An Open Letter to the AGI list
Matt- firstly, well said. Thanks for that perspective. To add, I would like to light a candle for theoretical researchers that are designers. This, as opposed to researchers who jump straight into coding modular tests, and/or designers who do purely-academic work and publish with the help of universities and other sponsors. In other words, independent researcher/designers such as myself and a few, others here. In general, we find ourselves out on an IP limb. In my view, contributors on the list tend to focus on vague-ish definitions, or hard experimentation, albeit for niche development of a version of AGI. And so often, one sees the "argument" either collapse, bog down, or go full circle to a re-examination of the term AGI. In retail services, this may even be referred to a "spinning". I would caution against any, established researcher to discount, or seemingly minimize the research of others. It is not part of the constructivist research approach, which is what I hope we're all trying to achieve on this list. After all, no-one here has found THE answer to an AGI architecture yet. I stand open to correction. This is where holistic researchers such as myself come into the picture. The research I do focuses on developing an architectural model of an AGI platform (all AGI possibilities). In theory, such an holistic platform would be able to position, and seamlessly integrate with all other, AGI endeavours, or niche achievements. As such, my work enhances, and is being enhanced by, the work of others. Your output serves as my input. I often do not understand the content of what some of you are sharing. My field of study is not EVERYTHING either. With some of your sharing, I even need to revert back to a dictionary and often go read a lot more than I have time for. But, I do, because I'm passionate about this topic. Part of my AGI journey is to try and bridge the semantic and experiential differences I encounter. Once the architectural principles of any of your output are sufficiently clear to me, I can use my own language of reference and start integrating it with the logical whole in my mind. To make sense of such an approach, I need to document my progress, and test it on the hand of existing and emerging theory. In addition, I need to "future proof" all my designs by attempting to peer 40+ years into the future. By comparison, when a researcher is swallowed up by the acute detail of developing an AGI, computational model, my perspective may seem to be an incumberence. However, bearing mind, whatever any of you produce, at some point it has to be seamlessly integrated with an holistic AGO architecture. I think such a mindset objective is most critical for the progress of AGI development in the world. In my experience, the greatest hindrance to the global development of AGI is our own minds. This is a view recently stated by a contributor. We try and make copied of AGI from our own minds, as if we are the super intelligent, the only role model. This may not be the case at all. I have learned that, as soon as I've reached a boundary of research, a doorway either extends the boundary. and/or another none opens up. I think research in AGI is a work in progress. Research requires a holistic framework, which we already have in computational-engineering approaches. Further, research also requires holistic architecture, which probably exists for some, but that may be of such strategic value, it is definitely not going to be shared. I plan to focus my research more pertinently from now on, to start integrating the understandable work here into the architecture I am evolving. I have so many papers still to go through, and would appreciate a fellow, holistic architect to join up with me in this goal. I also intend to share my results more progressively and sensibly with the community. Surely there must be lurkers here who has as sole intent to try and grab what others are doing for selfish purposes, but that one even finds at any conference in any case. It should not be the factor to stop AGI-networking possibilities on this list. May I then request, for the list, please could there be a sense of professional courtesy in terms of offering a degree of professional credit to individual contributors. Many of us have spent decades and tens of thousands of dollars on this research. We do not wish to just give it away like idiots, but we do wish to share in the interest of progressing AGI. Please try and institute a fairness to referencing contributors where possible, to quote accurately and correctly in full, and to give intellectual property credit when the work of contributors are being used. Else, this would have to turn into an academic site, forcing researchers to first spend a year writing and submitting white papers for every, little step of progress being made. The list would prove less useful and we'll probably never get anywhere fast. If this is indeed feasible, I'd be happy to share my body of research openly for what it is worth. I'd also appreciate all critical reviews and comments on my contributions. What say you? Sincerely Robert Benjamin ________________________________ From: Matt Mahoney via AGI <agi@agi.topicbox.com> Sent: Wednesday, 13 June 2018 9:02 PM To: agi@agi.topicbox.com Subject: Re: [agi] Anyone interested in sharing your projects / data models Among the many AGI designs and proposals mentioned in this thread, it was refreshing to see some actual results from Peter Voss's Aigo. (Also entertaining as my Alexa was listening and answering back while I played the demo videos). Experimental results are a lot more work to obtain than ideas, which is why most publishers and reviewers require them. I realize this is difficult for AGI, which I guess is why 85% of the papers accepted to the AGI conference still lacked a results section the last time I looked. My last 20 years of research can be summarized as finding experimental evidence (not proof) supporting the following hypotheses: 1. The best language models are based on neural networks. 2. Intelligence grows logarithmically with CPU time and memory. 3. Automating all human labor with AGI will probably cost $1 quadrillion. We recently learned that the best vision models are neural networks. My work suggests this is true of language too. It is based on testing thousands of versions of 200 compression programs since 2006 on a 1 GB text benchmark, found at http://mattmahoney.net/dc/text.html Text compression measures text prediction or modeling by adding a coder, which is a solved problem. The top models use dictionary preprocessing to convert words into tokens followed by PAQ style compression predicting one bit at a time using ad-hoc context features and shallow neural networks. They implement essentially toddler level language models with hard-coded lexical features, proximity based semantics and flat (n-gram) simple grammars and dictionaries sorted by grammatical role (i.e. grouping "monday" with "tuesday" or "brother" with "sister"). The models so far lack advanced grammars necessary to understand math, software, or complex sentences. Prior to my work on PAQ based compression, the best models were PPM (prediction by partial match) until about 2003. PPM predicts bytes rather than bits using the longest matching contexts. I started work on neural based compression in 1998, 5 years before achieving this result. The second hypothesis has several caveats. By intelligence, I mean text prediction accuracy. I show that human level prediction (which we have not yet achieved) implies passing the Turing test. Not everyone accepts the Turing test as general intelligence since it lacks non-text based processing like vision, music, and robotics, all requirements for AGI or automating labor. Also, my tests (with the same benchmark) only show a logarithmic trend over the range of a few bytes up to 32 GB and 1 to 10^6 operations per byte. If we assume that 10% of the human brain is used to process language, then the goal figure is 10^13 bits of memory and 10^14 operations per character. For my third hypothesis, please note I am estimating the cost of several billion human level intelligences, not just one human level AGI. The two pieces of evidence I produced in support of my claim are: 3A. My 1998 masters thesis where I showed the scalability and robustness of distributed indexing using computer simulations. Distributed indexing is an essential feature of an AGI design consisting of lots of independently developed and competing narrow AI such as my 2008 proposal. (The thesis is here: https://cs.fit.edu/~mmahoney/thesis.html ). 3B. I showed that recursive self improvement in a closed environment (boxed AI, sometimes proposed as a shortcut to AGI or a singularity) is impossible. http://mattmahoney.net/rsi.pdf Of course none of this disproves the possibility of other, less expensive routes to AGI. But logic based AI is probably not one of them (per my first result) and early progress does not predict success (per my second result). -- -- Matt Mahoney, mattmahone...@gmail.com<mailto:mattmahone...@gmail.com> Artificial General Intelligence List<https://agi.topicbox.com/latest> / AGI / see discussions<https://agi.topicbox.com/groups/agi> + participants<https://agi.topicbox.com/groups/agi/members> + delivery options<https://agi.topicbox.com/groups> Permalink<https://agi.topicbox.com/groups/agi/T731509cdd81e3f5f-Mda6e59327c21a47a77423b17> ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T731509cdd81e3f5f-M082efc12e9b986a6f6549b76 Delivery options: https://agi.topicbox.com/groups
