Knowledge and Artificial Intelligence

Knowledge and Artificial Intelligence

Reading Time: 1 minute

Knowledge and Artificial Intelligence

I just hit the publish button on this one, and I think it may be one of my better articles in a while. It’s an attempt at a bigger perspective on what’s happening with things like Deep Learning (and other forms of machine learning and artificial intelligence), as well as Google’s Knowledge Graph and even its “Knowledge Vault.”

In this piece, I try to explain how these new technologies fit together. I also try to show how  these latest breakthroughs are still part of a long arch in human history, a continuation of our efforts to pull tacit knowledge from the biology of our minds and bodies and embed it as explicit knowledge into a new container. 

I hope you enjoy this one. I enjoyed writing it. 

#knowledge   #knowledgegraph   #artificialintelligence   #knowledgevault  

People I believe might be interested in this one because of past comments or posts: David Amerland, Mark Traphagen, John Kellden, Daniel Schwabe, Mark Bruce, RobotEnomics,  Jeff Jockisch, Jeff Sayre, Teodora Petkova, Alex Schleber, Mani Saint-Victor, Meg Tufano, Leland LeCuyer, Iblis Bane, Chris Lang, Deen Abiola, Susanne Ramharter, Yifat Cohen,  iPan Baal, Max Huijgen, John Blossom     

(If I missed you, and you want me to notify you on future posts like this, just let me know).

No comments

  1. Ping Monica Anderson Andrea Graziano Tyger AC 

  2. thanks for the ping John Kellden 

  3. Bookmarking so I can read after work. Very interested in this!

  4. Bookmark for me too. Sounds really interesting, Gideon Rosenblatt

  5. Danny Brown Thought this conversation would be enjoyable for you as well!

  6. Marking to read later. Thanks for the ping, Gideon Rosenblatt!

  7. Gideon Rosenblatt awesome post, really pulled a lot of strands together. Bravo. 

  8. Okay I’m back. Definitely pulled a lot of strands together as David said. Great piece, Gideon Rosenblatt

  9. Stephan Hovnanian It’s important to note #Freebase  that Gideon Rosenblatt mentioned. It is how Microsoft creates its knowledge graph on BING. But don’t think MS is down and out. They are building their own #KnowledgeVault  in collaboration with Cornell University. 

  10. Chris Lang are they going to publish it on CDs? (just kidding!)

    Remembering the ill-fated Microsoft Encarta project….

  11. Can’t wait to dive into this!!!

  12. Going to read now.

  13. Freebase is one of the ‘existing knowledge’ sources in most academic research at the moment. It’s free to use so not limited to Microsoft. Chris Lang 

    Wikipedia is sometimes used as an alternative source to match new extracted knowledge against human knowledge, but its lack of structure makes that difficult.

    Freebase is setup explicitly to make machine reading easy. A lot of Wikipedia information can be found there as well.

    Good article Gideon Rosenblatt !

  14. Ouch, Mark Traphagen, I worked with that Encarta team! My first marketing gig out of bschool…so weird to think how quickly all that value was destroyed by a little crowdsourcing…nobody on that team saw it coming.

  15. Sorry Gideon Rosenblatt didn’t mean to hit you personally!

    Yeah, crowdsourcing and the ubiquity of the web.

  16. Thanks for the ping Gideon Rosenblatt.  An interesting piece indeed.  I particularly liked the distinctions between implicit and explicit knowledge, and of course, the idea that wisdom is the result of experienced knowledge.  I personally also like to distinguish between intellectual knowledge, (knowing something is so) and emotional knowledge, (actually believing and acting as though something is so).

  17. Not at all, Mark Traphagen – just tongue in cheek…It’s just funny when people remember Encarta at all. It was actually a pretty amazing product, but an amazing example of failing to see the consequences of where the technology – and the market it was shaping – were both headed. Makes me think about all this knowledge extraction work, and what its impact will be today…

  18. Thanks Iblis Bane. Me too. And in fact, I think that emotional intelligence is likely to be the focus of much more of my writing in the future, as I think that it might well have something to do with the route to volition in software…

  19. Thanks Max Huijgen, and you nailed it. That research paper on the “knowledge vault” helped me to see how these knowledge bases aren’t just end goals in themselves, but that they are (perhaps more importantly) validators for the algorithms that will be the key to really scaling this work. They are the prior knowledge to which the algorithms’ results are compared. It’s a really important insight, actually…

  20. Oh great…software with volition…just what we need… 😉

  21. To clarify what Gideon Rosenblatt just said: the so called knowledge vault itself is not itself a human validated knowledge base but one of the machine created ones. 

    (your sentence could be misunderstood Gideon and before you know the KV is some mythical new beast 😉

  22. It’s funny, Max Huijgen, even the paper itself seems to be a bit confused over whether they see the KV as an end product or a technique. I’m rushing out the door right now, but I remember noting that seeming confusion within the way they described it as I was reading it…. 

  23. Your brain upon viewing objects and people in your field of vision identifies them, makes predictions about the outcome of interacting with them, then via your past knowledge and past outcomes of such interaction Viola!!! *NOW- Let’s take that to the next step* and see what #TheMachines  can do?

    A knowledge graph would be vastly more effective if it did not alert you to things you already know. Think about what Google knows about me and your article Gideon Rosenblatt:

    >>> Google knows I read the post, I scrolled down the page, spent nearly five minutes on it via Google Analytics and I have commented and shared.

    >>> I saved the PDF to my research folder.

    >>> Google knows I have written about this subject and wrote a novella of fiction about predictive information and devices that can display that.

    Now, in a predictive search result Google should:

    >>> Not display your article, unless the search contained pointers that I want to find it specifically: IE; your name or something unique about the title.

    >>> Google should not display social posts to the article unless I have been joining conversations about such already at the time of search or normally do engage in conversations around the same time of day and since I have been on G+ for 3 years and Google social overall for 7, Google knows Chris Lang.

    Now, when it comes to display devices like #AugmentedReality  apps on your phone, Google Glass, #WEARables  like the watches and Google Now, IF:

    If Google could store what I already know and then choose not to display things I already have a great knowledge of, and Instead identify things new to me? WHOA (in a bill and ted’s voice) then they would have something, would they not Gideon Rosenblatt and Stephan Hovnanian + Mani Saint-Victor?

    So, by creating a Chris Lang profile or entity, and associating my interest graph, my relevancy graph and my personal knowledge graph now we can see the future of this?

    BB later, I will give you a while to all stew on that for a bit :]


  24. To be able to discern the difference between a cat person, and a dog person, there is a test for wisdom.

  25. Hans Youngmann Social data can do that. Look at David Amerland and the words he uses about Nike, his cat. My, Nike, she (I believe) and verbs like sleeping, hiding, playing.

    I on the other hand, talk about #RockyAndRusty  as the, they, theirs, and them. (I also share Rottweiler pics at times)

    I used Rocky and Rusty as kittens in a novella, where my fiction character owned them and Google knows my address so they know I life with Margie D Casados and her sister here. So in an algorithmic deduction Google could produce a prediction that I do not actually own the kittens in my novel, that they are based in real animals but that I am only partially a cat person :] 

  26. Thanks Gideon Rosenblatt, will read asap. In the meantime, for those who are not aware, I point out the existence of DBPedia (, which is an extraction of tabular data already published in Wikipedia, and in some sense has similar information as Freebase, published in RDF format. It is also a very popular source for facts, and many research projects use it as well. It is the hub of the Linked Data Cloud (which also includes Freebase…) see,

  27. Gideon Rosenblatt thanks for the ping and the great read. Made my mind imagining the cognitive advantages of the machine knowledge – the neuroplasticity of our brains used at its best, connecting multidimensional dots.

    Also I was intuitively driven to thinking about the “knowledge within” (which I believe is close to the Wisdom in the Pyramid) and also the idea of emptiness, e.g. the meaning of nothing, or the meaning of pauses.

  28. Gideon Rosenblatt the KV is a source of unchecked ‘facts’ so its just input for the knowledge fusion techniques the paper describes.

  29. Mani Saint-Victor nice vortex. Saving it for rereading. Thanks:-)

  30. Max Huijgen – The Knowledge vault should at it’s core should NOT be swayed by public opinion. That is the flaw of the current big databases like Wiki and IMDB that Google returns when we search say: How old is Scarlett Johansson? (oddly the OS in #HER )

    Just because we as humans believe that something is true, a machine learning database would be built off first data gleaned from humans, then refined by making predictions based on that data, then 3d verification models of the data refining what #TheMachines  can prove to be true. It’s the basis for machine learning and it’s how we learn too.

    Or the famous quote “Write Drunk – Edit Sober” that is often mis-attributed to Hemingway and the image to go along with it of the writer on the beach at the end of #TheRumDiary  that is again wrongly attributed to Hunter S. Thompson.

    #TheMachines  as we call the servers in our fiction, need to know what is truth and what is not. And by showing the people that use these machines the truth =not a Google result built on social shares or incoming links= they make us better. Where the fiction becomes interesting surrounding these ideas is when machines realize that we can all be manipulated by what we believe.

  31. Oh my, another step talking up the mythical beast of the non existing knowledge vault. PLENZES™ – Birth Of The Machines

    (love sci-fi btw Chris Lang, but not this comment 😉 

  32. Max Huijgen Remember that this is the basis of scientific theory and machine learning. >>> Store data >>> Make predictions via the data >>> Verify data thru 3d models >>> Discard anomalies >>> Retain what is believed to be true. Then rise, repeat.

    As far as knowledge vaults, they are also fact, Cornell and Microsoft just launched their own joint effort. Just as Google did and is with Stanford.

    And just like image recognition, it is not that the Google algorithm is better, it is that they have trained the machines over a decade to do so. >>> Make predicitons >>> Verify data >>> Discard anomalies.

    We see this all the time and it started first with Facebook image tagging. #TheMachines  asked us: Is this you tagged in the photo? Have you noticed a lot of images in your G+ feed suddenly asking if it was you in the photo in the last week? Again, data verification for algorithmic predictions; Google training their machines.

    Where the flaw comes into play, is learning from humans.

    Dan O’Shea Just shared this:

    Again machines can learn faster than we can, but who are they learning from? As Traphagen and I highly disagreed upon a few days ago: Most information online is shallow, self promotional and self serving. Especially when it comes to social :]

  33. Ho, ho, stop Chris Lang _Knowledge Vault_ in the context of Gideon Rosenblatt’s article has a specific meaning. 

    It refers to a scholarly article I happened to mention to a few people. 

    What do you refer to?

  34. Max Huijgen I am referring to machine leaning overall. Same context, just my viewpoint. 

  35. Max Huijgen I gotta go put some chicken on the grill. But look at perception over machine knowledge. When they killed authorship last week, the blog post I shared used a tombstone image saying Authorship was born June, 7th, 2011. That is perception and popular belief. It is also very incorrect. It actually creates misconceptions.

    When in fact Authorship was born in late 2009 and began appearing in tests in Google results in 2010. It was based on our old Google profiles when Buzz was still alive.

    So, I pose that any Knowledge Vault that depends on what any expert says is flawed at it’s core. Hence a real KV, would have to start from scratch, the machines would have to learn on their own, and gain their own knowledge on their own. That is what  #BigData  is all about: Store everything, make sense of it later.  

  36. As long as we differentiate between the ‘KV’ currently popular on G+ and your imagined true vault of knowledge, we’re all fine. 

    Enjoy your chicken while I start looking for my bed Chris Lang 

  37. Yes Max Huijgen we currently depend on a “knowledge graph” of human curated knowledge in Google search. BING uses then same based on Freebase in their results. Both are founding “knowledge vaults” rather than depending on others. Good night buddy!

  38. Max Huijgen we need to find you a cause like world hunger or something 😉

    (Just kidding my friend, thanks for keeping us on our toes ;-

  39. Wow, Mani Saint-Victor, you always bring an interesting perspective. Thanks for the thoughts and for the pointer to that piece on “artificial curiosity” – that one’s going to take some digesting. And speaking of the “organization of information” – did you see that TEDx talk by Max Tegmark that martin shervington and Tom Eigelsbach shared last week? Very much on point with that.

  40. Knowledge graph exists, human curated knowledge domain exists plentiful and if you want to call them vaults of knowledge it’s all fine with me Chris Lang 

  41. If mankind wants to progress it needs to get rid of sloppy thinking Mark Traphagen otherwise the machines will surely take over according to Chris Lang 😉

    Gideon Rosenblatt wrote a nice introduction to the hierarchy of knowledge assembly. The way we can progress from facts to possibly wisdom. 

    Crucial steps are in the lower half of that pyramid of knowledge. The conversion from data to facts. If we go wrong there all is lost and there are no stepping stones. Not for wise thinkers nor for future AI machines. 

    We need to get the basics right so when I noticed that David Amerland referred to a  ‘knowledge vault’  based on a public presentation, I pointed Gideon to an academic source article on that same subject.

    From the discussion I can see it’s still taken out of context and seen as more than it is. As I’m fully aware of the power of semantics I try to ‘defuse’ the word ‘Knowledge Vault’ as it suggests future omnipotent powers.

    Off to bed and planning to to fight world hunger … 🙂

  42. To get a better idea at a higher level of how KV works, check this related project:

  43. Good link, Deen Abiola. Thanks. 

  44. Tegmark makes great arguments, arguments that are hard to refute. But how is his position substantively different than Platonism? Perhaps it’s not and that’s fine but it would be surprising if a leading mind in physics turned out to be thinking along Platonic lines.

  45. Weren’t the ideals somehow more detached and separate, John Wehrle? I’m no philosophy expert, but I sense that what Tegmark is talking about is information patterns embedded in the physicality. 

  46. By the way, Wayne Radinsky, you may find this post of interest. 

  47. Max Huijgen Just so I don’t misunderstand you what do you mean by “Sloppy thinking”…?

  48. Deen Abiola I loved the link you shared above. Especially the Probability mention in the first screenshot. That is exactly what I am talking about above.

    That is the result of the verification models I was talking about in machine learning. Taking data, running it thru a 3d model for verification,  returning a percentage of probability then if the probability is high, committing it to the database or updating a prior record or a null value.

    That is a very exciting action taken by a machine. 

  49. Great topic and loved you article Gideon Rosenblatt 

    How about defining Wisdom as the ability to deal effectively with stimuli?  Sure, “experience” is a great way to gain wisdom, but it is merely a “path to” wisdom.  It is neither “the only” path, not the result of going through that path… 🙂 

  50. Thanks for the ping Gideon Rosenblatt. Something is up with the Mentions… I didn’t get a notification, and I only realized that you pinged me when I saw Mani Saint-Victor ‘s share…

    Anyhow, my 2 cents – knowledge + experience = understanding.

    Not the same as wisdom, but other layer.. 

  51. That idea is great Chris Lang unless I’m looking for things that I’ve read before and need them again…

    In your example, I’ll never find the same result twice.. Cause theoretically, I already know it…. And we all know how great the human memory is… 😉

  52. Yifat Cohen I am suggesting that Google’s personalization, enlighten and illuminate, within a knowledge graph. On a thread with David Amerland I talked about how a search for “Omega” returned nothing but links to the watch maker. The next day the same search returned links to a company named Omega that make industrial sensors for machines to interpret our surroundings.

    I would guess, and that is all we can do when we postulate how Google really displays results, that since I did not click any of the watches links the day before, a personalization algo tried giving me new results. And yes, I did click thru to the Omega Sensors site.

  53. That’s really a cool experiment Chris Lang. 

    My challenge is when I’m looking for info that I found before and no longer find it in future searches… and I wonder if Google is confused about what I really care about since I’m managing other client’s accounts and thus am searching and commenting on things outside of my personal scope of interests…

  54. Gideon Rosenblatt congrats on this getting reshared by Yonatan Zunger!

  55. Mark Traphagen It was? 

    (It’s possible that I’m just forgetting due to jet lag, but it doesn’t look familiar to me)

  56. So sorry Yonatan Zunger. It was this one you shared:

    I’m on the road too 😉

  57. Oh, that explains things. Hadn’t looked at the ripples yet. Thanks for the heads up on the other post, Mark Traphagen, and for sharing it, Yonatan Zunger. I really like Koert van Mensvoort’s Next Nature site, by the way. Sometimes there’s really weird, but interesting stuff there. 

    Yonatan, I’m sure you’re busy, especially if you’re on the road, but you might find this piece interesting – kind of a big picture take on the intersection between deep learning, semantic tech and human knowledge. 

  58. Yifat Cohen Google already does what I was talking about today. I have seen three very different versions of a search for “Google Kills Authorship” in the last week. That could well be the human nature of the web at work, linking ect as new articles and link structures emerge.

    All of them have Mark Traphagen and Eric Enge at the top for the news feed. But the results under that news item have changed drastically each time I have searched it. (congrats on 12.5K in shares)

    What is very interesting on that search? There is not knowledge graph displayed for that term. I find that odd.

    Gideon Rosenblatt Congrats on getting shared by a Googler, that is just plain validation that your facts are correct here.

    While I talked about how machines validate facts and data above, I would definitely digress that there is still a human side.

  59. Chris Lang thanks for the congrats. Given the way that Google is currently assembling knowledge panels, it does not surprise me that they don’t pick up a news item as one.  We are probably a ways away from them doing that.

  60. Also remember that we shouldn’t assume that everything that is in the Knowledge Graph is automatically displayed in a KG box in search.

  61. Mark Traphagen I do miss G+ cards in search though, now again that has been replaced by contacts cards for those you know, along with a very detailed above the fold links. If you don’t have a record of the person in Gmail, then you get the knowledge graph.  

  62. Gideon Rosenblatt It’s taken me a week to find the time to read your article. All in all, it is a great read!

    As I read it, I kept on thinking back to this article of mine which I’m guessing I shared with you a few years back. [See, The HyperWeb: it’s All About Connections (]

    At the start of my article, I too mention the Knowledge Hierarchy, which I call the Hierarchy of Visual Information. In your piece, you make a strong argument for how knowledge has and will continue to evolve as machine learning via AI plays a larger and larger role in our society. In my piece, I focus on the ramifications of increasing connectivity to pieces of datum and how that will lead to increasing complexity, how that will take us from today’s Web of Data to tomorrow’s Web of Knowledge.

    Our two pieces are different looks at the same emerging phenomenon.

  63. Thanks Jeff Sayre, and agreed, there’s a similar theme here. One of the interesting questions you pose is to what degree the evolution of the web will continue to remain open. Based on what I’ve come to see through writing this article, I’m coming to the conclusion that there are huge network effects that will come into play when it comes to running the AI that is able to extract meaning from the web. And that means that there’s a very high probability of concentration – a kind of winner-take-all effect that cuts to the heart of the questions you’re asking in that piece. 

  64. Interesting adjunct, thanks.  The question of maintaining the openness of the web is one that troubles me at times as well.

  65. It’s already changing from that “wild west” feel Chris Lang.  When I think about how it has changed in the last 20 years, the mind boggles.  Thanks for the link.  Have bookmarked to check it out when I have a chance.

  66. I do indeed have it, Chris Lang. Haven’t gotten to it yet as I had a bunch of other books queued up to read before that, but will. I’d prefer to take it off this thread though. 

  67. Chris Lang, I should have been clearer: I wasn’t talking about deleting that link. I really don’t mind people mentioning work they’re doing (as long as it’s relevant to the topic, which I think your story is). I just don’t want to divert the focus of the thread here any more with a conversation about the book. That’s all. 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Sign up here for the latest articles. You can opt out at any time.

Subscribe by email:

Or subscribe by RSS:

%d bloggers like this: