Types of AI you should know.pdf

https://pythongeeks.org/types-of-ai/

1. TypesofAIyoushouldknow
ArtificialIntelligenceisoneofhumanity’smostsophisticatedandamazingcreations
todate.Thatignoresthefactthatthefieldisstilllargelyunexplored,implyingthat
anyamazinginnovationweencountertodayisonlythetipoftheAIiceberg.Despite
thefactthatthispointhasbeenreiteratednumeroustimes,acomplete
comprehensionofAI’sfutureimpactremainselusive.
PeopleareconcernedabouttheinevitabilityandvicinityofanAItakeoverdueto
AI’srapidgrowthandformidableabilities.Moreover,theeffectsofAIinnumerous
industrieshavecausedbusinessexecutivesandthepublicatlargetoassumethatwe
arenearingthepeakofAIresearchandfulfillingAI’sfullpotential.Understanding
theperspectiveandcurrenttypesofAIwillprovideafullerpictureofcurrentAI
technologiesandtheroadtotheirdevelopment.
WhatisAI?
Theprocessofcreatingintelligentmachinesfrommassiveamountsofdataisknown
asartificialintelligence.Systemslearnfromtheirpastexperiencesandaccomplish
tasksthataresimilartothoseperformedbyhumans.Itimprovestheefficiency,
precision,andefficacyofhumanefforts.Tocreatecomputersthatcanmake
judgmentsontheirown,AIemployscomplicatedalgorithmsandprocedures.
TypesOfAI
TheextenttowhichanAIsystemcanimitatehumanabilityisusedasameasureto
determinethetypesofAI.Asaresult,AIcanbedividedintonumerouscategories
basedonhowwellamachinecorrespondstohumansintermsofdiversityand
efficiency.Inthisapproach,anAIthatcanperformmorehuman-liketaskswith

2. equivalentstandardsofaccountabilitywillbeconsideredamoresophisticatedsort
ofAI,whereasanAIwithrestrictedfunctionalityandperformancewillberegarded
asasimplerandlessdevelopedtype.
TherearetwoclassificationsforAIbasedonthiscriterion.Oneapproachisto
categorizeAIandAI-enabledtechnologiesaccordingtotheirresemblancetothe
humanmindandtheirabilityto“think”andpossibly“feel”likehumans.
Type1-BasedOnCapabilities
Basedoncapabilities,therearethreecategoriesofAI:
■NarrowAI
■GeneralAI
■SuperAI
1.NarrowAIorWeakAIorArtificialNarrow
Intelligence(ANI)
“Alexa!Setanalarmfor7A.M.”
AlsoknownasWeakAI,isalevelofAIthatinvolvesrobotsthatcanonlydoalimited
setofactivities.Atthisphase,themachinehasnoabilitytoreasonandjustconducts
aseriesofpre-definedoperations.
Cortona,Siri,Alexa,self-drivingcars,Alpha-Go,Sophiathehumanoid,andothers
areexamplesofweakAI.
2.GeneralAIorStrongAIorArtificialGeneral
Intelligence(ANI)

3. AGI,alsoknownasStrongAI,isthestepinthedevelopmentofArtificialIntelligence
whenrobotswillbeabletoreasonandmakedecisionsinthesamewaythathumans
do.Itisyettobedemonstratedbutisexpectedtodevelopintelligencethesameas
humans.Manyscientists,includingStephenHawking,believethatstrongAIposesa
threattohumanity’ssurvival.
“ThefulldevelopmentofAIcouldspelltheendofmankind.It’dsetoffonitsown,
re-designingitselfatabreakneckspeed.Humans,whosebiologicalevolutionis
slowed,wouldbeunabletocompeteandwouldbesurpassed.”
3.SuperAIorArtificialSuperIntelligence(ASI)
SuperAIisthestageofArtificialIntelligenceatwhichcomputers’capabilities
surpassthoseofhumans.MachineshavetakencontroloftheEarth,accordingtoa
hypotheticalscenariopresentedinsci-finovelsandmovies.
Givenourpresentrateofdevelopment,Ibelievemachinesarenotfarfromreaching
thisstage.
“Youhavenonotionhowfast—itisexpandingataratethatisnearto
exponential—unlessyouhavedirectexposuretogroupslikeDeepmind.Inthenext
fiveyears,thereisasignificantriskofsomethingextremelydangeroushappening.At
mosttenyears.!—AccordingtoElonMusk.
Thesearethevariouslevelsofintelligencethatamachinecanachieve.Let’slookat
themanytypesofAIandhowtheywork.
Type2-BasedOnFunctionalities
1.ReactiveMachine

4. TheyarethemostbasicandancientsortofArtificialIntelligence.Theyimitatea
human’sabilitytorespondtoavarietyofstimuli.BecausethistypeofAIhasno
memory,itisunabletousepreviouslyacquiredinformation/experiencetoimprove
results.Asaresult,theseAIsystemslacktheabilitytolearnthemselvesliketheones
weseetoday.
DeepBlue,thecomputerthatdefeatedinternationalgrandmasterGarryKasparov,is
anexcellentexampleofthistypeofequipment.
Thesupercomputerwasabletodetectallofthelegaloptionsavailabletoitandits
opponents.Itchosethebestfeasiblemovebasedontheoptions.However,because
thesemachineshavenomemoryoftheirown,theyareunabletolearnfromtheir
previousactions.
2.LimitedTheory
ThissortofAI,likeReactiveMachines,hasmemorycapabilities,allowingitto
leveragepriordataandexperiencetomakebetterdecisionsinthefuture.This
categoryencompassesthemajorityofthecommonlyusedapplicationsinourdaily
lives.TheseAIapplicationscanbetaughtusingahugeamountoftrainingdata
storedinareferencemodelintheirmemory.
Manyself-drivingcarsusethemtostoredatasuchasGPSlocation,speedof
neighboringautomobiles,size/natureofbarriers,andahundredothertypesof
informationinordertodrivelikeaperson.
Therearethreetypesofmachinelearningmodelsthatcanachievethisformof
LimitedMemory:
a.Reinforcementlearning

5. Throughseveralroundsoftrialanderror,thesemodelsevolvetomakebetter
predictions.ComputersaretaughttoplaygameslikeChess,Go,andDOTA2using
thistechnique.
b.LongShortTermMemory(LSTMs)
Researchersreasonedthatusingpastdatatopredictthenextiteminasequence,
particularlyinlanguage,wouldbebeneficial,thereforetheydevisedamodelbased
ontheLongShortTermMemory.TheLSTMlabelsmorecurrentinformationas
moresignificantandthosefromthepastaslessessentialwhenpredictingthe
followingpartsinasequence.
c.EvolutionaryGenerativeAdversarialNetworks(E-GAN)
BecausetheE-GANhasmemory,itevolveswitheachevolution.Themodel
generatesadevelopingentity.Becausestatisticsisamathofchance,notamathof
exactitude,growingentitiesdonotalwayspursuethesameroute.Themodelmay
identifyabetterpath,apathofleastresistance,asaresultofthechanges.The
model’sfollowinggenerationmutatesandevolvesinthedirectionofitsancestor’s
incorrectroute.
TheE-GANproducesasimulationthatisanalogoustohowpeoplehavedeveloped
onthisplanetinseveralways.Eachchildismorepoisedtohaveanextraordinarylife
thanitsparentintheeventofflawless,successfulreplication.
3.LimitedMemoryTypesInPractice
Whileeverymachinelearningmodelisbuiltwithafiniteamountofmemory,this
isn’tnecessarilythecasewhenit’sdeployed.
A.I.withlimitedmemoryworksintwoways:

6. Ateamisconstantlyupdatingamodelwithnewdata.
ModelsareautomaticallytrainedandrefreshedintheA.I.environmentbasedon
modelusageandbehavior.
Machinelearningmustbebuilt-inintothestructureofamachinelearning
infrastructureinorderforittosupportalimitedmemorytype.
ActiveLearningisbecomingmorewidespreadintheMLlifecycle.Therearesixsteps
intheMLActiveLearningCycle:
■TrainingData.Amachinelearningmodelrequiresdatatotrainon.
■BuildMLModel.Themodelhasbeendeveloped.
■ModelPredictions.Themodelmakespredictions,
■Feedback.Humanorenvironmentalinputsprovidefeedbackonthe
model’spredictions.
■Feedbackisconvertedintodata.Thedatarepositoryreceivesthefeedback
andstoresit.
■RepeatStep1.Continuetoiterateonthiscycle.
4.TheoryofMind
Itisthenextlevelofartificialintelligence,withlittletonoimpactonourdaily
existence.ThesetypesofAIareofteninthe“WorkinProgress”stageandareonly
availableinresearchlabs.Onceachieved,thistypeofAIwillhaveacomprehensive
understandingofhumanminds,includingtheirneeds,likes,emotions,mental
processes,andsoon.TheAIwillbeabletochangeitsownresponsebasedonits
graspofhumanmindsandtheirwhims.
ThetheoryofmindAIwasimplementedatHansonRobotics’Sophia.Sophiaisable
toseethankstocamerasinhereyesandcomputeralgorithms.Shecankeepeye
contactwithindividuals,recognizethem,andfollowtheirfaces.

7. 5.Self-AwareAI
ThisistheAI’sfinalstep.Itscurrentpresenceissimplyarumor,anditcanonlybe
foundinsciencefictionfilms.TheseAIsystemsarecapableofcomprehendingand
elicitinghumanfeelings,aswellaspossessingtheirownemotionalstates.Thisform
ofAIwilltakedecades,ifnotgenerations,todevelop.ElonMuskandotherAI
doubtersarewaryofthistypeofAI.ThisisbecauseonceanAIbecomesself-aware,
itmayenterSelf-Preservationmode,viewingmankindasapossiblethreatand
pursuingeffortstoeliminatehumanitydirectlyorindirectly.
BranchesofAI
Byemployingthefollowingprocesses/techniques,ArtificialIntelligencecanbe
utilizedtotacklereal-worldproblems:
■MachineLearning
■DeepLearning
■NaturalLanguageProcessing
■Robotics
■ExpertSystems
■FuzzyLogic
1.MachineLearning
Thescienceofteachingmachinestounderstand,process,andanalyzedatainorder
tosolvereal-worldissuesisknownasmachinelearning.
MachineLearningisdividedintothreecategories:
■SupervisedLearning
■UnsupervisedLearning
■ReinforcementLearning
2.DeepLearning

8. ItistheprocessofusingNeuralNetworkstoobtaininsightsandbuildsolutionsfrom
high-dimensionaldata.DeepLearningisasubsetofMachineLearningthatcanbe
usedformorecomplexissues.
3.NaturalLanguageProcessing
GenuineLanguageProcessing(NLP)isthestudyofextractinginformationfroma
naturalhumanspeechinordertocommunicatewithrobotsandexpandenterprises.
Amazonemploysnaturallanguageprocessing(NLP)tobettercomprehendcustomer
feedbackandimprovetheuserexperience.
4.Robotics
ItisabranchofAIthatfocusesonvariousrobotapplicationsanddisciplines.AI
Robotsareartificialagentsthatactinareal-worldenvironmenttocreateresultsby
takingresponsiblebehaviors.
Sophiathehumanoidisanoutstandingdemonstrationofartificialintelligencein
robotics.
5.ExpertSystems
Anexpertsystemisacomputersystembasedonartificialintelligencethatlearnsand
mimicsthedecision-makingabilitiesofahumanexpert.
If-thenlogicalnotationsareusedbyexpertsystemstotacklecomplicatedissues.It
doesnotrelyonproceduralprogramminginthetraditionalsense.Expertsystems
aremostlyutilizedindataadministration,medicalfacilities,loananalysis,andvirus
identification,amongotherapplications.
6.FuzzyLogic

9. Insteadoftheconventionalmoderncomputerlogic,whichisbooleaninnature,
fuzzylogicisacomputingapproachbasedontheideasof“degreesoftruth.”
It’sutilizedtoaddressdifficultchallengesthatrequiredecision-makinginthe
medicalindustry.They’realsoemployedinautomatictransmissions,vehicleclimate
control,andotherapplications.
Conclusion
Wemaybealongwayfromconstructingself-awaremachinesthatcanfixall
problems.However,weshouldconcentrateoureffortsonfiguringouthowa
computercantrainandlearnonitsownandmakedecisionsbasedonprevious
experiences.
IhopethisposthasclarifiedthemultiplekindsofAI.