Artificial Intelligence CS105

Team Meeting Time (10 minutes) Find yourself a team Find your team leader Talk about topics and responsibilities

Recap Attempt to understand and mimic human intelligence Turing Test State of the art AI Looked at techniques for representing knowledge for intelligent problem solving – Semantic networks and trees

Expert systems Knowledge based system that uses specialized set of information to solve problems – Doctors, Plumbers, Gardeners, Dentist – Specific problem domain – Applications are specific – Example: the troubleshoot in your windows Rule-based system: Based on a set of if-then or logical rules – Rules form the knowledge base Inference engine: determines which rules to be followed, tries to derive answers from a knowledge base

Expert systems MYCIN: Has the patient recently had symptoms of persistent headache or other abnormal neurologic symptoms(dizziness, lethargy etc)? HUMAN: Yes MYCIN: The CSF cultures will be considered to be associated with meningitis. Please give the date on which clinical evidence(symptoms, signs or lab tests) of meningitis first appeared. Please enter CSF finding. Is meningitis a hospital-acquired infection? HUMAN: Why? MYCIN: This will aid in determining the organisms (other than those seen on cultures or smears) which might be causing the infection. It has already been established that 1. The infection which requires therapy is meningitis, and 2. organisms were not seen on the stain of the pending CSF culture, and 3. the type of infection is bacterial, and 4. neurosurgery is not one of the relevant items from the history of this patient, and 5. the age of this patient is greater than ten days, therefore, if 6. the infection was acquired while the patient was hospitalized then there is evidence that the organisms (other than those seen on cultures or smears) which might be causing the infection are e. coli (75), staphylococcus-coag pos (3), pseudomonas-aeruginosa (5).

Recognizing images Intelligent machines (Computer Vision)

Pattern recognition Techniques that classify data (patterns) in an attempt to understand the data and take actions based on that understanding – A priori knowledge: Previous knowledge that does not get modified with new experiences – Statistical information extracted from the patterns – Example: Face recognition system – understanding pixels A priori or statistical information based?

Classifying data Linear Classifier: A technique that uses an object’s feature(s) to classify which group it belongs to x1x1 x2x2 Some equation negative positive Positive or negative?

Can this be classified? x1x1 x2x2

Natural language processing Branch of AI concerned with interactions and human languages Natural Language: Set of languages that humans use to communicate This problem is of strong equivalence – Ability to comprehend languages, extensive knowledge about the outside world and being able to manipulate it – Voice recognition: recognizing human words – Natural language comprehension: interpreting human communication – Voice synthesis: recreating human speech

Voice synthesis Artificial production of human speech – A system used for this purpose is called a speech synthesizer How do you synthesize speech? – Phonemes: The set of fundamental sounds made in any given natural language /K/ in Kit and sKill Select appropriate phonemes to generate sound of a word, the pitch might be tweaked by the computer depending on context – Recorded speech Same words have to be recorded multiple times at different pitches

Voice recognition Sounds each person make is unique – Vocal tracts: cavity in animals where sound that is produced at the sound source is filtered Systems have to be trained for vocabulary sets – Acoustic modeling: Statistical models of sounds Audio recording of speech and text transcriptions – Language modeling: capture the properties of a language, and to predict the next word in a speech sequence

Natural language comprehension Most challenging aspect! – Natural language is ambiguous – multiple interpretations – Understanding requires real world knowledge and syntactic structure of sentences Examples: Time flies like an arrow The pen is in the box The box is in the pen George: My aunt is in the hospital. I went to see her today and, took her flowers. Computer: George, that’s terrible!

Natural language comprehension Lexical ambiguity : Words have multiple meaning Time flies like an arrow Syntactic ambiguity : Sentences have more than one meaning The pen is in the box The box is in the pen George: My aunt is in the hospital. I went to see her today and, took her flowers. Computer: George, that’s terrible! Referential ambiguity: Ambiguity created when pronouns could be applied to multiple object Ally hit Georgia and then she started bleeding Who started bleeding? Ally, Georgia or someone else? Makes sense Makes no sense!

Natural language comprehension Systems must have these common components: – Lexicon: vocabulary, word and expressions – Parser: Text analyzer, inbuilt grammar rules, to form an internal representation of the text – Semantic theory: study of meaning and relationships between words, phrases – Logical inference: Process of drawing conclusions based on rules applied depending on observations or statistical models

How do we process information? FRONTAL LOBE Behavior Problem Solving Planning Attention Abstract Thinking Judgment Inhibition TEMPORAL LOBE Audition Language CEREBELLUM Balance, Posture Cardio, Respiratory centers OCCIPITAL LOBE Vision PARIETAL LOBE Touch Sensory combination and comprehension Number area

Neuron Brain is made up of neurons – An electrically excitable cell that processes and transmits information by electrical and chemical signaling – An excited neuron conducts a strong signal and vice versa – A series of excited neurons form a strong pathway – A neuron receives multiple inputs from other neurons Assigns a weight on each signal based on its strength If enough signals are weak-> inhibited state or vice versa

Artificial neural networks A mathematical model inspired by the structure and/or functional aspects of biological neural networks Artificial Neuron or Node Inputs: 1 or 0 Receives many inputs Assigned a numeric weight If effective weight of each neuron is above a certain threshold, output is 1

Artificial neural networks Training: The process of adjusting the weights and threshold values – Series of comparisons to desired results w w

Artificial neural networks You can train a neural network to do anything – No inherent meaning to the weights: Making it versatile Applications: – Pattern recognition – Classification – Modeling how are brain works