Audio Of The Week: A Decentralized AI Platform

My friend Gordon texted me this podcast and said “AI, blockchain, and homomorphic data. Trifecta!”

I gave it a listen and indeed some very interesting concepts are discussed in this one.

#blockchain#crypto#machine learning

Comments (Archived):

  1. RealBusinessOpportunity

    This a huge business opportunity that could provide real value to people. Fred should ditch the crypto hucksterism and focus on this stuff.

    1. Donna Brewington White

      Thank you. 😉

  2. Mike Cautillo

    Definitely excited to see where this goes…..it would appear that AI and blockchain could give birth to some interesting projects. Endor out of MIT is a pretty spectacular one that I’m certainly keeping my eye on also.

    1. jason wright

      Endor.Why does it need an ICO? Is a token essential?

  3. Joseph K Antony

    This is so timely. Was mentoring an early stage start-up almost exactly on these lines, except that they would focus initially on consumer days and generate equivalents of credit rating among others. This seems to have covered enormous ground.

  4. sigmaalgebra

    Let’s see: In the back of my refrigerator I may have some lettuce, tomatoes, and boiled chicken. Each is at least one year old and looks just awful. Not even a kitty cat would eat it. But, presto, bingo, if I just combine those three, then I can have a salad fit for the best restaurant in Paris of, say, 1900? Somehow the magic is in the combination?Uh, for AI, nearly all of it is intuitive guesswork that sometimes appears to work. Otherwise it is both in theory and in practice, a long way short of multivariate statistics back to, say,L. Breiman, et al., CART — Classification and Regression Trees, and to the early days of, say, SPSS (software, Statistical Package for the Social Sciences) and SAS (software, Statistical Analysis System) and lots more, e.g., the old IBM SSP (Scientific Subroutine Package). The relatively well polished texts on such statistics fill not just a bookshelf but a floor to ceiling bookcase. For just one reference, noticed an hour ago at Hacker News, downloaded and scanned quickly, there isCosma Rohilla Shalizi, Advanced Data Analysis from an Elementary Point of View. athttp://www.stat.cmu.edu/~cs…with an okay view of some statistics now popular.So, back to salad making, sure, to heck with fresh lettuce, tomatoes, and shredded, boiled chicken and use the rotten stuff AI, etc. instead?AI has some work that is new and good, but the good is nearly never new and the new nearly never good. AI has had some success in curve fitting with millions of variables and billions of data points. Okay.If you have some questions and some relevant data and want to know how to manipulate the data to get valuable answers to the questions, then okay. Then any such manipulations will necessarily be mathematically something, understood or not, powerful or not. So, for more powerful manipulations, especially ones can have some confidence in, proceed mathematically. The highly polished materials on applied math fill rows in the research libraries.New material from AI? After the hype and the salt water, there’s not much left.Go do some applied math or try AI and likely just waste time, money, and energy on pop culture hype. Choice is yours.Yes, applied statistics and nearly all the rest of applied math needs computing, from a little up to more than available so far. But that need for computing does NOT mean that the computer science departments are the best source, or even a good source, for the math.

    1. Lawrence Brass

      It’s the continuous decrease in computational and data storage cost that defines what algorithms go into ‘production’. More than a rediscovery it is a threshold that is crossed by a particular algorithm or group of algorithms to become cost effective.Evolutionary, not revolutionary, in my opinion. Technology is alive and it has its own momentum.Of course there is art and craft in the implementations and execution, that can make a difference.Interesting CMU link, Thanks.

      1. PhilipSugar

        I agree and disagree. You are right that as technology evolves it makes things that were known but not possible, possible: Evolution.What is revolutionary is how computational and storage continually evolve, and how many things they touch.For instance you make steel which makes skyscrapers big bridges, etc possible. Yes certainly better than 100 years ago, but not that much maybe a factor of 2.But from your avatar I am guessing we are similar in age. I’ll start at my business career not high school days when I remember using cassette tapes in my Apple III can remember buying a server and a hub in 1993. The hub cost something like 42k and had a 10mbs total capacity. 3mbs actual max. The server (Compaq) had a raid array at the time tons of memory I forget the specs, but I think it was on the same order. Because I remember for an office you spent $2k a head for networking, $2k a head for servers, and $2k a head for computers.25 years later you have a non blocking 48 port Gigabit Switch, that is 16,000 times faster, a server with a solid state memory raid measuring in the terrabytes, the the processors are so fast you don’t even worry because they are not the throttling factor. Total cost 4 times less.So you are talking about cost per performance of at least 50,000 times better in 25 years. And you can’t really compare that because 25 years ago it was not possible.I dropped a server that was not working out of a rack (yes, I’ll never hear the end of that one) I said I’d reimburse. We bought that thing 5 years ago for $5k, now worth $250.That IS revolutionary. Which I define as something you could not imagine.The fact that we can process with complex rules: 5B+ interactions, 1B+ financial transactions, on 100mm+ people in under 500ms response time would have been impossible to imagine.The fact I have a high speed networked super computer in my pocket that I can use to Skype you? Un-imaginable.

        1. Lawrence Brass

          It is a pleasure to disagree with you Philip :)But don’t get me wrong, I am constantly marveled and impressed by technology and human progress and by tagging something as evolutionary I mean no detriment at all.I am particularly impressed by the advances in materials science which often express later in capacity increases in the machines. I am also grateful to the comparatively few people that make this possible: the explorers and pioneers, the scientists, the entrepreneurs, the guys at the labs who gave and give their lives to their obsessions for the benefit of human knowledge.But taking out the emotion, these advances are “just better silicon” and we the people integrating with machines in unexpected ways. The social (networked) implications I must agree, are revolutionary.Revolutionary for me is something that comes with a paradigm shift included. So I get your point and we will have to work on a common definition or choose another word. :)Or maybe I have grew an optimist. I expect no less than immortality as the ultimate outcome of human technological (re) evolution. Possibly not for us but for the lucky generation that achieves it and, if our social structures haven’t changed enough by then, for those that can pay for it.For the record, today I am using a version 5.8 carbon based sleeve with code enhancements.( yes, we binged Altered Carbon last week and I have an.. er.. homomorphic device somewhere )

          1. PhilipSugar

            The pleasure is mine as well. I like thought out respectful banter, it makes me think.Let’s take YouTube as an example. Streaming video was “old” technology. But the advent of broadband and cheap storage and servers made it possible for them to come up with a model where you upload and view for free.So you could say evolutionary. BUT There are 1.3B people viewing 60B minutes of uploaded videos, 5B times a day. I’d call that revolutionary.I guess you can tell it’s a bit of a pet peeve of mine. I find many times when a technologist sees something they say oh, that is just an extension of that. Well it is, but it doesn’t make it less exciting or downplays the achievement of the team that built it. My two main issues are:Really, so why didn’t you do it?? Now you are jealous??Many times technologists lament that they are treated as second class citizens (Dilbert was based on this) it starts with this behavior.

  5. jason wright

    So this is the direction Numerai will need to go in.p.s. I didn’t know (until now) that Twitter now has app 2FA account security (e.g. Google Authenticator).

  6. Vendita Auto

    Thanks for the share most interesting to open my mindset in bite size pieces

  7. Twain Twain

    The most interesting question happens at 41.00 about how to value different data sets from different agents. How do you partition rewards and which parts of the information contribution matter?

  8. Lawrence Brass

    Very interesting. Thanks.For the coders, take a look at Part 5 in the following Andrew’s post, on homomorphic encryption: https://iamtrask.github.io/

  9. Jay Swartz

    Great recording. This is the type of technology I’m working on to build Provenance systems at Blackbox.ai. There are still challenges with real world business challenges, but the Deep Mine community is moving in the right direction.

  10. Kevin Hill

    Why the hell do you listen to these people? Every aspect of blockchain makes the AI problem harder, not easier. These people are selling you a load of crap and making you look like a fool.Fred, you’ve been a really strong source of knowledge for me for years, but this is like watching someone go through addiction. You seem unwilling to accept any input that contradicts your current beliefs.If you want to see what real problems look like in AI, I suggest you watch https://news.developer.nvid…or for a more academic framing of the same challenges: http://cs231n.stanford.edu/…And then think about how antithetical blockchain is to any of these issues.