cog science ( history of AI)

Computers and AI

AI implementation depends on sufficiently developed technology the origins of AI are,thus closely tied to the computer

AI( artificial intelligence)

AL is the study and implication of systems able to perform functions typically considered to require human intelligence or cognitive abilities

Expert systems

After limitations of contemporary AI became clear works shifted towards expert systems knowledge bases systems implement knowldgege gained from area experts Tendril (mid-1960s): suggest chemical structure of unknown compounds MYCIN ( mid 1970s) : diagnostic medicine

Babbage engines

Difference engine precursor to the calculator calculates and prints mathematical tables analytical engine general purpose progmabale computational machine adopts punched cards from jacquard loom implements ideas of memory the store and processor (the mill) not built- idea sisnot come to fruition until 10- years later

General problem solvers

Early AI problem-slovers were designed as general problem solvers most well known:Newell & simon's general problem solver (GPS) involve searching among possible steps and using simple reasoning successful at solving certain types of pussles unable to handle complex problems (most ptoplems )

MYCIN

MYCIN represented as a set of knowledge of about 500 inference rules it interacts with user asking questions to other necessary data

Neural networks (history)

McCulloch & Pitts 1943: purpose threshold logic units and artificial neuron able to compute logical functions like and or not these able to compute anything a digital computer can bebb:hypothesis that learning occurs via synaptic strengthenng (hebbian learning)

early movers in AI

Newell &simon : Logic Theories,GPS Samuel ( and IBM): AI program not limited to predetermined instructions mccarthy: Lisp ( high level programing language); shake robotics project minsky: work in artificial neural works

Lingustics

Not all advance AI depended n computers in linguistics chomskey promotes the formalization of linguistics computational linguistics/ natural language processors much work on knowledge representation

Four approaches AI machines AI machines that...

Think like humans act like humans Think rationally act rationally

A machine

a machine that acts like a human could pass as human

acting rationally

acting to achieve the best outcome acting to achieve the best expected outcome,in casees of uncertainty

Thinking rationally

application of formal logic to solve any solvable problem ac

The age of Big date

big data has shifted how many approach AI for some problems more date ----more improvement than alternative algorithms thus less emphasis in understanding the problem and rule based approaches for cognitive science big data is not always appropriate the amount of data by machines for learning may be significantly more than that available to huamns

Jacquard loom first programmable machine

built in 1805 programs: punched cards punched cards indicate which threads to raise sequence of punched cards used to produce intricate designs

Machine translations today

currently statistical models are most widely used based on large amounts of data

Natural Language processing (eliza)

first chatterbot 1964-1966 mimicked Rogeria psychotherapist ( most well known scripts,anyway) matches input to stored patterns to determine response upon initial encounter some people believed it to be real

state of the art

game playing (unpredictable enemies) robotics (roomba) machine translation waston(jeopardy! contestant ) synapse chip

The age of DATa

in recent years very large datasets have become available much possible due to internet ( language corpora ) companies collecting data from many users sciencntific date more easy to share( billons of base pairs of genomic sequences)

Machine translation 1960's

machine translation nd nature language processing were also areas of inert early on Naive belief that translation would be a simple task given a bilingual dictionary,word for word subsections would be east problems turns out to be much more complex multiple meaning for single spelling nuances concepts without direct word translations grammar differnces

tO THINK LIKE A HUMAN requires modeling the human mind ..

models of the human mind recognize both good and bad linguistic constructions reason in human fashion rather than deductively mimic learning patterns of humans this approach would align closely with the goals of cognitive science

required capabilities ignorer to pass as a human

natural language processing knowledge representation automated reasoning machine learning computer vision robotics

Competing computational approaches to AI

physical symbols system( newell & simon) AI consist of symbols and rules for their manipulation connectionism no symbols or explicit rules networks or simple units may be viewed as learning behaviors rather than rules

Natural language processing (shrdlu)

program with more advanced language abilities dictionary with semantic information knowledge about syntax: ability to erase sentences discourse knowledge (storage of command history ) reason ability navigate through a world of 3D objects carry out task correctly limits the world to a small domain difent iced blocks on a table ' a robot with a hand eye and ability to move blocks

State of the Art

robotic vehicles (DARPA challenges,autonomous vehicles by google,tesla ) natural language processing speech recognition( cell phone assistants ) auto summarization sentiment analysis journalism ( with predictable types of data; earning reports )

Neural netoworks (hisroty )

rosenblatt 1957: bills upon McCulloch & pitts with the preception minsky & papret (1960) perceptroois: shows limits of single layer preceptors (internet in neural networks declines for the next decade) 1986: rediscovery of back propagation learning algorithm; becomes popularized again

the goal of AI

strong AI- AI that is indistinguishable from a human;possibly attaining conscious weak AI- AI designed for some specific reasoning or problems solving task; no consciousness

computers

the earlist computers were built during wwII z3 (de; analyses of wing flutter ) heath Robinson,colossus (UK; codebreaking ABC,ENIAC (US;artillery tables,firing tables) e.g calculations to aim artillery at moving targets at a distance to to drop bombs from moving airplanes in different atmospheric conditions 1951: marvin minks builds SNARC, a computer modeled after neural networks

Ada lovelace: worlds first programmer

worked together with baggage made notes on babbages works realized that the analytical engine could be used for functions beyond mathematics wrote worlds first program ( an algorithm for analytical engine )

Birth of AI

workshop in AI 1956 at yarmouth college