Was wird uns KIalles bringen?

Prof. Dr.-Ing. Christian Bauckhage

we are living in exciting times . . .

big data+ affordable HPC+ deep learning systems= progress in AI

the media on AI

Feb 2016 Feb 2016 Mar 2016

May 2016 May 2016 May 2016

the history of AI

< 1940 philosophy, mathematics,engineering

since 1940 first modern computersColossus, Turing et al.Z3, Plankalkul, ZuseABC, Atanasoff et al.ENIAC, Mauchly et al.

1943 control theory, Rosenblueth,Bigelow, Wiener & von Neumannartificial neurons, McCulloch &Pitts

1944 game theory, von Neumann

1945 Bush on as we may think

1946 Monte Carlo methods, Ulam

1948 cybernetics, Wienerinformation theory, Shannon

1949 Hebbian learning, Hebb

1950 neural network computer,Minsky & EdmondsTuring test, genetic algorithms,and machine learning, Turing

1951 Shannon on computer chess

1955 computational linguistics,Chomsky

1956 alpha-beta search, McCarthybirth of AI, Dartmouth summerschool by McCarthy, Minsky,Shannon & RochesterLogic Theorist, Newell & Simon

1957 perceptron, Rosenblatt

1958 LISP, McCarthy

1959 Geometric Theorem Prover,Gelernter

the history of AI

1960s microworlds, Minsky, Papert,Sussman, Winograd, Abelson

1964 ELIZA, Weizenbaum

1965 fuzzy logic, Zadeh

1966 HMMs, Baum et al.

1967 k-means clustering, MacQueen

1968 semantic networks, Quillian

1969 Minsky & Papert on perceptrons

since 1969 knowledge-based systemsDENDRAL, Buchanan et al.heuristics, Feigenbaum et al.SHRDLU, WinogradSHAKEY, Rosen, Nilsson, . . .⇒ A∗ search, Hough trsf, . . .

1972 PROLOG, Colmerauer

1975 partial order planning, Sacerdotiframes, Minsky

1978 version spaces, Mitchell

1979 Stanford cart, MoravecBKG by Berliner (CMU) beatsworld backgammon champion

the history of AI

1982 Hopfield networks, Hopfieldself organizing maps, Kohonen

1983 SOAR, Laird & Rosenbloom

1984 Bayesian networks, Pearldecision trees, Breiman

1985 Boltzmann machines, Hinton &Sejnowski

1986 back-propagation, Rumelhart,Hinton & Williams

1987 behavior-based robotics, Brooks

1989 ALVINN, PomerleauZIP code reader, LeCun et al.

1991 ILP, Muggleton

1992 TD-Gammon by Tesauro (IBM)uses reinforcement learning

1993 support vector machines, VapnikMonte Carlo GO, Brugmann

1994 self driving cars (VaMP, VITA-2),DickmannsChinook by Schaeffer et al. (UofA)beats world checkers champion

1995 kernel methods, Vapnik & Cortes

since 1995 advanced agent technologiese.g. RAX (NASA)

1997 Deep Blue (IBM) beats worldchess championLogistello by Buro (NEC) beatsworld othello champion

1998 semantic web, Berners-Lee

the history of AI

since 2000 computer vision maturesspeech recognition maturesinformation retrieval maturesrecommender systems mature

2003 train game bots using playerbehavior data, Bauckhage et al.

2005 Stanley by Thrun (Stanford) winsDARPA grand challenge

since 2010 breakthroughs in computer vision,speech recognition, automatictranslation, text processing, . . .

2011 Watson (IBM) wins Jeopardy

2013 neural networks play ATARIgames (Google DeepMind)

2015 open letter by Hawking, Musk,Wozniak on ethics in AI

2016 AlphaGO (Google DeepMind)beats world GO championALPHA (U Cincinnati) beatsair-force fighter pilots

2017 Libratus and DeepStack (CMUand UofA ) beat poker pros

faces of defeat

Kasparov, 1997 Jennings & Rutter, 2011 Lee, 2016

neural networks in a nutshell

mathematical neuron

1 x1 x2 . . . xm

f(z)

y

w0 w1 w2

wm

−4 −2 0 2 4

−1

0

1

w0

z =m∑

i=1

wi xi − w0

f (z) = tanh(z)

“deep” neural network

1 2 3 4 5

6 7 8 9

10 11 12 13

14 15 16

17 18 19

20 21

dramatic progress in cognitive computing

text analysis and understanding

image understanding

speech recognition

robotics...

Socher et al., Proc. EMNLP, 2013

paradigm shifts in many areas

medical diagnostics

Ciresan, Giusti, Gambardella, and Schmidhuber, Proc. MICCAI, 2013Gulshan et al., JAMA, 316(22), 2016

Esteva et al., Nature, 542(7936), 2017

paradigm shifts in many areas

drug design, genetics

Ma et al., J. on Chemical Information and Modeling, 55(2), 2015Xiong et al., Science, 347(6218), 2015

a frenzy in finance

considerable investments by

BlackRock, Bridgewater,Schroders, MAN AHL, . . .

AI-based hedge funds / fintechs

Aidyia, Numerai, Sentient, . . .

research on predicting

stock momentum, volatility offutures, insolvency risk, . . .

reports accuracies of 53% – 60%

Krauss et al., Europ. J. Operation Research, 2016Ding et al., Proc. IJCAI, 2015

generating program code (A. Karpathy, github, 2015)

/* Increment the size file of the new incorrect UI_FILTER group information */static int indicate_policy(void){

int error;if (fd == MARN_EPT) {

/** The kernel blank will coeld it to userspace.

*/if (ss->segment < mem_total)

unblock_graph_and_set_blocked();else

ret = 1;goto bail;

}segaddr = in_SB(in.addr);selector = seg / 16;setup_works = true;for (i = 0; i < blocks; i++) {

seq = buf[i++];bpf = bd->bd.next + i * search;}

}bprm_self_clearl(&iv->version);regs->new = blocks[(BPF_STATS << info->historidac)] | PFMR_CLOBATHINC_SECONDS << 12;return segtable;

}

generating art

things are getting crazier by the day . . .

Google’s tensor flow, Dec 2015 Microsoft’s CNTK, Jan 2016

a Japanese farmer builds a cucumber sorting system

Google Cloud Blog, 2016

it does not even have to be neural networks . . .

June 2016

and the future is already here!

several “crazy” statements

by 2027, 75% of all S&P 500 companies will have disappeared

McKinsey, 2016

60% of all professions will be affected by automation

McKinsey, 2016

in 10 years, every process will be managed by an AI

C. Bauckhage, 2017

2016 2018 2020 2022 2024 2026 2028 2030

year

109

1010

1011

1012

1013

1014

1015

#sy

naps

es

bee

mouse

cat

human

The greatest shortcoming of the human raceis our inability to understand the exponentialfunction.

Albert A. Bartlett

thank you!

to find me online, simply google

“Christian Bauckhage”