http://agi-conf.org/2010/wp-content/uploads/2009/06/paper_6.pdf I refuse to call papers papers anymore, they are just text.
Let me override this discussion: There is only 3 ways to test for AGI: 1) The algorithm/body works like us. This won't tell us so much that it works good though. 2) Scores/compresses good on a large diverse set of tasks. This does tell us. 3) Generates useful outputs on a large diverse set of tasks ex. it generates good looking text/image/tool/job completions. This too, just as much sometimes. Wang seems to confuse 2 things together as the same thing, and try to combine them too: "Now we see that the empirical approach and theoretical approach of evaluation actually depend on each other for the meta-evaluation." No, generating solutions, and generating the actual procedure in real life to DO it (collect data/ find a cure/ give needle/ etc), are the same thing, we recognize them as a good thing, theories and working cures for humans made by AI are BOTH as useful, a cure for cancer of AI may seem untested crap until proven by a working invention but this isn't the case, ideas can be really accurate if you try. Ideas and inventions are same thing. This evaluation is the #3) I listed above! It's subjective, it sounds great as a completion, but YOU - not Lossless Compression, have to check it, very finely, to make sure it is the best solution, and must check thousands of tasks (solving building higher towers, solving cancer, inventing better storage devices). This is very wobbly, especially if it has no body to actually implement a full working idea to solve cancer and can only make semi-discoveries. That's, why we need to test for SOME accuracy, not that it can solve cancer fully. Way #3) is very wobbly, it's good but not actually telling you it is finding patterns in data. *We recognize/ predict the same way. **Humans wrote text, so it therefore is extremely **patterny** as humans are, because we predict text and our generated text we wrote is what we recognize/ predict. It is the inner patterns of a mind written out, or drawn if use images made by us. The world too does this, physics makes patterns. Our data is actually more **advanced** in evolution though.* "NARS is based on the belief that “intelligence” is the capability of adaptation with insufficient knowledge and resources. This belief itself is justified empirically — the human mind does have such capability" This is just pattern finding. you take only some snapshots of the world using only some diverse sensors and so on and can get something much more approx./ like a full atom by atom brute force simulation. Intelligence is using patterns/ experiences and making your homeworld into a pattern fractal to increase prediction accuracy and becoming a pattern (immortality), things that die die and things that are smart clone and last longer and outlive others. The future world will be a cooled-down metal dark less-dense airless gravityless place that only acts when there is danger from external inputs. Life is whatever acts like a statue/ metal, rocks are Life, not aliens only. Women and donkies and rocks are not sexy, we are born with the reward to predict/ see/ be near them because we need clones to help us with work and survive deaths of "employees" in the homeworld, parent DNAs combine to try likely good traits ex. man+wings, instead of just random mutations. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T466c51e997fec20a-Md79f88089c97c918b639a8fe Delivery options: https://agi.topicbox.com/groups/agi/subscription
