alg writer 2 - 24 10 19 these are algorithms, not sentences - why and how to take care of them: asks for computational verb, i.e. count (checks name of predicates, synonyms with data type of spec) - multiple interpretations (verbs) all related algorithms: part of speech like adverb may give new algorithm ideas all main algorithms: asks for mainness algorithms (based on needed algs, stories indicating algorithms) intelligence: simulate machine learning by using similar algorithm verbs in related algorithms (with different data) unique don't use same specification/algorithm/result twice. later: intelligence (research), why working (business side) and word2vec (all details) how to convert a manually entered sentence into an algorithm - uses combos of computational verbs, finds similar algorithms from spec with same words as those that trigger (use) these verbs - rules for how different wording of spec (with expanded conditions, which can be simplified and asked for in relation to other specs) trigger different combos of computational verbs specs written for major calls to help speed development old predicates checked for usefulness (used or changed) or new ones written new ones written: user asked for hypothesis of commands thought to be needed in predicate, then decomposition, other parts said, etc. - fine detail of args used added as go - analysis algorithms try to modify existing alg to reach goal, offer clues The alg should save configuration as goes for later machines learning - number choice points in case goes back - x just like cawp - question answerer that converts simple list prolog plan to code, asks for input/data - spec at each call - fills in gaps, can redo parts