chapter thirteen the artificial intelligence ii: embodiments and robotics

22
CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Upload: joanna-moore

Post on 02-Jan-2016

224 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

CHAPTER THIRTEEN

The Artificial Intelligence II: Embodiments and Robotics

Page 2: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

The IA ParadigmIntelligent Systems (and agents) are built

on three basic ideas.

modeling aspects of biological systems (including) human reasoning

abstracting general principles of intelligent behavior application of concepts to the design of “intelligent” systems.

Page 3: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Modeling Aspects of Biological Systems Biology is the study of living organisms

and

provides direct inspiration to automatically find solutions to optimization problems.

We can: improve mechanical design of the products that we use; discover innovative computer software; design electronic circuits.

Evolutionary computation models follow directly from natural evolution (diversity, survival, inheritance, and a selection process).

Page 4: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Abstracting General Principles of IAs

Embodiment : in regard to IAs this refers to the interaction with the environment through a physical body (robot).Situated. existence of an agent in a dynamic, rapidly changing environment which the entity can manipulate.

(The distinction: being situated emphasizes the view that

intelligent behavior follows from the environment and the agent’s interactions with it :

the embodiment is defined by the interactions with the environment itself. )

Page 5: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Applying the principles to the

design of IAsIAs include components that are considered critical in mainstream psychology. They are also well thought-out elements of human intelligence and summarized in the table on the next slide – Behavioral Elements of Human Intelligence.

Page 6: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Element of Intelligence Notes

Trial-and-error learning

Responses to external error that lead to satisfactory results will be repeated when the same stimuli are encountered: chess moves

Rote learning Direct association between stimulus and response: memorizing a list of items.

Operant conditioning

Highly developed form of learning involving positive and negative reinforcements of behaviors: gradually slowing a car at a red light. See Chapter 3 (Psychological Approach).

Reasoning

The ability to draw inferences appropriate to the facts and the situation. It is embodied in logical thinking and includes deductive, inductive, and abductive reasoning: Expert Systems – given the symptoms of computer failure, determine what needs to be repaired (or replaced).

Problem solving

Special-purpose solution relies on the circumstances that

pertain to the task; General-purpose solutions: can be applied to a broad

variety of problems: Means-ends analysis: the machine assesses the current

state of the system and chooses an action that will reduce the difference between the current state and the goal state.

Perception

The “intelligence” scans the environment using the sensory equipment it has at its disposal; the information is processed by internal processing mechanisms and converted to a set of abstractions that is made up of some combination of objects, features, and relationships.

Language Communication between human and machine.

Page 7: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

The Cognitive IA Model After Russell/Norvig

Learning and knowledge

CRITICCRITIC

LEARNINGLEARNING

RECOMMENDATIONSRECOMMENDATIONS ACTUATORSACTUATORS

COGNITIVE PERFORMANCE ELEMENT

COGNITIVE PERFORMANCE ELEMENT

SENSORSSENSORS

Environment

Learning goals

Feedback

percepts

actions

Performance standard

Page 8: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

How would the “GPS” develop a travel plan?

Using the street map below, develop a plan to go from the library to the university.

(Arrows indicate one-way street directions.)

Page 9: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

An IA includes Facts, Rules, Algorithms.

The Rules consist of IF-THEN assertions; If the antecedent (IF) is satisfied, the consequent (THEN) is invoked. For the problem on the previous slide, the IF-THEN rules follow:

If (at “library”) Then (follow one-way street to “Hospital”)If (at “library”) Then (continue in same direction to next intersection)If (at “intersection”) Then (turn right and continue towards “school”)If (at “school”) then (continue in same direction to “factory”)If (at “hospital”) Then (continue in same direction to “newsstand”)If (at “hospital”) Then (turn right and continue to “Park”)If (at “newsstand”) Then (turn right and proceed to “university”)If (at “newsstand”) Then (continue in same direction to “church”)

Page 10: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

“Blind” searches can determine the directions.

Blind searches contain no ancillary information about the problem.

Searches simply follow the IF-THEN rules. These may be accomplished in what are called “depth first” or “breadth-first” methods.

Software samples are readily available on the internet. A depth-first solution is shown on the next slide with the successful search shown in red.

A limitation with depth-first solutions is that an optimum solution may not be found – the search is over once the first solution is available.

Page 11: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Depth-first simulated search for finding directions

Page 12: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

A sample TSP problem

A

D

C

F

B

E

G

A sequence for the cities is bold (i.e., A .> C > B > E > G > F > D > A)

Page 13: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

The TSP Solution

Use the Evolutionary Computing Concept; a highly simplified diagram for a solution follows:

Arbitrary sequence

Calculate distance

Perturb sequence

New < old

Calculate distance

Perturb the city sequence by interchanging cities; if the cost is reduced the perturbation is the now the route; continue this until no lower cost is achieved. Keep a list of routes so that old routes are not repeated.

Page 14: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Expert Systems

Employs knowledge (facts); logic ( rules, primarily in the form of an IF-THEN calculus ); processes (algorithms for combining facts and rules)

WebMD is a popular Expert System that is freely available on the internet.

Challenges to Expert Systems: equivocation of facts; testing of the knowledge base may be difficult and unrealistic; “experts” disagree regarding what is important or may overlook information that they take for granted.

To overcome some limitations, Expert Systems may include statistical information (“there is an 80% chance that the bacteria is chicken pox”)

The CYC project seeks to expand knowledge to include metaknowledge with commonplace facts known to all.

Page 15: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Why not build a brain?

Blue Brain Project: started after 15 years of systematically dissecting the microanatomical, genetic and electrical properties of the elementary unit of the neocortex .

15,000 experiments in rat somatosensory cortex

first milestone (2006): creating, validating, and researching the neocortical column of a rat [brain]. This is the smallest functional unit of the neocortex which was described in connection with Hawkins’ work and is posited to be the part of the brain responsible for higher functions such as conscious thought.

accomplished using a supercomputer that was able to represent ANNs but with an added advantage of being able to simulate biologically realistic neuronal models.

a more challenging phase aimed to: simplify the column simulation leading to simulation of connected columns. This will culminate in being able to simulate a whole neocortex; in humans this is about 1 million such columns.

Completing this simulation has the potential to lead to an understanding of the very nature of consciousness itself.

Page 16: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Robotic Embodiments – Ultimate IAs robot – mechanical embodiment of an IA that can function

autonomously.

“autonomously” function with help from a human operator; is able to adapt to changing circumstances in the world that it inhabits; continue to work if a part breaks just as a human can continue to function if we only break a finger; is able to navigate and interact with the changes in world circumstances.

Page 17: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Robotic Models Siegwart and Nourbakhsh (2004): traditional aspects of the AI associated with IAs such as: an internal model of the world that does robotic planning and execution of locomotion based on that model.

Rodney Brooks (1999): subsumptive approach to IA design with its development of complex IAs based on integration of simple, “non-intelligent” components

Cynthia Breazeal: “social robots” particularly those imbued with “emotion”.

Ronald Arkin (Arkin, 1998): behavior-based reactive control and action-oriented perception for mobile robots and unmanned aerial vehicles. In this regard consider the “moral issues” that robotic development might engender (i.e., moral dilemmas faced by an automata on the battlefield regarding the taking of human life).

Page 18: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Hierarchical Architecture

SENSESENSE PLANPLAN ACTACT

The environment is modified.

IncludesSensors andFeatureextraction

Controls

actuators

Creates a model, a plan to complete a task:

Produces commands for the actuators

Page 19: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Second primitive behavior

Third primitive behavior

environment

sensor 1

sensor 2

sensesense actact

Primitive behavior

Reactive Architecture after Brooks

Page 20: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Simulating Robotic Planning

Start

Obstacles (walls)

Goal

Robot track

The robot senses and avoids the obstacles and arrives at the goal

Page 21: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

The robot is trapped in a box

Start

Robot track

box goal

Page 22: CHAPTER THIRTEEN The Artificial Intelligence II: Embodiments and Robotics

Challenges and Prospects for Implementing IAs

Computers are not close to achieving the perceptive, finely-honed reasoning and manipulative capabilities of adult humans. Machines can currently demonstrate the intellect of a low order insect.

The more we try to replicate human intelligence the more we may learn to understand human intelligence.

Designing computer-based machines that are intelligent is not the same as building computers that simulate intelligence.

We should not be obsessed with mimicking human intelligence.

Neural interfacing is an emerging technology that will permit us to directly control actuators using electromyography.