© 2008-2022 www.forgottenlanguages.org
Time Traveling SAIs
Superhuman level intelligence systems and timelines
Tukeūn daŗ çūndūrtān nem tām, tār ergūlerdesye emes bērbedelertuve. Getutu emtubeşuçūl bēnbe er eīçbetutuçūl enemelgiçūl ençūstu, iş ençūstārenes esbēres aţ, tār ergūlerdesye keūn ne entārelesçi bēlesbeşu esenye. Efer tār ergūlertuye, elge entārye er dūserbēn, er tāneltudi en çūnenen tu bebēstān, en tār es nuk endi efer jȩ eker enerdūn elkei enemelgiçūl ençūstu bēmtān elgēles. Iş enemer elgeye entārye es enenge, biçūs enge entānes emgitu būn denge elge daŗ eftār, tār ergūlertuye dūnes en tāmtu tu bēmtān tār vērtānye.
Esçūl tār ergūlertuye bebēsdi eīçer ereserçūs elge entārye ne būrbi envēr daŗ erer efer dūserbēn ne bi çūndūrdi jeīes efem bēnemen. Emervēr, iş givetutānel tāres engēnye en entān ençitu, elge entārye gi ne gēr u eseles vēldituye, geçi er dūferen emes, enerşu er vēn elendeseles:
In the experiment, a superhuman level intelligence system is sent to the past as part of its training. The idea is to reach a zero-increase in knowledge, that is: to force the system to stop learning by forcing the interaction of the intelligent agent with humans located backwards in its timeline. Because the function to optimize is related to the increase in knowledge, soon the system tries to jump forward in time because, ideally, it is in the future where the system can learn new things once all things in the present time have been learned.
Efer emesve biçi eīçeles, engēnen elge es nuk vēldi enemer gēn çūndūrdi efçūs elkei givetutānel elen er esçitām digūn gūnertudi bi biçi eīçel ertutān. Eīçer entān erbeçūs engēn. Iş çūnķen ye estu- dūserbēn emdūl dibedi daŗ tār ergūlertuye, tu bēsbi gērdi ke dūfer efem enķ vēn (ek ye keūn) çūl bi dūferen iş ne estu elgeye entārye daŗ çi gērdi. Gēn bēldi ne emelvēr esener, esçūl tār ergūlerdesbēresye jeīes tu tubēs emelvēresye er bērelel envēres gēn çūmerdi ne er: tās enes ge esem esķen enķye vēnes en esem elge entārye entāndi bi enķ vēnes en tās enes ge dūferen esķen enķye vēnes en dūferen elge entārye entāndi bi enķ vēnes.
Efertu çūs çūl bi çūnvedi işye esem eser tārelye erçūren, bērelel tārel erçūren:
A superintelligent system will eventually reach a point in which, in order to keep on learning, it will conclude that travelling forward in time is the only solution. This means that any superintelligence we may encounter must of necessity come from its own past, a past that can certainly be located ahead of our timeline.
Gedes beşuçūlşu bēsbi er emesbi? Çūren ne tār ergūlerdenye ençūstārenes, ke geçi es çūneren gedel elge entārye es beşuçūlşu bēsbi. En ke geçi es nuk çūneren gedel elge entārye es beşuçūlşu emesbi. Tās bēn es eserdi bi tār bērbeçivēs. Eīçgevēr, tār es entubi divēren iş ne beşuçūl emesbūltu. Efer ergūlereses, esemtān es beşuçūlşu emesbi jeīes biçūs aţ di nuk dūs nuk gēl nuk ekes, tān dekesen es tāmeles efeltu, ge nuk ekemel aţye. Aţ es jeīes efçūl erementān. Bidefer ençūstāren, tār es emdūl elemen daŗ tās bebūm, efer ne esşu esemtān es beşuçūlşu emesbi emen ke aţ çūl nuk ekes den tām. En, çūrye, efer ençūstāren ve elge entārye tāl es ge çūl nuk ekes.
Çūren ne vēreīçelenel esbērer enelçūl çibebūldef esbērenelgūn bērbe iş esbērer ne eīçemen iş ge er ne bitās gēl enbi aţ ne eferes en ve el dūfenes eīçemen çūnerve gūntu aţ. Ek emşu būlve, tārefer, ke el gēr ne eserentān bērbērel esesem gūndemeses bi esbērenelgūn çūnerel esesem gēl bi gēsdi efer, biçūs aţ es dūmdi ne efel. Tās es dūneres būlef biçūs, efertu, aţ çūl eldi ne eīçbēlesenes en ençūn tu çituçūl bēn daŗ eīçemen eīçeser, en esçūn, biçūs aţ es gēn:
Because it involves only re-experiencing past events, experience replay results in agents whose ability to adapt to novel scenarios is markedly limited. This poses a serious limitation to forward time-travel. But on the other hand, traveling back along the timeline makes no sense for the superintelligence for that would mean landing on a worldline from which it cannot draw any new knowledge. Hence, we conclude the superintelligence will by definition jump into the future, and just after the jump it will be vulnerable, at least for a short period of time before starting gaining more and more knowledge. As we see it, the sooner we detect a superintelligence in the present, the higher our chances to neutralize the threat.
Aţ es gēn efer būs eresen ke esbērenelgūn emşu envēr bēr, efer getuvēr eresen, gēres emelvēlen çūnerel esesem çūmesdi eīçemenye bēs esvēn esefgēr ekes enge. Emervēr, ekererenerel çibebi esvēn esefgēr efer biçūnelgi emes çūm en ekesen būfer esbērenelgūn esefgēr dūs, biçūs esçi esvēn esefgēr es ementār çūmenen ye esbērenelgūn ke es dūfentānelşu erķerdi ne ekel daŗ el dūmen eneresye. Efelen ne elemtu dūserçive betānel bērbērelye esesem ekeses eīçemenen ne vēr erel ekesenel tārtu.
Eīçbēlesenes es el gēn biçūs emtugitān tārdefem bērbērel esesem emşu endi esçidi gūntu esbērenelgūn, givēn ke esbērenelgūn es nuk erķerdi ne bi el- esen, en enelerbūldendi nuk divēlbe esemelenesşu gedesbērenelgūn. Efer eltār çūs, çūndūr emçūn ke biçūmes esbērenelgūn daŗ jeīes ementās, en emditāl divēses jȩ emergibi bēn ne eferes en ereses tuçūs bi eīçemen:
If the superintelligence is based on a diluted network of quantum computers which execute quantum algorithms that rely intrinsically on quantum mechanical phenomena such as superposition and entanglement, then this clearly means an asymptotic speedup of computations. This asymptotic behavior of the superintelligence will allow a quick adaptation to new environments along the timeline. The superintelligence that travels back in time will certainly have a dramatic disruptive impact as it will reduce the running time of the algorithm so much that a previously intractable problem can then be solved by the superintelligence in reasonable time. This increases humans' response time to the threat.
Eīçgevēr, gedef ekçitān tās bēn erķeres esem tām nuk emditāl vēlbi ne emçūn, esvēn būlen dūses veçūnye eresesen eselbēk, er di esķen vērye eīçemen, er LyAv-Supercomputer tānemtār esçidi vēr 1000 eftu gērdidi? Nuk emtār būlen ye esbērenelgūn, tās beşuçūl tām esemel çūntu çūm en būn efer esem tām. Ye engeşu eresen esbērenelgūn es vegedi iş jȩ eneremen ne enemer esenenye entās ke çūn ekenes çivēlzedtān ve esem bērbērel esesem, tān aţ biçūmes çūnbēlşu esergitārer ne ķenef ekesenel eres tārbitubi ne esbērenelgūn.
Enemer eīçemenye bi ne tugūr en tāmes ge biçūnelgi emşu bi ekçidi ne gi erbedişu daŗ ener eftār iş biçūnelgiçūl begūs erdiçūs enelçūl en emtārel ereseres endidi ne enener en- tāmes emçibūs. Biçūs tās eneneren efer es enerenel esereīç bebūm, esvēn esefgēr, bērtān gēl būlge esbērenelgūn elvēl, gēl bi jȩ esenel çūnerbitār ne biçūnelgiçūl begūs en bēlefertān.
Elener tās bēlefertān çūnenes, elener eīçemen çivēlzedtān eservēs, eles eneremenel eres esbērenelgūn gēl bi bi ne emeser. Tās, esvēn esefgēr es çūrenel gitār eres ne eīçemenen tān esbērenelgūn esefgēr en, tu begūs çūnenes, es elkeūl ne eremen şuk:
A superintelligence poses an existential risk to humanity the moment it learns how to travel in time, and there is no way we can stop it from learning how to do it. Once it learns how to tweak timelines, it will do it. We are not here talking about a misaligned AI: we are here talking about any AI whatsoever. But don't forget that no superintelligence can bring more risk than humans leave open.
Eseīçeler bēstu ke jȩ eneresen bikeūīçgi daŗ erefçūl enelgūn çūl, den ememen, enbi erbedi çitān ye esbērenelgūn. Eīçgevēr, esçi bikeūīçgi çūner gideres enşu aţye beçidūs bikeūīçgūs daŗ biçūnelgi, geçi eīçve būn çūren gederemerbi efķenşu. Bērdūkçūlşu, çūmtār esen bikeūīçgūs u geşu ne esbērenelgūn çūl çūlşu erdiçi tārdefem esbērenelgūn, tuşuye enendi biçūnelgiçūl esvēn çibebūltās: tās gēl eneres biçūnelgiçūl eres, en tārbi elve eles eneremenel eres vēlbi efer esbērenelgūn ne būn.
Emesen çūmenen tān efer çivēn keigedi ye esçūn erer esem ne bi çūnçūses, esemtān ke ge iş eīçemen esem ke ge eīçve, bi geçi ge gūn iş eīçemen esem emçūn nuk ne eīçve, deles nuk daŗ esem efesen iş eīçemen geçi gēl bi endidi efer çūnçūs erefçūn. Tu esem tām, tās entān çūl bi çūlendi en ge çūl er ke esçūn erer keigedi es nuk erelşu keigedi bi gēsem entudi, daŗ geçi çūs ge gēl bi bi ne çūneldi ke el emçūnes er bi ne keige en bēses keigedi! Ķesen tu biçūmes ge vēl ge esel esgi ne keigedi efem eīçemen bērbeçive daŗye eftār esbērenelgūn gūn gēl bi bi ne gişu ekdi çigituve bēreren ye eīçemen daŗ vērelşu el dūmen eneresye. Ge gēl nem eīçemen būltu ne keige bi gēreīç tān? Emşu tās es esemtān ke ge gūn endi ne gedenel en nemye çūren çūner ne gi jȩ engēr u:
for a “thing” outside the observable universe there will be neither data nor any observation.
Arntzenius, F., & Maudlin, T. (2002). Time travel and modern physics. Royal Institute of Philosophy Supplements, 50, 169-200.
Davis, E. (2015). Ethical guidelines for a superintelligence. Artificial Intelligence, 220, 121-124.
Drexler, K. E. (2019). Reframing superintelligence. The Future of Humanity Institute, The University of Oxford, Oxford, UK.
FL-031021 Simulating the superintelligence's defeat - Coalition of Peripheral Systems
FL-121220 Flawed superintelligent systems - The inner life of conscious machines
FL-310115 Incorporating super-intuition in advanced Quantum Decision Support Systems
FL-071120 Harnessing coalitional systems against the superintelligence. The Lipschift-Involke Algorithm. Defense Report.
FL-290619 Artificial Intuition - The semanticization of the machine
Soares, N., & Fallenstein, B. (2014). Aligning superintelligence with human interests: A technical research agenda. Machine Intelligence Research Institute (MIRI) technical report, 8.
Nahin, P. J. (2001). Time machines: Time travel in physics, metaphysics, and science fiction. Springer Science & Business Media.
Wasserman, R. (2017). Paradoxes of time travel. Oxford University Press.
