Signatures of Post-Biological Intelligence: extending the search spectrum Dr John R Elliott Reader in Intelligence Engineering International Astronautics Association SETI Committee; Post-Detection Task Group Leeds Metropolitan University, UK.
The Signal Universe Long chain Polymers Structured All Highly predictable Phenomena e.g Pulsars Language . ..? Music D.N.A Random e.g White Noise
Would a machine construct [evolve] a communication system based on logic or an optimised form of natural language encoding [no redundancy - 100% pattern utilisation]?
We need not just target [listen to] areas of the Universe which support biological life.
We look at the likely signatures and contrasting structures such non- biological communicators may present and discuss how such contrasting forms of information exchange can aid, extend and refine our detection and decipherment capabilities. To achieve this, we first look beneath the Veneer at known biological communicators and the common structures they demonstrate. ...do machines, or do post-biological agents expand the signal universe?
The Human Language Template: some core features Surface Structure : all language [communication] demonstrates similar structural surface entropic structure, as does information rich phenomena ; Internal Structure : more specific to language – as increased dependency is calculated, the entropic value diminishes at an approximate -1 slope [log- log graph] ; Cohesion : how sub-types of an information [communication] system bonds with other sub-types, to build narrative; Cognition : the constraints of a system that reflects the processing limits of the users. This results in chunking of blocks of information, which often share an imbedded relationship, within the narrative [discourse]: e.g. Phrases.
The Efficient Compromise ....linguistic reciprocal altruism Rhythm and structure – you may not understand the utterances but you can usually recognise what language is being spoken; Evolution of language – evidence and many simulation show that the efficient relationship of frequency and word length are a constant.
Human Language Evolution - a simulation Over time the simulation follows similar paths for successive trials of grammar rule induction. This begins with a large set of arbitrary rules when the size of the E- language is small, which expand to an even larger set as the E-language expands, before eventually ‘settling down’ to a concise and relatively small set of rules when the E-language is extensive. From such simulations, basic structural properties of language emerge over time through the complex dynamical process of social transmission.
Human Language Evolution - a simulation Grunts, babble, emergent structure.........refinement to efficiency, where principles of least effort and reciprocal altruism are embodied. Yep...me in the morning Humans do it.....and so do Dolphins These simulations also demonstrate that frequently used words, such as function words, eventually evolve to be the shortest. Ultimately, we evolve: • a very small number of very common words • a small-medium number of middle frequency words • a very large number of words that are infrequent
Dolphins It is of interest to note that while some dolphins are reported to have learned English – up to 50 words used in correct context – no human being has been reported to have leaned dolphinese. Carl Sagan
Dolphin ‘Chatter’ 85% 2 20% 60% 20% 20% 108 3 137 80% 67% 20% 162 5 17% 13% 57% 13% 14% 7 27% • Number of distinctive sounds ≈ 163
A comparison of structure = Dolphin
(Robot) Silicon chat Lingodroids are robots, which use an onboard camera, sonar and a laser range- finder, to map the space around them . The language, which sounds similar to the tones on a phone, is 'spoken' aloud by using a microphone and speaker. Experiments conducted in this project are a useful insight into how machine intelligence may develop communication. These words have been invented by the robots themselves , using a variety of games to establish correlations between specific words and places, directions, and distances. And recently Lingodroids have been teaching themselves brand new words for different lengths of time.
Some examples of the Lingdroid vocabulary Geographical location examples: yifi, kiyi, gige, mira, xala, soqe, sihu, juhe, rije, pize, tuto, kopo, heto, qoze, yaro, zuce, xapo, zuya, fili, fexo, pucu, reya. Distance measurement examples: puga, puru, vupe. duka, ropi, puga, huzu, hiza, kobu, bula, Temporal examples: kafi, puni, fohu, qija, fedi, tofe Unlike human discourse, robot [Lingodroid] vocalisation requires no inbuilt redundancy - like dialing phone numbers. So, optimisation of ngram usage - for a given word length - is both possible and desirable, for purposes of efficiency.
Robot Language Change, Learning and Sharing Language Change ↔ Learning Period When different communities of agents meet, their different languages are acquired quickly: high frequency words are rapidly well established. Comprehension well established, even if terms not used by other community.
Computer code, Robot & Human Entropic Profiles 7 6 5 4 Computer code Lingodroid Human Language 3 2 1 0 1 2 3 4 5 6
140 120 100 80 60 40 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 LinCos vs English Semantic chunks / phrases
Bigrams = 676 Trigrams = 17, 576 Tetragrams = 456,976 Pentagrams = 11, 881, 376 The above number of combinations is based on 26 letter alphabet, which is, of course, an assumption. We use words that have a high relative frequency of use peaking at the range 8 to 10 letters. If we actually were able to utilise all combinations, this would provide us with a possible lexicon of 141,167,095,653,376 words. Or rounding down to 140 Trillion (USA) or 140 Billion in old UK currency!
Conclusions – for now Entropic Signatures – similar initial profiles but differences do occur as higher entropic values are calculated; Redundancy and Ngram frequencies contrast significantly; No evidence (yet) for morphology in lingodroids; The distribution of functional elements to logically relate content elements, in LinCos, are not constrained by cognition, as it is not a spoken conduit; Typical distributions found in natural language are currently absent in evidence available.
Conclusions – for now The signature of machine communication will be seen in the relationships and structure of the comprising patterns that represent semantic content, as it is clear that initial entropic profiling will be similar for both machine and biological intelligent communicators. So, whether Morse code, Ogham style encoding, Tonal encoding......or whatever conduit is used to encode and communicate information, it is the relationship of the patterns that is important, when ascertaining the type of author (man or machine): their constraints, behavioural categories, frequency - relative percentage – of combinations utilised and redundancy.
Recommend
More recommend
