1 BBS#Tel#Aviv#2016
IA:#Hard#vs#So7#Compu<ng#
Soft Computing
Subconscious Sub-simbolic
Connectionism (Hebb 40’s)
Computa<on# (and# mental# processes)# as#
the# emergent# process# of# interconnected#
networks#of#simple#units.
#
Neural#networks#(deep#learning) Swarm#intelligence
Hard Computing
Conscious Symbolic High level
Computa<on# as# manipula<on# of# symbols#
via#structured#rules#
RuleLBased#Systems#(Op<miz.#and#search) Logic#and#Constraint#Programming Ontologies
Probabilis<c#computa<on
…
2 BBS#Tel#Aviv#2016
Topics#and#Techniques
Knowledge#representa<on#
Logic Ontologies Sub0symbolic
Problem#solving
Search6(in6large6spaces) Op:miza:on
Adversarial6search6(games) Constraint6Sa:sfac:on6Problems
#Reasoning
Inference Planning PROLOG6and6CLP
Uncertain6reasoning
Probabilis:c6reasoning Markov6decision6processes Bayesan6networks
Neural6networks
Feed6forward6network Recurrent6network Deep6learning
Learning6and6data6analy:cs
Supervised Unsupervised Renforcement Induc:ve
3 BBS#Tel#Aviv#2016
Topics#and#Techniques
Communica<on
Natural6Language6Processing Human0machine6communica:on Automa:c6transla:on
Percep<on
Acous:c6percep:on Visual6percep:on
#Robo<cs
Sensors6and6aOuators Autonomous6decision6making
Expert6systems
Medicine Finance
Natural6programming
Gene:c6algorithms Ant6compu:ng
Cogni:ve6modeling
Cogni:ve6states Emo:ons
4 BBS#Tel#Aviv#2016
Logic#and#AI#
Many#different#logics
Proposi<onal,#firstLorder,#modal#(epistemic,#temporal,#deon<c)#…
Many#techniques#for#represen<ng#knowledge#and#reasoning Deduc<on,#Induc<on,#Abduc<on#…
Many#languages#and#tools
PROLOG,#Constraint#programming
As#the#base#of#many#Expert#Systems
5 BBS#Tel#Aviv#2016
PROLOG#(Kowalski#1973):#logic#programming
Example:#sum#of#two#integer#numbers
1. sum(0,X,X).
2. sum(X+1,Y,Z+1)
!sum(X,Y,Z).
Meaning:#
1.#the#sum#of#0#and#X#is#X
2.#if#the#sum#of#X#and#Y#is#Z#then#the#sum#of#X+1#and#Y#is#Z+1
Many#possible#queries
Fact Rule
sum(2,4,Z).############Answer#Z#=#6 sum(2,X,5).# #########Answer#X#=#3#
6 BBS#Tel#Aviv#2016
Deduc<on#
E#(deduce)
#
parent(X,Y)#:L#mother(X,Y).
parent(X,Y)#:L#father(X,Y).
##
mother(mary,vinni).
mother(mary,andre).
father(carrey,vinni).
father(carry,andre).
parent(mary,vinni).
parent(mary,andre).
parent(carrey,vinni).
parent(carrey,andre).
U
→B T
L
In#logic#programming
Deduc<on#allows#to#derive#consequences#of#the#assumed:
B#can#be#derived#from#A#if#B#is#a#logical#consequence#of#A##(A#|=#B) Given# the# truth# of# the# assump<on# A# follows# the# truth# of# the#
conclusion#B###
#
7 BBS#Tel#Aviv#2016
Induc<on#
parent(mary,vinni).
parent(mary,andre).
parent(carrey,vinni).
parent(carrey,andre).
mother(mary,vinni).
mother(mary,andre).
father(carrey,vinni).
father(carry,andre).
parent(X,Y)#:L#mother(X,Y).
parent(X,Y)#:L#father(X,Y).##
U
E+EL parent(mary,carrey).
→
B T#(induce)
#
∀e+#∈#E+:###B∪T#|=##e+####(T#è#completo)
∀eL#∈#EL:####B∪T#|=##eL####(T#è#consistente)
Induc<ve# reasoning# allows# inferring# B# from# A# where# B# # does# not# follow#
necessarily#from#A
The# premises# are# viewed# as# supplying# strong# evidence# for# the# truth# of# the#
conclusion
In# some# case# deriva<on# of# general# principles# from# observa<ons:# # we# have#
seen#white#swans#we#derive#the#rule#“all#swans#are#white”
In#LP
8 BBS#Tel#Aviv#2016
Abduc<on
Abduc<on#allows#to#infer##A#an#explana<on#of#B
In#other#words,#the#precondi<on#A#is#inferred#(abduced)#from#
the#consequence#B.
Given#a#theory#T#and#an#observa<on#O#we#infer#(abduce)#and#
explan<on#E#Such#that
T#U#E#|#=#O
T#U#E#is#consistent#
9 BBS#Tel#Aviv#2016
Abduc<on#in#LP
E
parent(X,Y)#:L#mother(X,Y).
parent(X,Y)#:L#father(X,Y).
##
mother(mary,vinni)#∨##father(mary,#vinni).
mother(mary,andre)#∨#father(mary,andre).
mother(carrey,vinni)#∨#father(carrey,vinni).
mother(carey,andre)#∨#father(carry,andre).
parent(mary,vinni).
parent(mary,andre).
parent(carrey,vinni).
parent(carrey,andre).
U →
B#(abduce)
T
T∪B#|=##E T∪B#|=##IC
Spesso#si#usano#anche#“vincoli#di#integrità”#per#controllare#la#
generazione#di#ipotesi.#
E
parent(X,Y)#:L#mother(X,Y).
parent(X,Y)#:L#father(X,Y). mother(mary,vinni).
mother(mary,andre).
father(carrey,vinni).
father(carey,andre).
parent(mary,vinni).
parent(mary,andre).
parent(carrey,vinni).
parent(carrey,andre).
U →
B#(abduce)
T
mother(X,Y)#→#female(X).
father(X,Y)###→#male(X).
female(mary).
male(carrey).
IC
T∪B#|=##E
10 BBS#Tel#Aviv#2016
Expert#systems#and#decision#support#systems#(1980#L)
Knowledge#based#systems#(expert#systems)
solve#problems#in#a#limited#domain#
have#performance#similar#to#an#expert#of#the#domain use#knowledge#representa<on#and#search#in#large#spaces build#dynamically#the#solu<on
search#and#solu<on#genera<on#guided#by#rules
.
11 BBS#Tel#Aviv#2016
Example#of#rule#based#system:#simple#diagnos<c#problem
OBIETTIVO: prescrivere una medicina adeguata per un determinato paziente in base ai
risultati di alcuni esami di laboratorio: prescribe(Drug).
BASE di CONOSCENZA Fatti:
gram(neg).
not(allergic(antb)).
Regole:
R1: gram (neg) → id (ecoli).
Se il risultato dell'esame è gram-negativo allora l'identità è enterium-coli R2: gram (pos) → id (strep).
Se il risultato dell'esame è gram-positivo allora l'identità è streptococco R3: id(strep) OR id(bact) → ind(pen).
Se l'identità è streptococco o bactero allora è bene indicare penicillina R4: ind(X) AND not (allergic(X)) → prescribe(X)
Se è bene indicare una certa medicina e il paziente non è allergico a tale medicina, allora si può prescrivere tale medicina al paziente
R5: id(ecoli) → ind(antb).
Se l'identità è enterium-coli allora è bene indicare antibiotici
12 BBS#Tel#Aviv#2016
Motore#di#Inferenza#o#Controllo#(backward)
Match R5 (X/antib)
Match R1
Act R1 Goal: prescribe(Drug)
prescribe(X) Match R4 (Drug/X)
ind(X) not allergic(X) Act R4 AND
Match R3 (X/pen) ind(pen) ind(antib) YES
Act R3 Act R3
id(strep)
id(bact) id(strep)
Match R2
gram(pos) Act R2
NO
NO
Act R5 id(ecoli)
id(ecoli)
gram(neg) YES
13 BBS#Tel#Aviv#2016
Example#of#working#session
Sys: È la coltura Gram positiva ? User: NO.
Sys: È la coltura Gram negativa?
User: SI.
Sys: Il paziente è allergico agli antibiotici ? User: NO.
Sys: Allora prescriverei una cura a base di antibiotici.
Sessione di spiegazione:
Sys: È la coltura Gram positiva ?
User: WHY? < perchè è importante determinare tale informazione?
Sys: Tale informazione è di aiuto per determinare se (R1) l'identità dell'organismo è streptococco.
User: WHY?
Sys: Tale informazione è di aiuto perchè per tale organismo è indicata la penicilina ( R3 ).
Sys: ... Allora prescriverei una cura a base di antibiotici.
User: HOW? <come puoi concludere tale responso?
Sys: Ho utilizzato la regola R4: ho prescritto antibiotici poichè sono indicati per tale infezione (prem1) e il paziente non è allergico (prem2).
User: HOW Premessa 1 ?
<come puoi concludere che sono indicati gli antibiotici?
Sys: Ho utilizzato la regola R5: sono indicati antibiotici poichè l'organismo che ha causato l'infezione è ecoli.
14 BBS#Tel#Aviv#2016
Some#recent#applica<ons###
(from#the#Expert#system#with#applica<ons#journal,#2017)
The# impact# of# microblogging# data# for# stock# market#
predic<on:# Using# Twiper# to# predict# returns,# vola<lity,#
trading#volume#and#survey#sen<ment#indices
Detec<on# of# idea# plagiarism# using# syntax–Seman<c# concept#
extrac<ons#with#gene<c#algorithm
A#compara<ve#study#on#base#classifiers#in#ensemble#methods#
for#credit#scoring
WordLsentence# coLranking# for# automa<c# extrac<ve# text#
summariza<on
….
15 BBS#Tel#Aviv#2016
Applica<ons#of#AI:#categories
“Normal”#human##ac<vi<es#(very#difficult#for#machines)
Natural#Language#Processing#(NLP)#
Vision#### #########################################################################beper#subLsymbolic Movement##and#robo<cs
…
Formal#ac<vi<es Games
Mathema<cs#and#Logic
….
Specialized#ac<vi<es
Expert#and#decision#support#systems#
Reccomender#systems Diagnosis
Planning
…
}
16 BBS#Tel#Aviv#2016
Some#examples
Chess
Limited#number#of#states Explicit,#non#ambiguous#rules
In#a#sense#“easy”#(but#huge#search#space)
Natural#language#processing#(Watson) Ambiguous,#context#depended#
Congi<ve#states
Robo<cs#and#autonomous#systems In#a#physical#ambient
Dynamic,#realL<me In#part#non#symbolic
17 BBS#Tel#Aviv#2016
Natural#language#understanding#and#QA
Watson#(IBM)#won#at#Jeopardy,#2011#!#
Jeopardy:#Given#an#answer#must#find#a#ques<on Language#understanding#+#Ques<on#Answering Uses#4#terabytes#of#data#(Encylopedia#etc.) Analyses#200#M#pages#of#content#in#3#second
18 BBS#Tel#Aviv#2016
Prac<cal#applica<ons#of#Watson
Financial# domain:# Bridgewater# Associates# (managing# 160# $# bn)# hired# the# chief#
developer#of#Watson#to#create#a#system#for#managing#daily#opera<ons.#Long#term#
goal#is#to#have#in#5#years#¾#of#the#managing#decision#done#by#so7ware
Health#domain:#Watson#Oncology#is#a#cogni<ve#compu<ng#system#deevloped#at#
Memorial#Sloan#Kepering#Cancer#Center##to#interpret#cancer#pa<ents’#clinical#
informa<on#and#iden<fy#individualized,#evidenceLbased#treatment#op<ons.
Personalized#tutoring##The#Teaching#Assistant#of#the#2016#Ar<ficial#Intelligence#
course##at#Georgia##Tech#was#a#program#(based#on#IBM#Watson).#It#was#answering#
students#ques<ons#onLline#with#a#success#rate#of#97%.#
Weather#forecast#Watson#used##to#analyze#data#from#over#200000#sta<ons
19 BBS#Tel#Aviv#2016
Element Number
of cores
Time to answer one Jeopardy! question
Single core 1 2 hours
Single IBM Power 750 server 32 <4 min
Single rack (10 servers) 320 <30 seconds IBM Watson (90 servers) 2 880 <3 seconds
Memory:
20 TB 200 million pages
(~1 000 000 books)
~1 000 000 million lines of code 5 years development
(20 men)
John# Searle:# “Watson# Doesn't# Know# It# Won# on# 'Jeopardy!’# IBM# invented# an#
ingenious#program—not#a#computer#that#can#think.”
Noam#Chomsky:#“Watson#understands#nothing.#It’s#a#bigger#steamroller.##Actually,#
I#work#in#AI,#and#a#lot#of#what#is#done#impresses#me,#but#not#these#devices#to#sell#
computers.”
Not#crea<on#of#a#new#algorithm#but#ability#to#quickly#execute#hundreds#of#proven#
language#analysis#algorithms#simultaneously#to#find#the#correct#answer.
[12][Both#hardware#and#so7ware#are#needed#for#efficiency#and#results:
20 BBS#Tel#Aviv#2016
10/01/17 14:16
Page 1 of 1 file:///Users/Mau1Nuovo/Downloads/DeepQA-1.svg
Question analysis
Query decomposition
Hypothesis generation
Soft filtering
Hypothesis and
evidence scoring Synthesis Final merging
and ranking
Hypothesis generation
Soft filtering
Hypothesis and evidence scoring Evidence
sources
Primary search Question
Answer and confidence Candidate
answer generation
Supporting evidence retrieval
Deep evidence
scoring Answer
sources
Trained models
“The#system#we#have#built#and#are#con<nuing#to#develop,#called#DeepQA,#is#a#
massively#parallel#probabilis<c#evidenceLbased#architecture.#For#the#Jeopardy#
Challenge,# we# use# more# than# 100# different# techniques# for# analyzing# natural#
language,#iden<fying#sources,#finding#and#genera<ng#hypotheses,#finding#and#
scoring#evidence,#and#merging#and#ranking#hypotheses”.#
[1]#[1]#Building6 Watson:6 An6 Overview6 of6 the6 DeepQA6 Project.# AI# Magazine# Fall,# 2010.# David#
Ferrucci,# Eric# Brown,# Jennifer# ChuLCarroll,# James# Fan,# David# Gondek,# Aditya# A.# Kalyanpur,#
Adam#Lally,#J.#William#Murdock,#Eric#Nyberg,#John#Prager,#Nico#Schlaefer,#and#Chris#Welty