Matt, you mention three applications *** face recognition or language translation or a self-driving car ***
Facebook has now pretty convincingly solved face recognition, via a simple convolutional neural net, dramatically scaled A self-driving car can, I suspect, be achieved via a narrow-AI integration of various components, without any general understanding Machine translation is harder than the above two areas, but if one is after translation of newspaper text or similar, I suppose it may ultimately be achievable via statistical ML methods. Although, the rate of improvement of Google Translate has not been that amazing in recent years -- it may have hit a limit in terms of what can be done by these methods. The MT community is looking more at hybrid methods these days. But all these problems are ones that have been focused on lately, precisely because they are useful and *can* be addressed fairly effectively by narrow-AI statistical machine learning methods on today's big data/hardware... If you picked other problems like being a bicycle messenger on a crowded New York Street, or writing a newspaper article on a newly developing situation, or learning a new language based on real-world experience, then you would find that statistical / ML methods aren't so useful... The goal with OpenCog is not to outdo CNNs or statistical MT on the particular problems for which they were developed. The goal is to address general intelligence... I know that OpenCog is not yet impressively functional. You are justified to ignore it till it's impressively functional, if it doesn't match your intuition. Just like early skeptics of rocketry, who didn't understand or accept the underlying theory, were justified to ignore early rocketry experiments until the rockets had actually blasted way up into the sky. I get the feeling you think that the ML technologies underlying today's MT and face recognition can scale up to take care of all human-intelligence-level capabilities, if we just throw more machines at them, and improve the algorithms a bit. If so, I can't prove you wrong, but my intuition strongly disagrees. You ask how OpenCog could help you make a better Watson? Actually that's an easy question to answer. Watson is a fairly specialized expert system, and its ability for uncertain inference and generalization is not so strong. I suspect that integrating OpenCog's inference capability with Watson would result in a much better Watson. But this would be a fairly large R&D project, not an immediate "plug and play" integration. And Watson is closed source so I'm not sure this is something I'd be psyched to pursue anyway... -- Ben G On Sun, Mar 23, 2014 at 9:52 AM, Matt Mahoney <[email protected]> wrote: > On Sat, Mar 22, 2014 at 5:40 PM, Piaget Modeler > <[email protected]> wrote: >> The group is debating whether to only sell and at what price, or to give >> away a "free" >> Community version, and charge for support and the "Enterprise" version. The >> fear is that >> a free Community version will never generate any real sales on support on >> the Enterprise >> version. > > I don't think that is the issue. Software is not something you sell > any more. We give away software because that is how you sell yourself. > You can develop better software if you make it free and open source > because you get free testing and feedback from users. It is the > approach I have used successfully in developing data compression > software. I am glad that OpenCog uses this approach too. > > I think the issue with not getting offers on OpenCog is that it > doesn't do anything yet. As I understand it, OpenCog is a complex > knowledge representation system with hundreds of classes, a > "programming language" for developing AI. But it lacks a knowledge > base and lacks testing on real-world applications. It is not clear to > me how it could be helpful to do, say, face recognition or language > translation or a self-driving car. How would OpenCog save me time in > developing a competitor to Watson? > > AI has three very difficult and expensive components: software, > knowledge, and computing power. IMHO, software is the least expensive > part. OpenCog doesn't address the larger obstacles. In order to > convince me that it is useful, it needs to be tested on a real world > problem with all the components, like solving some problem using > millions of images on thousands of processors. > > -- > -- Matt Mahoney, [email protected] > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/212726-deec6279 > Modify Your Subscription: https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com -- Ben Goertzel, PhD http://goertzel.org "In an insane world, the sane man must appear to be insane". -- Capt. James T. Kirk "Emancipate yourself from mental slavery / None but ourselves can free our minds" -- Robert Nesta Marley ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
