| … | … | @@ -2,7 +2,7 @@ Evidence |
| 2 | 2 | ======== |
| 3 | 3 | A primary intellectual activity is the gathering of evidence to |
| 4 | 4 | determine the nature and structure of things. It is closely related to |
| 5 | | the task of discrerning similarity, and of reasoning by analogy. |
| 5 | the task of discerning similarity, and of reasoning by analogy. |
| 6 | 6 | This git repo is a rumination of these tasks, plus an attempt to express |
| 7 | 7 | these ideas as functional code, expressed as OpenCog [Atomese](https://wiki.opencog.org/w/Atomese). |
| 8 | 8 | |
| … | … | @@ -77,7 +77,7 @@ The judgement call is often hierarchical in nature. For example: do the |
| 77 | 77 | father's and mother's names match? Well, perhaps they do, but there is |
| 78 | 78 | some alteration in spelling: Maria vs. Mary, Elizabeth vs. Betty. One |
| 79 | 79 | can assign some probability or confidence in the belief that these are |
| 80 | | the same parents. To butress that belief, one can look at siblings: if |
| 80 | the same parents. To buttress that belief, one can look at siblings: if |
| 81 | 81 | two children are born three months apart to the same mother, then |
| 82 | 82 | this is a piece of negative evidence that absolutely demands a rejection |
| 83 | 83 | of a match. That is, of course, only if the recorded birthday is |
| … | … | @@ -89,9 +89,9 @@ agglomeration of the supporting evidence. A preponderance of evidence |
| 89 | 89 | is desired. This can be taken as a mathematical mean or average, but |
| 90 | 90 | it can also be some more complex algorithm extending across the |
| 91 | 91 | structure: for example, "voting in new members to the club of facts". |
| 92 | | In comp sci, similarity is measured with cosine distance, Jaquard |
| 93 | | distance, mutual information, Kullbeck-Liebler divergence or any one |
| 94 | | of dozens or hundres of different similarity measures, which can be |
| 92 | In comp sci, similarity is measured with cosine distance, Jacquard |
| 93 | distance, mutual information, Kullback-Leibler divergence or any one |
| 94 | of dozens or hundreds of different similarity measures, which can be |
| 95 | 95 | aggregated and weighted in a large variety of ways (random forests, |
| 96 | 96 | collections of experts, membership clubs, etc.) |
| 97 | 97 | |
| … | … | @@ -101,11 +101,11 @@ often having the same name, but not always; often having the same |
| 101 | 101 | timestamp, but not always; often appearing in the same directory, but |
| 102 | 102 | not always; often having the same size, but not always; often having the |
| 103 | 103 | same content, but sometimes having all-zeros where the data should have |
| 104 | | been. Or perhaps MP3's with missng bytes, or images with steganographic |
| 104 | been. Or perhaps MP3's with missing bytes, or images with steganographic |
| 105 | 105 | watermarks. The challenge is to find the copy of the file that is least |
| 106 | 106 | likely to be corrupted, and is most likely to be "correct". Despite the |
| 107 | 107 | fact that all data is directly available from the operating system, |
| 108 | | using very pure and well-defineed, structured software interfaces, this |
| 108 | using very pure and well-defined, structured software interfaces, this |
| 109 | 109 | turns out to be a hard problem. |
| 110 | 110 | |
| 111 | 111 | Project Goals |
| … | … | @@ -119,7 +119,7 @@ academia, dating back to the 1960's. |
| 119 | 119 | |
| 120 | 120 | Here, the idea is modified: The jigsaw connector tabs are not strict |
| 121 | 121 | yes/no matches, but have some confidence or likelihood. This confidence |
| 122 | | or likelylihood is obtained from hierarchical subrelations: do |
| 122 | or likelihood is obtained from hierarchical sub-relations: do |
| 123 | 123 | sub-pieces also fit? This confidence may also be obtained by spreading |
| 124 | 124 | over the local neighborhood: is coherence maintained over not just |
| 125 | 125 | nearest neighbors, but also next-nearest-neighbors, etc.? In Link |
| … | … | @@ -132,10 +132,10 @@ they come from a "lexis", a dictionary of possible forms created by |
| 132 | 132 | professional linguists. That lexis is presumed to encompass the |
| 133 | 133 | morpho-syntactic structure of language. The lessons of structural |
| 134 | 134 | linguistics and syntactic analysis from the 1990's-2010's is that such |
| 135 | | dictionaries are always incomplete, and that no hurculean effort by grad |
| 135 | dictionaries are always incomplete, and that no herculean effort by grad |
| 136 | 136 | students can ever come to a conclusion. The task is overwhelming. At the |
| 137 | 137 | same time, transformers, such as BERT, and now, more generally DL-NN |
| 138 | | LLM's have shown that statistica techniques, taken from corpus |
| 138 | LLM's have shown that statistical techniques, taken from corpus |
| 139 | 139 | linguistics, are far more adept at encoding the structure of language. |
| 140 | 140 | |
| 141 | 141 | Thus, part of the problem is to obtain structural factoids. That is, |
| … | … | @@ -144,7 +144,7 @@ environment. This requires the perception of not just individual data, |
| 144 | 144 | e.g. a single birth record, but also the perception of the "shape" of |
| 145 | 145 | the structure: birth records have a date and a name; but the date may |
| 146 | 146 | only appear at the top of the page, with a list of names below it. Some |
| 147 | | classes of birth records provvide additional info, others do not. An |
| 147 | classes of birth records provide additional info, others do not. An |
| 148 | 148 | extreme form is the obituary: all of us know what to expect; but the |
| 149 | 149 | actual informational detail can be spotty or it can provide a wealth of |
| 150 | 150 | info. Roughly speaking, no two obituaries have the same "shape", even |
| … | … | @@ -194,7 +194,7 @@ homunculus. |
| 194 | 194 | Although the word "structure" here is meant to evoke a sensation of good |
| 195 | 195 | olde-fashioned symbolic AI, there is no presupposition of avoiding |
| 196 | 196 | neural nets. Indeed, the structure of a given fixed DL-NN is described |
| 197 | | symbolicly: its written in ROCm or CUDA, its written in Python or |
| 197 | symbolically: its written in ROCm or CUDA, its written in Python or |
| 198 | 198 | TensorFlow. These are structural, symbolic languages. Insofar as there |
| 199 | 199 | is progress at the meta-level, then it is in the use of vibe coding to |
| 200 | 200 | design transformers. Which is indeed an interesting approach. But the |
| … | … | @@ -206,8 +206,8 @@ Methodology |
| 206 | 206 | Little has been done or decided as of yet. However, it is expected that |
| 207 | 207 | the general theoretical foundations, and the actual algorithmic |
| 208 | 208 | explorations will be based on the following systems, theories and |
| 209 | | infrastrcture: |
| 210 | | * [Atomese](https://wiki.opencog.org/w/Atomese), a strutural description system. |
| 209 | infrastructure: |
| 210 | * [Atomese](https://wiki.opencog.org/w/Atomese), a structural description system. |
| 211 | 211 | * [Sheaves](https://github.com/opencog/atomspace/tree/master/opencog/sheaf), |
| 212 | 212 | a general theory for assembling. weaving together structural |
| 213 | 213 | information. |