Interview with Udi


DN: Udi, thanks for taking the time for giving this interview.

UB: My pleasure, thanks for having me, Dibau.

The Problem

DN: 1st of all, can you describe in a nutshell what the Matrix project is - what’s the Why behind it?

UB: Sure. The Web Wide Matrix project is basically an attempt to solve a big problem I personally have. Almost every day, I encounter some big stream or collection of information, that I really want to go over, but know that it will take few hours or a full day, which obviously I can’t allocate to. My twitter feed alone receives some 20K messages a day, much of it very important for me, but even within it I’ll often get to a resource, say some interesting person/blog/publication, which also contains thousands of texts which I’d really want to go over.

So, the problem is that it is important for me to go over all of these streams of information, but it will take many hours or days & I only have a few minutes free for doing that.

DN: I must say that most people don’t have this problem, you know. For them, most feeds are just for entertainment, & they’re fine with consuming just a fraction of them.

UB: That’s the thing - it depends on whether you think there’s real value there or not. I only have time to read say 0.00001 of the feeds I receive each day, but I’m making so much value from just these few items, so it’s reasonable to believe that I could make up to x10,000 more value if I could go over the entire feed. I don’t know about you, but I’m quite frustrated knowing I could create x10,000 more value every day.

DN: Yea, so it’s basically the problem of not being able to ingest the amounts of information you want to process.

UB: Exactly, some people call it Information Overload.

DN: & can you give me some examples of some these feeds you refer to?

UB: Sure. So for me they include: Twitter feed from people I follow, Blogs & tech news aggregators I subscribe to, Email accounts, Slack communities, research papers.

The idea

DN: OK, so what are you trying to do about it?

UB: I started thinking about, & identified 2 underlying problems: 1 is that the information in these streams arrives as an assorted feed of texts, similar to a baggage carousel you’ll find in an airport. This interface is fine for few dozen items, but completely ineffective for thousands. The 2nd underlying problem is that I need to go over all the streams myself, which given my time constraints also implies that it can only work for few dozen items but not thousands.

To address the 1st problem, I started thinking about interfaces that do allow you to consume the items you’re interested in, out of a selection of thousands of items. A good interface for that is a supermarket, or for that matter, any store, library or organized space, where you have tens of thousands of items, but it only takes you few minutes to go find & discover the things you want.

DN: Wait, let me stop you for a second. I agree that it only takes my few minutes to find what I’m looking for in a supermarket, & discover other stuff, but can you explain what makes the Supermarket so much effective versus the baggage carousel? Just the fact that it’s sorted?

UB: No, there are 2 human faculties that are utilized in an organized space, such as the supermarket, but totally don’t help in an assorted feed, such as the baggage carousel. One is what I call Topsight Vision - the ability to see an overview of everything. Humans are quite good in that - if you stand on a sky-scraper, & are asked to find some visible object below - out of the thousands of visible objects - it will usually take a person less than a minute. The 2nd faculty utilized in an organized space is Spatial Memory - the ability to build an internal map of the space, & remember where each thing is located. So, after the 1st time I’ve found where my Seiten cubes are located in my supermarket, I will know exactly where to find it or related products such as Jackfruit stripes.

DW: I see. So there are real reasons why this type of interface is so much effective vs the existing ones.

UB: Right.

So, once I identified the problem with the interface & an alternative interface which is much more effective, I started thinking on ways to automatically convert streams of assorted texts into an organized space that I can consume the way I’m shopping in a supermarket.

To address the 2nd part of the problem - the fact that I need to go over these streams myself - I immediately imagined having a team of assistants, that are smart enough to go over the streams of texts for me, & make sense of them, so that I won’t need to do that myself.

The combination of these 2 ideas is the Web Wide Matrix. Imagine this: you encounter this big streams or feed of texts that you want to go over. You allocate a full floor in your office building - say 10K sq. You ask your team of assistants to print out the entire stream, & start organizing it in the floor, like a supermarket, where the departments represent categories, the isles sub-categories, the shelves are topics & the items on the shelves are sorted by concepts within these topics. You enter the floor once the assistants have finished, & you immediately get an overview of the entire stream, & can easily see the stuff that interests you & have a good sense of what’s available & what you wish to focus on & consume.

That’s what we want to build, tho obviously in bits & not atoms - VR & not real office buildings.

DN: OK, that wasn’t a nutshell.. Let me see if I got it right: so this Web to Matrix project comes to solve your personal information overload problem like this: every feed is converted into a large VR room, with all of the feed content displayed like a store or supermarket..

UB: ..any organized space - could be a cloth closet or a library, could be a miniature city or a replica of your home.

DN: & you achieve this using a team of smart bots that basically process the information for you?

UB: correct - they basically take-in the raw feed content - throw it on the floor, read it & start organizing it according to the way I want it to be arranged & displayed.

DN: So, it’s very personalized to your preferences?

UB: Well, I try hard to encode the Knowledge of how the smart bot should do all this in a very simple to write document - which I call: a Training Course. My plan is that it will be so easy to write, that other people can change it to meet their needs or download Training Courses written by others.

DN: I see. I think I get it, although TBH I don’t see myself using it IRL.

UB: Well, as long as I will be able to use, I’m good [laughing]. I definitely see myself using it - I need it & therefore I’m building it, even if no one else wants it. And just as an anecdote: one of the difficulties & challenges I encountered when building it, is that when I work on programming this thing & want to test it, I can’t do much further work because since I’m testing it with real content, I find myself swollen & drilled in into it - it’s really difficult for me to stop using it, once I start testing it.. [laugh]

Use cases

DN: interesting. I have another question: what are basically the use-cases. Who do you think may want to use this - besides yourself - & for what purpose?

UB: Lets start with actual important stuff. There are people whose work is to analyze huge streams of information or data, & find stuff that requires stuff that requires attention & handling. So one example would be the people that need to analyze streams of events related to security in order to say: here’s a real security breach that we need to handle right now. These security analysts need to go over & analyze streams of incidents or events. Unfortunately, the amounts of data they need to process, is horrendous. And they actually don’t use much visualization, I think because they need to see the actual data & not an aggregation or visual summary.

But, I believe that when applying the Web Wide Matrix to this problem, the full data will be visible to them, & will be organized & categorized, for example, as a massive store with products, or a book library, that’s well organized. & if you see everything - the human perception has this really powerful capability of being able to find a needle in the haystack, especially if the haystack is nicely organized on a big space, & not all hay look the same. [sigh] I don’t know, maybe this metaphor doesn’t so much work..

DN: bad metaphor, Udi.

UB: I can tell you, if I see a stream of millions of lines of text that all look the same, or on the other hand see them as different objects representing their meaning in a big space, I think I will be able to find the things that are of interest to me much better in the organized space, especially if I navigated this space so many times, so that I know that if something pops up in a certain area, I know what it is just because of its location. There are real big advantages in leveraging those 2 human capabilities - the Top-Sight Vision & the Spatial Memory - that will analysts do their work much more effectively, that’s my hypothesis, not yet put to test - I didn’t experiment it yet.

Now, there are many others that do similar work. There are analysts of operations log files, say of data-center server farms. There are analysts that go over market or trading or sales data. So in the work of all those people that need to consume all this information, I believe the Web Wide Matrix could be transformative.

DN: But, aren’t these types of jobs expected to be replaced soon by AI that could automate it & perform it much better?

UB: Maybe so, however, as much as I like machines, I’m on the humans side - I want to empower them to be x1000 more effective in their capacity & work. I still believe in the human brain & what it can come up with, even with all the AI hype. I mean, you can automate pattern recognition, but if you empower people to leverage more data & information, they can find new insights & come up with new ideas based on their ability to correlate & synthesize stuff.

DN: I didn’t think about it that way.

UB: Another use-case is for individuals or organizations that need to meta-learn from a large corpus of knowledge or information, basically, to extract insights by aggregating a lot of texts. An obvious example is when a municipality or corporate wants to extract insights or ideas from a large open-ended survey. Another more concrete example would be a biotechnology startup developing some devices for diagnosing diseases, that needs to extract information on correlations between genes & mRNA’s out of hundreds of thousands of research papers. Just going over all of these text documents would take a team of people many days. But if you could have a team of smart bots that will analyze, aggregate & organize them in a large space where you can see all the aggregated relations between different genes & mRNA’s, you’ll be able to quickly understand & utilize the insights from this large corpus of texts, basically, find the genes correlations that are most interesting to pursue. Instead of having a team of people working on this for days, it can take one person less than an hour.

DN: Yes, but it’s possible to write software today to process all these research papers & extract these correlations.

Processing content

UB: But you’ll need to write this sophisticated software, which will again take a team of engineers a few days or weeks to develop. The idea of the Web Wide Matrix is that information is only partially-useful if it can’t be processed, so we’re trying to make this processing as easy & seamless as it is to just access it. You’ll have a team of bots that will do the processing for you, & you’ll just need to inform them how you want the information to be analyzed & organized.

DN: But it surely takes time to teach these bots how to extract correlations between genes & an mRNA from research papers. It’s not trivial to understand text & classify whether a mention of a gene & a disease is actually a description of a positive correlation found between them.

UB: Very true, but that’s why the knowledge for training these bots to perform their task is done using very simple documents - we call them training courses - that provide the seed knowledge for doing this job, in a very high-level manner. Once there is a large repository of such training courses, you can quickly find ones that answer your needs, or almost answer your need, & just tweak them a bit to do the job. I usually compare it to HTML - the language used to write web pages. Since it’s a very simple language, people were quickly able to develop web sites for any purpose. Having the ability to view the source of Web sites you encounter, also helped people to take examples & tweak them to produce what they need.

DN: This is also the case with coding in general - programmers that needs to perform a certain task, usually do a quick search to find some good example on the Web, and then just modify it to fit their needs.

UB: Exactly, the plan is to have a large repository of these training courses documents, that you can quickly search & utilize to come up with a course that trains your team of bots to meet your needs. & the key here is that the courses only provide the seed knowledge to get them started - once they start generating the organized spaces, you can provide them with feedback, that will cause them to explore more options to improve their training & make the organized spaces even more fitting to your exact needs.

DN: OK. I think I get the main function of this thing - converting huge streams of content into an organized 3D space, using a team of AI bots that get trained to process the data & generate this 3D space. And I think I get your argument that this new interface will be more effective for consuming information. I must say, however, that I don’t get how this magic is achieved, can you describe the architecture that enables this to work?


UB: A core part of the Web Wide Matrix, is that we change the interface for effectively consuming information, but this change depends on 2 underlying changes. The 1st is the introduction of a an abstraction layer over the web, that enables uniform access to information, for both humans & machines. & the 2nd enables the seamless processing of this information, as an integral part of the access layer, which results in these organized 3D spaces.

DN: Abstraction layer over the Web? What do you mean by that?

UB: OK. So to access content on the Web, you need to connect to many Web sites, each having its own API interface, its own content format &c. This makes it difficult for the teams of smart bots to both discover & easily process & understand the content. So we wanted to introduce this abstraction layer that basically pipe the content from the Web into a different model that’s easier to process. Basically, we wanted to make the interface of consuming information for machines more effective, just as we’re making the interface for humans more effective, & we do that by applying the concept of organized space to the whole Web, just as we apply it to a single stream of content.

DN: I don’t understand - why would a 3D interface to the Web help machines process it?

UB: Well, a 3D interface is how we perceive the world, & it’s based on the concept of space, where things are located at some coordinates, & we are able to sense them, move them around & store them in an organized manner. Think of a group of baboons looking for food. The Web today is like a giant locker room, & the baboons need to open each one of them & make sense of what’s exactly the thing inside them & whether it’s food or not. On the other hand, if you convert this locker room into a house or a supermarket, the baboons will have a look at each room or department, figure out which rooms are the most likely to have food in, & quickly discover the food they like & eat it. All this just because you applied the concept of Space - things are located & organized - & Sensing of content - things look & smell differently & you can perceive them.

DN: you’re saying that the smart bots use vision to process information?

UB: not exactly, I’m saying that the Matrix is a spatial paradigm for accessing information. Each piece of information has location, & an external indication of its contents that allows it to be sensed, just like appearance & smell. We achieve that by changing the metaphor used for representing information today, which is the metaphor of files & folders - text documents piled inside folders, piled together. This metaphor leads to the current way people consume information - they read paper by paper, web page by web page, email by email. There’s no concept of space. So we needed a different metaphor, & came up with the concept of Buildings - these composite storage units we live in.

DN: I’m not following - files are the way we store information in computers. The Web is a protocol for sending information. The Matrix is using something else? Buildings?

UB: The existing metaphors are the way we imagine & reason with stuff - visualize them in our heads. The metaphor also influences the interface & functionality of objects. Files & folders, & how they are communicated over the Web was nice, but doesn’t cut the job anymore. We decided on a different metaphor, for the things that contain information. Buildings are composite storage units, located in a 3D space. In the Web Wide Matrix, everything is a building. They are composite, because they can be infinitely nested: for example, you have a building with multiple floors, each floor has apartments with rooms, each room can have closets & drawer cabinets. Each one of them has floors & drawers. Each one of them can have boxes inside it with further compartments & so on. We call each one of these nested containers Buildings, & their content is information. So we can have an atomic building, that just contains some information, & have external attributes that inform how it is sensed, & we have composite buildings, that have floors, & rooms, each one containing further buildings. This is different than files & folders, because every building has an address - it’s coordinates in the space, in respect to other buildings, & every building can be sensed.

(To be continued.)