Four short links: 9 Mar 2009

Hardware, open source, and AI today:

  1. Geek Tour China 2009 — how did I miss this? Bunnie Huang has led a tour of China manufacturing for hardware hacking geeks. Read the blog posts from participants: here, here, here, here, and here. Just go ahead and add these bloggers to your feed reader: sweet sweet candy they post. My favourite: American Shanzai, asking where are the USA hackers like the Chinese who make working phones out of packets of cigarettes? But read the posts for giant single-digit LED clocks, markets of components from torn-down phones, and 280km of velcro/day machines.
  2. Open Source Hardware Central Bank — an interesting idea to fund the manufacture of larger runs than would be possible with self-funding, so as to achieve modest economies of scale. “Looking at Open Source Software, it’s a thriving ecosystems of communities, projects, and contributors. There are a few companies, but they mostly offer “paid-for” services like consulting, tech support, or custom code/build-to-order functionality. I’d like the same for Open Source Hardware. I’d like the money problem to go away for small contributors like me and others. And I’d like to help guys like Chris and Mike and Mark and David and Jake build more cool stuff because it’s fun.”
  3. Wolfram Alpha — everyone is skeptical because it smells like AI windmill tilting mixed with “my pet algorithms are the keys to the secrets of the universe!”, but it’ll be interesting to see what it looks like when it launches in May. “But what about all the actual knowledge that we as humans have accumulated? […] armed with Mathematica and NKS I realized there’s another way: explicitly implement methods and models, as algorithms, and explicitly curate all data so that it is immediately computable. […] I wasn’t at all sure it was going to work. But I’m happy to say that with a mixture of many clever algorithms and heuristics, lots of linguistic discovery and linguistic curation, and what probably amount to some serious theoretical breakthroughs, we’re actually managing to make it work. Pulling all of this together to create a true computational knowledge engine is a very difficult task.”
  4. Open Source, Open Standards, and Reuse: Government Action Plan“So we consider that the time is now right to build on our record of fairness and achievement and to take further positive action to ensure that Open Source products are fully and fairly considered throughout government IT; to ensure that we specify our requirements and publish our data in terms of Open Standards; and that we seek the same degree of flexibility in our commercial relationships with proprietary software suppliers as are inherent in the open source world.” Great news from the UK!
tags: , , ,
  • Wolfram has experience (see his web site) assembling encyclopedias that are interactively “computable” – his firm maintains an encyclopedia of math.

    Therefore, his firm has some interesting “core strengths”. They have some clues about taxonomies, cataloging, and, as he puts it, “curation” generally.

    I think he’s saying that he was one day guesstimating the minimum size of a useful “general knowledge” computational ‘pedia and looking at his business’ experience for guesstimated costs of production and came up with “it’s worth a shot” – and now he’s reached a point in the experiment where “um… this is coming out pretty good.”

    So he’s got a “formula” – a business method, of a sort – for hiring experts and giving them some training and setting them loose to add data to this new ‘pedia following some “curatorial” discipline and practice. On the other side, he’s got a functional/lispish symbolic computing environment from which he derives the syntaxes that those experts use to express the data they collect. And he’s got a bunch of hackers organized around collecting useful algorithms to run on that data set – with an emphasis on “composability” of algorithms – and hackers working on UI.

    So it’s got a very narrow domain “natural language” part trying to design a query language that is technically simple to parse and precise in meaning but that is a close match to how people normally talk. Queries are compositions (“output of X becomes input of Y”) of various basic algorithms (e.g., “solve this equation for Z looking up other values as you need” for the earth-sun distance example).

    NKS, my guess is, impacts this project partly in convincing him of the irreducibility of the goal to solutions which eliminate the experts and in influencing his tendencies in business modeling and management style.

    Silicon Valley traditional model firms would have trouble undertaking something like this because it depends on having “experts” to “curate”. Traditional easy-in / easy-out start-ups and their analogous projects in established firms tend to shun reliance on experts like the plague. In pursuit of robustness, normal start-ups and big projects tend to plan around the assumption of most of the grunt work being done by inter-changeable, least-common-denominator tradesman programmers. So, nobody in that market would seriously contemplate a project that required a staff of mainly experts. That tactical limitation of most firms is what gives Wolfram his opportunity here.

    So: A programming language for queries that resembles natural language well enough that people can guess a correct syntax for many queries without any training in the language per se. A query engine that interprets queries in that language and that is “extensible”….

    I’ll say it differently:

    In Mathematica he has what is known in the trade as a “proof assistant”. You feed it axioms. You feed it rules of reasoning. It helps fill in steps in mathematical proofs, it helps to carry out tedious steps, and it helps to visualize mathematical structures.

    Ok, now, realize: Mathematica freely mixes what ordinary (non-mathematician) people would think of as number crunching and logic. It’s as happy to tell you the zillionth digit of pi as it is to tell you that Socrates must be mortal given that he’s a man, etc.

    So: how about instead of looking for really esoteric math theories we feed it “axioms” from encyclopedias, and news stories, and official publications, etc. etc. We can screen-scrape that stuff at amazing rates if we put our minds to it to work efficiently and focus usefully.

    So now we have a proof assistant armed with all of this real-world data and there is a huge, huge space of possible “theorems” to be deduced from that data using pretty banal rules (like, how to convert units or how to solve a simple algebraic equation or what the equation is for the phase of the moon).

    In the hands of an expert that’s quite a powerful tool, assuming we have experts keeping the axioms up to date and always adding and improving the “extensions” that teach the proof assistant how to answer various questions.

    One step further, then: devise a “natural” (in the sense of requiring little or no training to at least get started) query language as an interface to this newly armed proof assistant.

    It’s beautiful, simple (conceptually), richly complex in a horizontal way, an interesting challenge financially — it sounds very promising.


  • Sorry to self-follow-up but a really lovely irony just struck me:

    By building around experts and curation, and taking the NKS-indicated stance that the goals are irreducible (experts and curation necessary) Wolfram is actually anti-AI.

    This isn’t AI windmill tilting.

    Google – page rank and AdSense – that is AI windmill tilting.


  • I’ll shut up after this. Really.

    Version 2.0 of Wolfram’s thing will start to add subjunctive case. It’ll be like the “what if” aspect of spreadsheets but applied to a much broader set of reasoning techniques (not just equations over spreadsheets). “Assume the oceans rise 1 foot. Now, in this world, what is the total area of Manhattan? How long does it take a canoe to travel shore-to-shore from New Jersey? Ok, assume the oceans rise 2 feet – how do the answers change?”


  • Falafulu Fisi

    It would be interesting to see what Dr. Wolfram’s system can comes up with in terms of knowledge inference for his new product.

    I believe that the man is up to something exciting. As a Mathematica user myself for over 10 years now (mainly to proto-type my numerical algorithms before coding the tested algorithms in Java), I can see his (Dr. Wolfram’s) brain in his product, ie, Mathematica, which is an amazing tool. I would say that Mathematica is the best commercial CAS (computer algebra system) available out there today for doing symbolic computations.

    I am looking forward to see what this product can do when it comes out.

  • > Geek Tour China 2009 – how did I miss this
    How *did* you miss that? Apparently I should feed you with more links… :-)