse each other in intuitive ways. (“This thing is like this other thing in one dimension, but also not like it in another dimension” kind of stuff.)
I remember what.cd seemed to have an amazing network graph for browsing through artists and discovering new stuff... sadly I think that's forever lost in RIAA-hell
TBH I blame the advent of streaming platforms, and the public’s acceptance of brainlessly following recommendations from algorithms.
I’m sure that each of the big platforms - YT, Apple Music, Spotify and Tidal - have someone in their team who wants to implement this type of thing, but they’re probably drowned out by overbearing algo marketing types.
They need to take a long hard look at the popularity of the rock family tree publications, or at least consider the type of traffic that whosampled.com gets from people going into a relational rabbit hole.
Zooming in and out seems to be their approach for that problem, the more you zoom out the most specific nodes lose their labels; for a music graph I would guess that would be the songs, and even further out maybe join artists by genre (or decade)
But yeah that approach may be limited as well for such an ambitious project, in the long term I think they should have a list of categories you may check (or uncheck) to show only the subjects you are interested at the moment, e.g. health, money and education.
Love this discussion and totally agree about the challenges of visualizing a network graph with lots of nodes! (I'm Lisa, head of design at System.) We'll be doing a lot more work on the graph in future releases to allow the user more filter controls.
Some kind of hierarchy - but nested abstractions seem difficult to name and unhelpful to use.
What about picking a single instance (actual song) as representative?
A platonic form, at that level of abstraction. Expanding that node would give "more like this" - consisting also of actual instances, representative of further sets.
In a similar space, I wish someone would make a graph like this for materials required to produce something, for society-bootstrapping. For example, the different tools and materials required to make a functioning water well, and how to make those tools and materials.
Cool! I can't promise any contributions, but I hope it succeeds. 2 things: it looks like this is a platform for generalized build/tech trees, and I think that is the right approach. I think you need a "poster child" project though, like what Wikipedia is to Mediawiki. Pick something that everyone can benefit from, for example the "civilization build tree". And start drawing attention to it, which will draw attention to engine powering it.
Imo, the biggest area to focus will be reducing the barrier to entry. We're people who are comfortable with version control, but your average person isn't. The underlying software can still be version controlled, but you need a way for normal people to add/edit/remove nodes so that they can start contributing. You also need a system for inferring experts in different areas, and allowing them elevated privileges.
hah, i wanted to build that same thing too. but limited to food only. should have been thinking bigger. never got around to it though. good luck! hope it's as cool as it sounds
I understand that these relations are based on some sort of notion of "evidence" via scientific studies. My question would be - there seems to be a rather low amount of studies per relation in most cases, how is the specific study that establishes a relation here chosen? In reality, in many fields we often see completely contradictory studies, e.g. in psychology or nutrition.
Is there really just one study about this? What would happen if there was one study saying that there is such a relationship, and one saying there isn't?
To me it seems that this kind of system leads to a weird simplified spaghetti view of the world, which at best might be an ok topic explorer, but at worst almost like a conspiracy theory automator (following the threads of more or less representative studies of things being "related").
(note: I am the Director of Data Science at System)
Thank you for the comment. This is a great observation and it is true that many of current relationships on System are not comprehensive (i.e. they are not based on a representative sample of overall scientific consensus in that field). Some topics/relationships are more comprehensive than others (e.g. relationship between food fortification and anemia: https://system.com/view/topic-relationship/yQrSS5fcTQs/d9wUM...)
System is still in its infancy. Compare it to early days of Wikipedia. Over time, our goal is improve the depth and breath of knowledge on System through various methods including community engagement and partnership with domain experts.
Meanwhile, if you are interested in exploring a specific topic, please let us know through our slack community (link on the platform) and we will be happy to prioritize them.
First of all, I'd like to say that this looks like a great project and I wish you the best of luck. I've done a bit of work on building knowledge graphs from semi-structured data and I know that every aspect of it is challenging. Obviously there's the data pipelines, ETL, semantic matching/categorization, statistical models, etc. Just building a simple UI for presenting a large knowledge graph was more challenging than most front end work I've ever done.
Question: if the goal is to build a knowledge graph that can "explain how anything in the world is related to everything else" how do you measure progress toward that goal? And how do you measure the quality? Just having a bunch of topics and relationships is not a great metric in my opinion. Obviously this is still very early, but here's an example I found in about 30 seconds of clicking around:
Thank you so much. We'd love for you to join our Slack community (link on system.com).
Great question. There is no ground truth that we are modeling System after, i.e. there is no causal model of the world out there (to use Pearl's framing). So I'm not sure we can know how far along we are epistemologically. More practically, for the next few years we have plenty of work to just represent all the existing corpuses of scholarship! The truer and arguably more meaningful test of progress though is how decisions are improved — for users, for organizations — that use System.
Re completeness, as I wrote below, System Search results are not necessarily comprehensive — but they will be. System is in the early stages of its development as a public resource and you should expect that knowledge will be missing. The knowledge base will be constantly growing and improving and evolving as knowledge does. Our community will play an important role in relating what we expect or know should be related.
First, thanks! If you'd like to reach out and learn more or talk about your learnings from building something like this we'd be very interested (we have a Slack community and a direct contact form on the site).
As for your questions - we have tools for assessing the reproducibility (in the statistical sense) of models and relationships added to System, as well as tools for users (and built in to the platform itself) to assess the relative statistical strength between any two relationships that you find on the site.
And, yes, we're early on in the process of writing (peer-reviewed) evidence on various topics, and as you note, the value of seeing these systems will grow with how detailed the topics are covered and the overall number of the world's topics shown to be related. I hope you'll stay engaged to see!
That choice will hopefully always be true in a free society.
I founded System because the biggest challenges we face in the world — from COVID to climate change — are systemic, yet our data and knowledge are organized into silos. I believe this fundamental incongruity makes it impossible to think, plan, and act systemically. As a result, we are stifled in our ability to reliably predict outcomes, make decisions, mitigate risks, and improve the state of the world for everyone.
System is a shared tool for systems thinking — and, we hope, a springboard for collective action.
We have great respect for freebase (see comment below on metaweb) and WA. System offers a different lens on data and knowledge rooted in the statistical associations between things in the world.
So I tried to find "language". The only hit was "language development". This turns out to have a handful of links, which are factors that have been found to have a bit of influence on language development. No idea in what direction, and quite a few from what appears to be the same source. But the source article is not cited. Why? The extremely limited information that is available suggests some kind of self-reporting, which is, as we all know, highly unreliable.
But, with a bit of searching, I did find the source. It's some random overview article from Family Matters (issue 91, 2012), a publication of the Australian government. It is based on survey data. It doesn't even seem to be peer-reviewed (which is the lowest bar in scientific publishing, IMO), and seems to contain all the classical ingredients that gave us the replication crisis (NHST, high N, just throwing everything in a GLM, assuming absence of significance is absence of causality, and that's just superficial reading).
On the other hand, every plain bit of knowledge on language development is missing. Where is the part that relates language development to the actual process, or even the child's parents or guardians, the influence of multilingualism, or the multitude of other potential factors? Not to be found. Adding them in the current style will not overcome the problems. I've got a bunch of articles on a specific (language-related) topic which reach opposite conclusions. I'm sure they exist for language development, too.
"The System", as it is now, will definitely, 100% lead anyone trying to understand language development to the wrong conclusions.
This looks like something that could be cool, though I'm having trouble finding a clear description of what it's really supposed to be.
In particular: What do the nodes and connections on the graph represent?
For example, once this project is complete, would this project contain nodes for "Hacker News" and "HTML"? If so, would they be directly connected, or if not, what would the shortest paths between them look like?
It'd be neat to see a concise, technical description that sketches the conceptual-framework and briefly the methods that implement it. For example, what's the top-level meta-model? And how does the particular form of the graph follow from it?
---
EDIT: Read through a lot of online stuff trying to trace down the underlying theory...
I'm having some trouble finding the technical details behind the ideas; critiques seem to commonly mention a lack of technical-depth in the underlying conceptual-framework.
It'd seem helpful to have a technical-description for folks who're well-versed in math, modeling, etc..
Hi, I'm the founder of System (www.system.com). System is a free, open, and living public resource that aims to explain how anything in the world in related to everything else.
- We formed a Public Benefit Corporation, committed to open knowledge and advancing systems thinking, to operate System.
- Our mission is to relate everything, to help, the world see and solve anything, as a system.
- System is built on top of a novel, large-scale graph platform that gathers and organizes evidence of statistical associations between things in the world.
- Like Wikipedia, the information on System is available under Creative Commons Attribution ShareAlike License, and topic definitions on System are sourced from Wikidata.
- Anyone will be soon able to contribute evidence of relationships to System using a variety of tools. v1.0-beta is read only. The determination of what datasets, models, and papers statistics are retrieved from currently falls to members of our team and to users who are beta testing the tools we've built to contribute to System.
- We invite you to join a diverse community of systems thinkers from all walks of like who are coming together to build System.
At the end of your intro video you ask the viewer to imagine what could be possible with such a system. But that’s putting the onus on the viewer, who has likely never thought about such a system, rather than the creator who is selling the vision. I’d encourage you to give some concrete examples on what could really be achieved here.
When everything is related to everything, it’s hard to get anything actionable out of such a model. Further qualifying the edges should also matter a lot… is something correlated? Causal? Indirectly related? How far does the causality propagate? For example, could changing the formula for toothpaste affect obesity? I’d imagine it would be easy to draw a graph connecting these things, but it’s probably difficult to know if a causal change is likely to produce the desired result.
This reminds me a lot of cybernetics, which ultimately failed. I’m be curious for your thoughts on that field and it’s relationship to your endeavor.
Relationships on System carry several parameters that address your question. For example, in what population was this measured/what time period, a normalized measure of the statistical strength, statistical significance, the direction of the relationship when possible, the sign of the relationship, and a measure of the reproducibility of the evidence. You can read more in our docs: https://docs.system.com/system/how-system-works/relationship.... Our aim is to synthesize (or meta-analyze) all of this evidence and associated metadata in such a way that helps users take actions. An open causal model of the world, to use Pearl's framing.
Love the question re cybernetics. I am inspired by the writing of Mary Catherine Bateson on the matter. She has argued that the tragedy of the cybernetic revolution, which had two phases, the computer science side and the systems theory side, has been the neglect of the systems theory side of it. We chose marketable gadgets, she says, in preference to a deeper understanding of the world we live in.
I went looking for mentions of Semantic Networks, RDF, triplets, taxonomy, or other common knowledge-related terms but I found nothing. Thoughts about pre-existing knowledge graph systems and how they may or may not apply to System? Thanks!
The post title immediately had me excited, because between the looming threat of dangerously fragmenting perspectives on truth and the fundamentals of automated artificial reasoning oriented towards meaning relevant to real world problems, ontology is of great interest. It's been the cause of much of my typing.
Unfortunately, the first impression was not good. I grant that the amount of data so far is limited and other questionable experiential aspects of the site aren't a knock on the underlying goal, but I can't help but feel that the way it's being done is not the inevitable future. That may only be a surface appearance, but when people put something out there and promote it, I assume they believe it's a fair representation of where their vision is headed.
There's also this nagging concern that the aim of this project can lead it to become an authoritative source on relationships which other software and services are forced to consume in order to compete, because the first open project of its type to reach escape velocity in respect to its data momentum may cement itself as the standard which suppresses the competition-promoting value of the data being forkable (not unlike Wikipedia).
15 years from now when it's passed a usefulness threshold, maybe most note taking apps, Microsoft Word, search engines, dictionaries, machine learning based services and more may come to rely upon it to improve access to context. At that scale it has non-negligible influence and the alpha-male status of it may limit evolutionary diversity around the intensity of the relatedness of things.
Don't get me wrong, a project like this is necessary anyway regardless of the potential dangers if only to head off a Chinese implementation to brainwash the world. To rephrase Netflix wanting to become HBO before HBO could become Netflix, don't become China before China can become you. The last thing we need is a monopolistic ministry of truth produced by runaway network effects that don't care about Public Benefits Charters. :)
Speaking of the Public Benefits Charter, Google never significantly defined "Don't Be Evil" in less vague terms, so the removal of it was harder to identify as a canary. The charter doesn't clarify what "scientific thinking" is, at a time when we are redefining what racism is with reckless abandon. I also didn't see any assurances about whether the charter would survive acquisition by another company and whether such an acquisition would be an exception to the promise to not sell user data. It also doesn't specify whether there's an active preservation clause.
We've seen Chinese acquisitions of sensitive companies pass by without notice providing further evidence that governments cannot be trusted to preserve their countries, it's up to the people. Companies themselves need to integrate a binding promise (if legally possible) to bar those outcomes and to carry that promise forward to acquirers which means any company operating in a state or country where such a promise could not be legally enforced doesn't qualify as a suitable parent company. Maybe that's wishful thinking, not a lawyer.
In any case, I think for this kind of project there is a likely leapfrog scenario that may cause it to be supplanted if the company doesn't do its own leapfrogging.
Additionally, there's the concern about whether the majority of the data qualifies as articles of fact which supposedly can't be copyrighted and could potentially nullify most of the impact of the CC-SA license. Luckily, all facts are fiction now, so you might be able to skirt by that one! Hope that doesn't mean embedding lies will be a future strategy to establish copyright snags for preserving enterprise customers. :/
By internal contradictions, do you mean conflicting evidence in the relationship between topics or metrics? That will (and does) come up regularly - peer-reviewed studies investigating the same topics have differently measured (or contradictory) results. We have tools for assessing the statistical quality of submitted relationships (through things like statistical reproducibility, algorithm type, statistical controls, etc.), so unreproducible or statistically unlikely relationships will be clearly seen as such. Building tools to programmatically test reproducibility of evidence is definitely something we've thought about (if that's the "formal verification" you are talking about).
Ultimately the goal will be to (statistically) approximate the sum total of all evidence between pairs of topics, and also to provide users with the tools and sources to assess (and apply!) that evidence.
2. The criteria for reproducibility seem a little rough to me? They seem to be sort of distant from things like registered replication, publication bias analysis etc.
3. My gut impression is you need some kind of meta-scientific model, eg something that models the probability of studying an association, the observed association conditional on that, effects on heterogeneity of effect sizes etc.
4. Along those lines, I wonder if there's an implicit schema of looking for nonzero associations and documenting them rather than reporting best estimated strength of known association? Maybe not.
5. I'm curious how you define nodes/topics versus subnodes/subtopics. I suspect defining the nodes/topics and their boundaries would become tricky?
RE your first question, one of their answers from product hunt may be useful:
> Q: Is this supposed to be open source version of Google's knowledge graph?
> A: At their essence, KGs are based on semantic relationships, e.g. coffee is a beverage, apple and banana are fruit, diabetes is a disease, etc. System is based on statistical relationships (collected and synthesized from data, models, and papers): A predicts B, C is caused by D, E and F are highly correlated, G and H change together, etc. [...] We hope these will be complementary ways of understanding the world -- one based on language, the other based on statistics.
I think internal contradictions are more of an issue for a Knowledge Graph, which try to infer things and have to make conclusions based on possibly contradictory evidence. System just tries to present the available connecting evidence without making object level conclusions itself.
Great question. We present all the evidence behind a relationship (on "evidence cards" that show the source, strength, sign, direction, population, controls, and reproducibility). The evidence cards on a relationship page may conflict, and this is clear for users to see and evaluate. We also generate a natural language synthesis of the evidence. We are working on enhancing our meta-analysis of the evidence to flag these kinds of conflicts. And our community will surely play an important role (as is the case on Wikipedia).
Hopefully, I will get to use this as a source for validating more opinionated system dynamics and agent based models. Community driven projects are going to have a more linear growth curve, but I'm excited to see where it goes and contribute. There's a big discoverability issue in science that causes our decisions to be less informed, but this is obviously a massive issue to tackle. Best of luck!
We obtained the domain from a company that was using it for a very different purpose. They appreciated our mission and we arrived at a reasonable price. And we're grateful to them.
I clicked on "unemployment" and the results were unsatisfying to say the least. All it told me was that unemployment is related to being out of work or unemployed. I love the idea though.
I agree - to me it displays suicide & crime & health.
Btw. personally I expected something like health & education & <other stuff, e.g. economic> => I guess that it depends on the point of view, if "upstream"/"downstream", cause/effect?
My mindset was therefore expecting an upstream/cause link, but it looks like that it shows the downstream/effect of what is being queried?
(note: I am the Director of Data Science at System)
Thank you for the question. We don't distinguish the upstream and downstream effect at the moment. However, we do encode directionality of pieces of evidence that construct those relationships. If you drill down on any relationship, you will find directionality arrows. (note that I am not using directionality as the evidence of causality. That's something that we are currently working on).
TL;DR Because the biggest challenges we face in the world — from COVID to climate change — are systemic, yet our data and knowledge are organized into silos. I believe this fundamental incongruity makes it impossible to think, plan, and act systemically. As a result, we are stifled in our ability to reliably predict outcomes, make decisions, mitigate risks, and improve the state of the world for everyone.
So, we built System. A new way to organize data and knowledge into systems based on the evolving relationships between everything in the world. A shared tool for systems thinking — and, we hope, a springboard for collective action.
We share much in common with Wikipedia, are deeply inspired by what it does for the world, and hope to be used alongside Wikipedia. Our openness and CC license, our use of Wikidata as the source of definitions, for example, are common. But System aims to explain how anything in the world relates to everything else — based on statistical evidence. It is not an encyclopedia.
Yes, it is much easier to build a holistic picture of a topic when you can see connections like this. I think many people are not interested in viewing data in this way as they prefer to drill down into the details.
(note: I am the Director of Data Science at System)
We are building System with multiple personas in mind. While a domain expert might be interested in drilling down to understand the source of every piece of evidence, an average user may just want to understand the system of using pacifier for their newborn.
The atom of System is a single statistical association that carries its context (population, covariates, etc). As you zoom out, we summarize and combine pieces of evidence (through semantic matching and meta-analysis).
This method provides an incredibly powerful framework to tailor to different needs and level of expertise.
I was thinking the same. Quite sad really: if simply nobody else was using it for something useful yet, it should just be available... I hate that domain squatting is legal.
I'm fine with normal trade. I might use such a domain for personal use if this is my nickname or something, and if someone wants to have it and I want money for that, that's fine (it's annoying for me to move after all), but just buying whole dictionaries of domains... should be illegal for a scarce resource.
Frankly that'd be better than the status quo. If there are a million the same games with a different word as title from the same company, it's easy enough to show this isn't legit. You'd have to do at least a tiny amount of effort and raise the cost of registering whole dictionaries.
Also, I'm not claiming to have a perfect solution for enforcement, just more generally that I wish we'd try to do something about it.
The use case I had for something tangentially related was an economic factors graph for linking weighted macro risk exposure in publicly traded companies, where for a given thing or company you knew about, you could link it to interest rates, suppliers, commodities and their prices, competitors, countries and locations of operations, investors, myriad other things, etc. It was a risk arbitrage discovery tool where within a few hops, you could find potential exposures that would affect prices. Data sources would be just piling stuff into the graph. sort of how this System tool appears.
Thanks very much. We hope System will be complementary to Wikipedia. They share a core open ontology that will allow for possible future interoperability.
That is very much the risk and one we have taken on as part of our tech and culture from day one. Today, System considers parameters like evidence reproducibility, significance, and statistical strength. But there is a lot more to do here. As a Public Benefit Corporation, we've codified it in our charter that we must consider and share the potential unintended consequences of each major release. And we'll be publishing our first such report shortly.
Excellent approach! I was actually thinking as I was writing that Wikipedia could be ab excellent source, just producing a graph starting at the actual links within Wikipedia, and then doing the conceptual linkage (using knowledgeable humans to start until an AI develops that is actually able to embody conceptual understanding). Obviously there's IP rights issues involved, but I'd be surprised if a reasonable deal couldn't be made.
This is really cool. I built https://thicket.io a while ago, which had related goals but which System really puts to shame. Thicket took a simpler approach to graphing knowledge, so it's fascinating to see how you've structured the site and all the work you've done extracting meaning from your evidence.
I hope you have better luck with user engagement than I did!
Interesting, very ambitious. How much did you draw inspiration from early Lisp machine research, eg for “expert systems” with inference engines and knowledge graphs?
http://knowledgegraph.today/paper.html
V. Interesting approach, but does making connections statistical avoid making system a consistent system as in in Gödel incompleteness? How does system represent things like the liar paradox (if it does?).
(note: I am the Director of Data Science at System)
Thank you for the very intriguing question. System consists of statistical relationships and their causal counterpart (A correlates with B and A caused B). Theoretically, we don't claim completeness of the causal graph even in the steady state.
I don't understand this. Surely there are lots of things that are connected in ways that aren't mapped here:
COVID-19 and Vaccination,
Real Estate Price and Housing,
Socioeconomic Status and Happiness are all nodes that are not connected in the graph - but are obviously connected in the real world.
Is this a WIP until everything is connected to everything else?
I wouldn't say a direct connection between those things is "obvious". Seems like you could just as easily argue that they are "obviously" indirectly related (ex. COVID-19 would effect social distancing, which would effect population density, which would effect real estate and housing).
But I agreed that the "nodes" I just made up are arbitrary, and that you could make an argument that everything effects everything else in some way. So it's not clear to me what gets to be a node and what constitutes an edge.
A noble effort, I'm all for exploring it, but it seems like pie in the sky.
In brief, System is designed to maximize precision in how evidence is captured and represented, while also ensuring that information about the same or similar things is grouped together. This is key to building and representing one system. This translates into three types of nodes of increasing specificity: Topics, Metrics, and Features. More here: https://docs.system.com/system/using-system/topics-metrics-a....
Search results are not necessarily comprehensive — but they will be. System is in the early stages of its development as a public resource and you should expect that knowledge will be missing (just like the early days of Wikipedia and Google). The knowledge base will also be constantly growing and improving and evolving as knowledge does. ICYI, you can join our slack community to discuss this work further (the link is on system.com).
There is great work out there that mines Wikipedia for semantic relationship (co-occurrence of topics for example, parent-child relationships, etc.). But that methodology would not provide the statistical evidence that is the building block of System. Relationships on System are statistical in nature. A predicts B, C is caused by D, E and F are highly correlated, G and H change together, etc. By organizing these (billions and billions of) statistical relationships, anyone will be able see anything that's important to them as the system it truly is, rather than the silo we often see today.
No, Wikidata is an open database of semantic definitions and relationships. System is a public resource that aims to explain how anything in the world is related to everything else based on statistical evidence. Semantic vs statistical is the difference.
System is possible today because of Wikidata and the advancement of open knowledge: All definitions on System are sourced from Wikidata. System will contribute back to the open knowledge commons with a new, free, open, and living knowledge base of statistically-based relationships between things in the world.
>System is a public resource that aims to explain how anything in the world is related to everything else based on statistical evidence
People have made a game out of finding spurious correlations that are both impressive and funny.
For now the site seems to have a focus on Medicine. That's great because we spend a whole lot of money running RCTs and collecting trial data. But the stakes are also very high.
How do you make sure that System doesn't accidentally become a public resource that explains how anything is (spuriously) related to everything else by confounders and unfortunate correlations?
Relationships on System are gathered, stored, and presented with a variety of contextualizing fields designed to help System and users evaluate and weigh the evidence. These include Strength, Sign, Direction, Population, Controls, and Reproducibility.
ICYI we discuss and review these methodologies on our slack community (link on system.com).
Is the pope catholic, or statistically related to catholicism?
Is Chicago related to Illinois by published evidence?
Thank you, and congrats on the launch. I loved freebase, and desperately want it back.
If this works, please don’t sell.
(note: I am the Director of Data Science at System)
We love wrestling with these types of questions at System. The examples that you gave are "semantic" on System. You find those connections in Wikidata for example. (Q19546 -> Q9592 -> Q1841 or Q1297 -> Q1204). A relationship is statistical (as defined on System) if you can estimate its strength statistically, in a population and with certain statistical confidence.
We're big fans of metaweb (and had one of their founding engineers as an advisor early on).
At their essence, knowledge graphs (like metaweb) are based on semantic relationships, e.g. coffee is a beverage, apple and banana are fruit, diabetes is a disease, etc. System, instead, is based on statistical relationships (collected and synthesized from data, models, and papers): A predicts B, C is caused by D, E and F are highly correlated, G and H change together, etc. While statistics (probabilities for example) can definitely be used in a KG (and certainly in large scale ones), the nature of the relationships themselves (x is a movie, x stars y) are semantic.
By organizing these (billions and billions of) statistical relationships on System, anyone will be able see anything that's important to them as the system it truly is, rather than the silo we often see today.
We hope these will be complementary ways of understanding the world -- one based on language, the other based on statistics. Importantly, System leverages the same core ontology as Wikipedia (i.e. Wikidata) so the definition of "coffee" on System is the same as on Wikipedia. So these two ways are very intentionally interoperable.
You can read more about System's methodologies in our technical documentation: docs.system.com/system.
You're probably not serious about your comment. But... thanks anyway. I found this IBM design website with some elements that align in concept with the ones at System.
I love this question and will need some time to answer thoughtfully. I'm actually very flattered by the IBM reference. :)
(I'm Lisa, head of design at System.)
We attempt to redirect to our mobile site about.system.com/mobile and are working on a mobile friendly version.
Apologies that you are experiencing issues. Could you send us more details through our feedback tab or our slack community or email hello@system.com(If you could also attach a screenshot, it would be really helpful to see what you are seeing).
I used desktop mode on Firefox mobile on Android which prevents the redirect and renders your app at a desktop resolution (and changes the user agent to look like a desktop browser).
I was able to navigate the graph, select nodes, navigate references and so on. This is obviously not a true E2E test but I encountered no bugs in my usage.
I don't know what drivel means, but I also thought that claim was a bit out of place when I saw it in the video (the only thing I can see on mobile...). Sure, yeah, we have droughts and climate change on our hands, but what does that tell me about your product? Instead of using scammer tactics of instilling fear and urgency with a looming problem that only they can fix, just tell me what the product is. I had to scroll quite far down in the comments to find it (specifically this comment: https://news.ycombinator.com/item?id=30689872) because the video is just too abstract.
Edit: looked up drivel, learned a new word today. Not quite how I'd describe this product though...
Hopefully the welcome screen (post video) more clearly spelled out the product purpose. You can also read our full product guide here ICYI ("Using System"): https://docs.system.com/system/
> Hopefully the welcome screen (post video) more clearly spelled out the product purpose.
Oh, if that was supposed to show on mobile, it didn't for me. Or maybe I closed the tab too soon since it said that no mobile experience was available and that I could watch the video instead (figuring that, after it finished, that was all for now).
