Artificial Intelligence

Introduction
 AI Generations
• 80’s:
– we start worrying about hardware. Maybe
hardware could be scary.
– Maybe this technology that we're building could
turn against us.
• 90’s
– we realized that software can be scary
you're moving from hope that this technology
can make our lives better to, what if it doesn't?
 What If
• AIs will defeat us or rise up, to a little bit more
• What if they're indistinguishable from us?
• What if we get replaced?
• The fear here is not so much about a war, but
a replacement.
Reality Check!
• we are and what we can and can't do.
 The Express Tribune
AI miracle cure in the making
By Safdar Rizvi
Published: July 12, 2019

• Artificial Intelligence (AI) could soon be

playing an important role at hospitals
across the country, performing
important diagnostic tests and even
surgeries with greater precision than
surgeons. Such medical advances are
expected under the federal
government’s first Artificial Intelligence
project.
• The AI project is expected to cover
human resource training for various
health fields, including medical
diagnostics related to MRI, X-Ray and
CT scan. The technology will allow
diagnostic testing through robotic
examination.

https://tribune.com.pk/story/2011286/1-ai-miracle-cure-making/
Presidential Initiative for Artificial
Intelligence & Computing (PIAIC)
 Elon Musk Worries about AI
• - That the biggest risk we face as a civilization is actually
artificial intelligence.
• - What if this stuff works?
• - What if it works really well?
• -What if it doesn't do what we expect or what we want?
• - How do we define
• - what we want and how to keep these things safe?
• - What's the difference between what we can do, what
we can't do?
_ _ _ __ _ _ __ _ _ __ _ _ __ _ _ __ _ _ __ _ _ __
AI: Chapter 1: Introduction 15
What is Artificial Intelligence? (again)

AI: Chapter 1: Introduction 16

Acting humanly

() 19
 Case Study
Pac-man Game
• Pac-Man game as a benchmark test bed originally developed for UC
Berkeley’s AI introductory course CS188 available at
http://ai.berkeley.edu/project_overview.html

Equivalent Layout in (X, Y) Coordinates of Hurdles and Valid

Capture image from Pac-man game
Movement as a grid of 18 * 9 tiles

Ghosts  DFS, BFS, UCS, and A* Algorithms

Environment  {Pacman, Ghost1, Ghost2 Tile Positions}

Department of Computer Science IIUI () T hesis Proposal 20

 Case Study
Game Agent Development
• Designing this function is complex and it must be implemented for specific games.
• The agent replaced human player input through keyboard and provide next move for maze chase
game as Pac-Man.

Tools Used:
• MATLAB/ Simulink Enviournments
– Neural Network Toolbox
– Fuzzy Logic Toolbox
• Python IDLE 3.6
– “Pymatlab” package
• SPSS: Statistical Package for the Social Sciences

Modules Developed
• Module 1: Accessor methods has been modified to access the state data
of all three agents of Pac-man game environment
• Module 2: Middle Tier Function received (x, y) coordinates of Pac-man
agent from trained network and pass it to Pac-man Game to run
• Module 3: A hurdle and wall control function

Department of Computer Science IIUI () T hesis Proposal 21

 Acting Humanly
• The Turing Test (1950)
– Can machines think?
– Can machines behave
intelligently? Human
?
• Operational test for Human
intelligent behavior Interrogator
– The Imitation Game

AI System

AI: Chapter 1: Introduction 22

AI: Chapter 1: Introduction 23
AI: Chapter 1: Introduction 25
AI: Chapter 1: Introduction 26
 Thinking Humanly
• 1960’s cognitive revolution
• Requires scientific theories of internal activities of
the brain
– What level of abstraction? “Knowledge” or “Circuits”
– How to validate?
• Predicting and testing behavior of human subjects (top-down)
• Direct identification from neurological data (bottom-up)
• Cognitive Science and Cognitive Neuroscience
– Now distinct from AI

AI: Chapter 1: Introduction 30

 Proposed Framework

Department of Computer Science IIUI () T hesis 32

 Case Study
• We used public fMRI data which are available at
‘http://cnl.web.arizona.edu/spm.htm’
• Region of Interest(ROI)
– Brodmann areas 10 and 47 from the
prefrontal cortex of human brain

Block 1 Block 2 Block 3 Block 4 Block 5 Block 6 Block 7 Block 8

20 sec 20 sec 20 sec 20 sec 20 sec 20 sec 20 sec 20 sec

10 TRs 10 TRs 10 TRs 10 TRs 10 TRs 10 TRs 10 TRs 10 TRs

4 easy birds 4 controls 4 hard birds 4 controls 4 easy birds 4 controls 4 hard birds 4 controls
33
 Experimental Setup

Department of Computer Science IIUI () T hesis 34

 Three Conditions: (i) hard-to-learn (ii) easy-to-learn
and (iii) familiar birds as control condition
H: Hard birds (5 seconds)
E: Easy Birds(5 seconds)
B: Baseline (crosshair fixation), ~1 seconds
4 easy birds blocks & 4 hard blocks.
Each block lasts for 5s and has 80 images
Each image was presented for 5s

time
time

RH

NrOfSlices: 17 SliceThickness: 5 mm
FoV: 220 mm Matrix: 128 x 128
NrOfVolumes: 126
Visual Stimulation
Protocol :
• Objects in LVF
• Objects in RVF
• Fixation
time LH
time
 Slice Terminology

VOXEL
(Volumetric Pixel)
SliceThickness In-plane resolution
e.g. 4 mm e.g., 220 mm / 128
= 1.72 mm
1.72
mm 4 mm

SAGITTAL SLICE IN-PLANE SLICE

1.72 mm

Number of Slices
e.g., 17

Matrix Size
e.g., 128X128

Field of View (FOV)

e.g., 220 mm

36
Activated Areas

Response to
Response to
easy Bird
Hard Bird

37
 Data Pre-Processing
Image time-series Kernel Design matrix Statistical parametric map (SPM)

Realignment Smoothing General linear model

Statistical Gaussian
Normalisation inference field theory

Template p <0.05

Parameter estimates
38
 Experimental Setup
Performance Measures

• Following performance evaluation measures have used.

1. Mean Square Error (MSE)
2. Accuracy
3. Sensitivity
4. Specificity
Statistical validation methods
• K fold cross validation (10 and 20 fold)
• bootstrapping
• 70-30 training/testing
• Tools Used
– SPM ( Statistical Parametric Mapping)
– MATLAB/ Simulink Enviournment
Data Pre-Processing Tools fMRI Data
– mricorn
– AFNI
– FSL
() 39
 Introduction
Human Decision Making
The enormous success of applications of functional brain
imaging techniques such as EEG and fMRI to make it possible
in the study of human decision-making.

• Types
– Risky decision making
– Social decision making
• Rewardless Decision making

• Examples
• what you want to wear today
• choice of beverage in a café
• weather or not buy a new car at the moment

() 40
 Problem Statement
Human Decision Making

Different patterns have generated during

stimuli. All such information has passed into
visual area and frontal areas of brain. All
information is mix in regions (BDs 10 & 47).
The real challenge is observe the pattern of
these regions and to classify the patterns.

() 41
 Related work
• Objective
– Identifying the most dominant voxels of prefrontal cortex in the human brain which helps
a person in rewardless-related decision making using ANN technique.
– For classification of decisions made by respondent and for identification of the most
dominant voxels of human brain for decision making, they analysed the Brodmann area
10 and 47 from the prefrontal cortex of human brain

• Methodology
– Multilayer Perceptron (MLP) neural network
– Evaluate training accuracy based on 40 voxels
• Issues
– Due to limited data, the algorithm is often stuck on local optima with very low fitness.
– The performance of backpropagation and test statistics is also compromised due to local
gradient data[38]
– When the training size is not large enough, ANN will have to face the problem of
overfitting and reflect biases [39]

F. Ahmad, I. Ahmad, and W. M. Dar, "Identification and classification of voxels of human brain for rewardless-related decision making
using ANN technique," Neural Computing and Applications, vol. 28, pp. 1035-1041, 2017.
() 42
 Research Issues
1.The applicability of the hybrid techniques in
the domain of cognitive neuroscience is yet to
be explored

2.To find the appropriate Artificial Intelligence

algorithms

3.There is a great need to use large data to avoid

the problem of overfitting.

() 43
 Research Goals
Goal : Development of a model in the domain of behavioural neuroscience, to
find strongly activated brain regions, their specific cognitive processes and
stimulus-trigered activities during rewardless-related decision making using
functional magnetic resonance imaging (fMRI) data

To achieve our goal, following research questions have been identified.

1. What is an effective mechanism to cognitive states classification for

decision-making voxels ?
2. How to derive a measure of classification of rewardless decision making
fom the likelihood ratios between the models and which model gives best
results?
3. While identify the dominant activation patterns, which training method
produces the best results?
4. Which method is best to find significantly activated brain regions

() 44
 Acting Rationally
• Rational behavior
– Doing the right thing
• What is the “right thing”
– That which is expected to maximize goal achievement,
given available information
• Doesnot necessary involve thinking –e.g., blinking
reflex-but involve thinking should be in the service of
rational action.
• We do many (“right”) things without thinking
– Thinking should be in the service of rational action

AI: Chapter 1: Introduction 46

 Unpack
• What does maximize mean?

• What are we maximizing over?

• What are the algorithms for finding optimal things in that sense?

• What is utility?

• What are we trying to achieve?

• What is the expectation here?

• How does our uncertainty over the environment

 what AI can do today?

Quiz: Which of the following can be done at present?

• Play a decent game of table tennis.
• Drive safely along a curvy mountain road.
• Buy monthly grocery from the web
• Discover and prove a new math theorem.
• Converse successfully with a person for an hour
• Perform a surgical Operation.
• Translate spoken chinese in English real time.
• Fold a laundry and put away dishes.
• Write an intentionally funny story.
Artificial Intelligence (AI) Subcategories
 Introduction
• Digital Games
– Play Games
– Model Players
– Generate Content
• Robotic Surgery
– Complex Actions in Critical Situations
– Accurate Every Time
• Autonomous Driving
Modeling Modeling
– Behaviour
No more accidents to human error
Human
– No more traffic jams Behaviour Behaviour
• Industrial Automation
– Efficiency
– Precise Quality Control
– Safety
• Healthcare (Assistive Robotics)
– Elderly Care
– Rehabilitation
– Special needs
• Conversional Agents
– Assistance
Imitation Learning
– Recommendation
– Therapy
• Others
– Modeling Internet and Web
– Social Networks
– Criminology

74
()
 References
• http://ai.berkeley.edu
• Prof. Mian Muhammad Awais
– CS 531
• Lahore University of Lahore, Pakistan