Upload
danilo-fraga-dantas
View
217
Download
2
Tags:
Embed Size (px)
Citation preview
References
Computational epistemology: an overview
Danilo Dantas
Computational epistemology: an overview
References
PART I: Which epistemology?
Computational epistemology: an overview
References
Quine’s proposal
The stimulation of his sensory receptors is all the evidenceanybody has had to go on, ultimately, in arriving at hispicture of the world. Why not just see how this constructionreally proceeds? Why not settle for psychology?(Quine, 1969, p. 75).
Epistemology, or something like it, simply falls into place asa chapter of psychology and hence of natural science.(Quine, 1969, p. 82).
Computational epistemology: an overview
References
Naturalized versus traditional epistemology
Aim Method ReductionTraditional epistemology normative a priori noNaturalized epistemology descriptive empirical yes∗
Table: The ∗ is true of some naturalized epistemologies (e.g. Quine, 1969),but not of all (e.g. Goldman, 1986).
Computational epistemology: an overview
References
The desirable traits to epistemology
1. To be normative-grounding;
2. To employ non-controversial methods;
3. To be emancipated, but to benefit from empirical data.
Computational epistemology: an overview
References
PART II: Computational epistemology (CE)
Figure: http://xkcd.com/329/
Computational epistemology: an overview
References
Approaches to AI
Thinking Humanly Thinking Rationally“[The automation of] activitiesthat we associate with humanthinking, activities such asdecision-making, problem solving,learning (...)” (Bellman, 1978).
“The study of the computationsthat make it possible to perceive,reason, and act” (Winston, 1970).
Acting Humanly Acting Rationally“The creation of machines thatperform functions that require in-telligence when performed by peo-ple” (Kurzweil, 1990).
“Computational Intelligence isthe study of the design of intelli-gent agents” (Poole et al., 1998).
Table: Russell and Norvig (2010)
Computational epistemology: an overview
References
Epistemology as the description of the ideal agent
The ideal (but finite) rational agent is a finite rational agentwhich acts to achieve the best expected outcome in all possibleenvironments, and which does it using the less possible amount ofprocessing and time.
Computational epistemology: an overview
References
The desirable traits to epistemology
1. To be normative-grounding;
2. To employ non-controversial methods;
3. To be emancipated, but to benefit from empirical data.
Computational epistemology: an overview
References
Normative-grounding
S has grounds to believe that p in s ←→ The ideal agent believes that pin s
S is warranted to believe that p in sS is justified to believe that p in sS has reason to believe that p in s
S knows that p in s ←→ S believes that p in s& p is true& The ideal agent believes thatp in s
Computational epistemology: an overview
References
PART III: Methods and an example
Computational epistemology: an overview
References
What is 2SAT?
2SAT is the problem of determining whether a given propositionallogic formula in two-conjunctive normal form (2CNF) is satisfiable ofand providing an assignment that satisfies it.
E.g. does any assignment satisfies (C ∨ ¬D) ∧ (A ∨B) ∧ (¬A ∨ ¬C)?
Computational epistemology: an overview
References
Formalizing problems
If a problem can be described as a search problem, we may use theformalization in proposed by Russell and Norvig (2010, p. 66):
I The initial state;
I A function which returns the available actions in a given state;
I A transition model, which specifies the result of a given actionin a given state;
I The goal test, which determines whether a state is a goal state.
I A path cost function, which takes a list of pairs state-actionsand returns a number.
Computational epistemology: an overview
References
2SAT as a search problem
I The initial state is [x1, ..., xn], where x1 = x2 = ... = xn = 1.
I The available actions are to change the value of any number ofconstants pi from 0 to 1 or from 1 to 0.
I The transition model returns, for each action, the state withthe resulting assignment.
I The goal test is whether an assignment render the formula true(classical logic rules).
I The path cost function returns the number of changes in thetruth value of constants.
Computational epistemology: an overview
References
2SAT as a graph
[1, 1, 1]
S
[1, 1, 0]
[1, 0, 1]
[1, 0, 0]
[0, 1, 1]
[0, 1, 0]G
[0, 0, 1]
[0, 0, 0]
G1
1
2
1
2
2
3
Computational epistemology: an overview
References
Building agents
The design and test of a putative ideal agent have 3 stages:
1. The choice of a hypothesis to the ideal agent for a given problem,and the building of a model of the agent based in this hypothesis;
2. The implementation of the model in a computer simulation;
3. The analysis of the data from the simulation.
Computational epistemology: an overview
References
The agents for 2SAT
I Truth table agent;
I Truth line agent;
I Simplification agent.
Computational epistemology: an overview
References
Analyzing the agents 1
In order to be implementable as a model of the ideal agent, an agentmust meet some requirements:
1. to have consistent dispositions;
2. to be translatable into a programming language;
3. to be computationally accurate and feasible.
Computational epistemology: an overview
References
Analyzing the agents 2
In analyzing data, there are 5 important measures:
1. the accuracy rate;
2. the solution cost;
3. the time and space requirements;
4. the lower bounds.
Computational epistemology: an overview
References
Truth table agent: the plots
2 4 6 8 10
50
100
150
200
250
300
350
400
Solution cost
Constants
Pa
th c
ost
2 4 6 8 10
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
Time complexity
Constants
Assig
nm
en
ts
2 4 6 8 10
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
Space complexity
Constants
Assig
nm
en
ts
Computational epistemology: an overview
References
Truth table agent: solution cost
[1, 1, 1]
S
[1, 1, 0]
[1, 0, 1]
[1, 0, 0]
[0, 1, 1]
[0, 1, 0]
[0, 0, 1]
[0, 0, 0]
1
2
1
31
2
1
Computational epistemology: an overview
References
Truth table agent: time and space requirements
Constants Lines Time Space1 2 2× 10−6 seconds 2 bytes2 8 4× 10−6 seconds 8 bytes5 160 3.2× 10−5 seconds 160 bytes10 10240 1× 10−3 seconds 10 kilobytes20 4.1942× 107 1.0486 seconds 20 megabytes50 2.2518× 1015 35.7 years 50 petabytes100 2.5354× 1032 402 trillions of years 1.1259× 1017 petabytes
Computational epistemology: an overview
References
Truth line agent: the plots
2 4 6 8 10
50
100
150
200
250
300
350
400
Solution cost
Constants
Pa
th c
ost
2 4 6 8 10
200
400
600
800
1000
1200
1400
1600
1800
2000
Time complexity
Constants
Assig
nm
en
ts
2 4 6 8 101
2
3
4
5
6
7
8
9
10Space complexity
Constants
Assig
nm
en
ts
Computational epistemology: an overview
References
Simplification agent: the plots
2 4 6 8 10
1
1.5
2
2.5
3
3.5
4
4.5Solution cost
Constants
Path
cost
2 4 6 8 10
10
20
30
40
50
60
Time complexity
Constants
Assig
nm
ents
2 4 6 8 10
5
10
15
20
25
30Space complexity
Constants
Assig
nm
ents
Computational epistemology: an overview
References
Humans: the plots
2 4 6 8 10
0.5
1
1.5
2
2.5
3
3.5
Solution cost
Constants
Pa
th c
ost
2 4 6 8 103000
4000
5000
6000
7000
8000
9000
10000
11000
Time complexity
Constants
Tim
e (
ms)
2 4 6 8 100
1
2
3
4
5
6Accuracy rate
Constants
Err
ors
(%
)
Computational epistemology: an overview
References
PART IV: CE and other sciences
Computational epistemology: an overview
References
The desirable traits to epistemology
1. To be normative-grounding;
2. To employ non-controversial methods;
3. To be emancipated, but to benefit from empirical data.
Computational epistemology: an overview
References
Is this still philosophy?
Computational epistemology: an overview
References
PART V: The 2nd year paper
X
Computational epistemology: an overview
References
The Bayesian agent
1. The Bayesian agent holds degrees of belief in accordance with theaxioms of the probability calculus;
2. The Bayesian agent employs traditional probability calculus toolsto calculate degrees of belief;
2.1 In particular, in acquiring new data, the Bayesian agent updates(some of) its old degrees upon these data using Bayes theorem.
3. ∗ The Bayesian agent holds beliefs in propositions when itdegrees of belief in that proposition is higher than a threshold.
Computational epistemology: an overview
References
The defeasible agent (Pollock, 1995)
1. The defeasible agent adopts beliefs in response to construingarguments, provided no defeaters have already been adopted forany step of the argument;
2. The defeasible agent must keep track of the basis upon which itsbeliefs are held;
3. The defeasible agent must keep track of defeated inferences, andwhen a defeater is itself retracted, this should reinstate thedefeasible inference.
Computational epistemology: an overview
References
The Wumpus world
PIT
1 2 3 4
1
2
3
4
START
Stench
Stench
Breeze
Gold
PIT
PIT
Breeze
Breeze
Breeze
Breeze
Breeze
Stench
Figure: Russell and Norvig (2010)
Computational epistemology: an overview
References
References
Bellman, R. E. (1978). An Inrrocluction to Artificial Intelligence: Can ComputerThink? Boyd & Fraser Publishing Company, San Francisco.
Goldman, A. (1986). Epistemology and Cognition. Cambridge: Harvard UniversityPress.
Kurzweil, R. (1990). The Age of Intelligent Machines. MIT Press, Cambridge,Massachusetts.
Pollock, J. L. (1995). Cognitive carpentry: a blueprint for how to build a person. TheMIT Press.
Poole, D., Mackworth, A. K., and Goebel, R. (1998). Computational intelligence: Alogical approach. Oxford University Press, Oxford, UK.
Quine, W. V. (1969). Ontological Relativity and Other Essays, chapter EpistemologyNaturalized, pages 69–90. New York: Columbia UP.
Russell, S. and Norvig, P. (2010). Artificial Intelligence: A Modern Approach 3rdEdition. Upper Saddle River,EUA: Prentice-Hall.
Winston, P. H. (1970). Learning structural descriptions from examples. technical reportmac-tr-76. Department of Electrical Engineering and Computer Science,Massachusetts Institute of Technology, Cam- bridge, Massachusetts.
Computational epistemology: an overview