Influencing policy (training slides from Fast Track Impact)
What Artificial intelligence can Learn from Human Evolution
1. What we can learn from human evolution..
By
Abhimanyu Singh
Enrolment No.: 0441189907 MBA (SEM)
Under the able supervision of
Mr. Nidhish Shroti
Faculty & ERP Consultant, CDAC-Noida
2. Objective
To try to understand the behavior of human
intelligence and to find out points where artificial
intelligence can be benefitted from it.
3. Introduction
Human intelligence has developed for billions of
years through the process of evolution. Bit by bit,
features were sifted and deployed, ensuring we got
the best.
We humans now want to replicate this marvel. We
want to create an intelligence of our own, we want to
create the Artificial Intelligence. The only catch is
that we don’t have as much time.
I would like to put light on some of these features
that can be replicated to get closer to the goal of AI.
4. Artificial Intelligence
Attempt to make computers do things that right now
humans do better.
Related not only to Computer Science, but also to
Psychology, Physics, Anthropology, Biology,
Philosophy and so on..
Currently broken into specialized sub fields.
Amusingly enough, though AI has not evolved much,
people have already started working on the so called
“Robot Rights”.
5. Examples of Current Approaches to AI
Neural Networks
An artificial neural network (ANN) is a mathematical model or
computational model based on biological neural networks. It consists of
an interconnected group of artificial neurons and processes information
using a connectionist approach to computation. In most cases an ANN is
an adaptive system that changes its structure based on external or internal
information that flows through the network during the learning phase.
Fuzzy Logic
Fuzzy logic is a form of multi-valued logic derived from fuzzy set theory
to deal with reasoning that is approximate rather than precise. In binary
sets with binary logic, in contrast to fuzzy logic named also crisp logic, the
variables may have a membership value of only 0 or 1. Just as in fuzzy set
theory with fuzzy logic the set membership values can range (inclusively)
between 0 and 1, in fuzzy logic the degree of truth of a statement can
range between 0 and 1 and is not constrained to the two truth values {true
(1), false (0)} as in classic predicate logic. And when linguistic variables are
used, these degrees may be managed by specific functions, as discussed
below.
6. Human Intelligence characteristics
Motivated.
Emotional.
Simulator.
Regularly updated by the 5 senses.
Self Aware and conscious.
Predicts future, and can plan.
Can group things and show belongingness to groups.
Not bound by rules.
Social
7. Motivation
Provokes a person to do or not to do something.
Some of the popular theories of Motivation
Maslow’s Need Hierarchy (5 Factors)
ERG Theory (3 Factors)
Hertzberg’s Two factors Theory (2 Factors)
8. The Two Basic Factors
We can reduce all of them into the following two basic
factors of motivation,
Greed
Fear
We struggle to achieve things we want (Greed).
We Avoid things that we fear.
As we grow, we learn what we want to attain and what we
want to abstain.
Interestingly, we are not “Hard wired” to absolutely follow
these rules, e.g. when needed, we may even plunge into fire
to save someone we love, despite the fear of getting burnt.
9. Motivation for Computers??
This approach can make them autonomous or “Self
Motivated”, i.e. they don’t need to be told to do or not
to do something every time, they react to a situation
they sense around them.
Two Priority based stacks can be used to replicate
fear and greed.
Will assist in making decisions based on past
experiences.
Can be preloaded with a few basic instincts like
humans.
10. Social Behavior
Since it is possible that we may have conflicting
motives, we need to authenticate the sources to deal
with such situations.
Each person it “knows” is assigned with a priority,
which we can refer to as “trust”.
In case of conflicting motives the, higher priority
source will be preferred.
In such cases, the other person’s priority may be
decreased to mark out non trustable sources.
12. Emotions
We often consider emotions as hindrances to our
intellectuality.
Hitler tried to speed up the process of natural
selection and hence the speed of human evolution by
eliminating the old, weak and sick people.. Was his
decision intelligent??
Emotions bring rationalities to our decision making
process.
Behave as constraints to our motivations.
E.g. We do not start gobbling up sweets wherever we
see them, even if we like them very much.
13. The 5 Basic Emotions
Pleasure: The Reward for doing things we like. (Dopamine)
Pain: Physical pain draws attention towards a
malfunction, Mental pain associated with social
reasons.
Anger: The feeling that provokes us to fight against the
immediate danger.
Fear: The feeling that provokes us get out of a situation
where we cannot fight in case of danger. (Amygdala)
Disgust: The feeling that repels us from possibly
harmful objects.
14. The Law of Diminishing Returns..
States that for each unit being added for an activity,
the returns keep diminishing.
Required for emotions.
This helps in getting used to things and moving on.
Makes us dynamic, regularly urging us to try
something new.
Lack of it will lead to an almost static life, keeping us
in a state we already are.
Experiments have shown lack of it could lead to
death.
15. Senses
We have 5 Senses, namely the sense of Smell, Taste,
Touch, Sound and Vision.
Of all the senses, 3 senses can prove to be very
important for AI, i.e. Touch, Vision & Sound.
The sense of touch can be recreated easily, including
the feeling of heat and pressure.
The problem arises with the Sense of Vision and
Sound.
These provide the highest details of the surrounding
environment, and we will be focusing on these two.
16. Natural Language Processing
We use dictionary based lexical parsing.
Store words and their meanings in data dictionary.
Store people, place etc. and their identity in object
dictionary.
Learn new words while conversation, typically by
typing or in some cases through voice recognition.
Create replies based on grammatical rules and on past
experiences.
17. Problems with NLP
No physical world-logical world connection.
“Understanding” heavily marred by ambiguity present
in the sentences humans use.
Conversation has to be error free for proper
absorption.
We humans not always talk sense.
Most of the things we often say, has internal
meanings which are not understandable by computer.
Recent and older conversations has to be available so
as to talk sense and not be repetitive.
18. Eye Sight
We believe that we see whatever is present in front of
the eyes.
In reality, what we see is actually a recreated
interpretation of things in front of our eyes.
Even 2-Dimensional images are also interpreted in 3
Dimension.
Follows HSI color model, and not the RGB model.
Has two modes, Day Vision (Using Cones) and Night
Vision (Using Rods).
19. Eye Sight Continued..
We can identify reflection, and feel presence of
transparent & fluid objects.
We have different algorithm for face recognition.
Alphabets are interpreted differently (lack of it is Dyslexia).
Numbers are interpreted differently (lack of it is
Dyscalculia).
20. Image processing
Heavily noisy environment.
Need a lot of interpretation.
Consume a lot of Processor power.
Edges not clearly defined.
3-Dimensional objects have almost uncountable 2-
dimensional footprints, leading to almost useless
comparison of interpreted objects.
Reflections and transparency increase complexity.
22. What can be done..
The entire visible area doesn’t need to go through
detailed scan.
We can run an iteration of algorithms such as
RGB2HSI convertor, Edge detector, Histogram based
Object finder, Motion detector, Distance finder etc.
on the entire visible area.
Using all the above parameters, we can find Points of
Interests (POI) in the Field of View (FOV).
The points of interest are then sent for further deeper
analysis like Face recognition, character recognition,
Object Combiner etc.
23. The Language of the Brain
What is the language that the brain of a newborn
infant thinks through?
It doesn’t have any knowledge of any language, and yet
it thinks.
There must be some language that the brain thinks
through, and the NLP can be superimposed on it.
It is also compatible with the senses of vision, smell,
touch and taste.
Do we have such language for computers?
The Chinese room hypothesis asks exactly this
question.
24. 3 Dimensional Object Based
Thought Arena
Perhaps we can use our very own OOP concepts for
this purpose.
We need to create a 3D canvas where we can import
objects that we desire or see.
For our convenience, we can use the term Thought
Arena for it.
We can create or remove rules in/from this arena.
(Laws of physics, like gravity etc.)
We can assign attributes to the objects, morph them,
and tag them.
25. Thought Arenas
Research has shown that we can think about, up to 4
different things at a time.
This could mean that, we may be having up to 4
Thought Arenas in our brains.
One of these is obviously
dedicated to the real world,
and has got the highest priority.
Others may be dormant or
running in the background.
During sleep, these could get
activated causing dreams.
26. How does it work?
When we hear a voice coming from behind us, what
happens?
We instantly know that there is a person behind us.
We can gauge out who (s)he is, we can determine
distance etc.
For vehicles we even find out the speed and direction.
All this without even taking a look behind !!
The reason is that we instantly import the human
object or vehicle object into the Thought Arena and
assign attributes to it.
27. Contd..
The same happens with vision too.
We keep collecting data from the surrounding
environment.
We create a list of objects and their position in time
and space.
Even motion is stored as object, and helps us find
patterns in it.
This is remodeled in the Brain and a simulation is
started, which may lead to certain results.
Based on the results and instructions in the Motivation
stacks, the computer can react to situations.
28. 3D Objects
We normally create 3D
objects using Simple
Nodes and Vertices.
Each object has a
center of mass, which
is used as a reference
for activities like
collision detection,
movement control etc.
for the entire object.
29. Required 3D Properties
We need special types of Nodes which are
used for points where other nodes may or
may not be connected.
It will help in working with ambiguous
environment.
Also, instead of each object having a
single center of mass, we need each node
to have a center of mass.
The vertices can then behave as bonding
element.
Much like an atom, but bigger in size.
30. Time Based Memories
Short Term, Mid Term and Long Term Memories are
present to boost the response time of brain, so it can
be used for AI too.
Objects, that are of regular and daily use, are kept in
short term memory and so on.
Similar concepts are used in computers, e.g. HDD,
RAM and Cache. However they work on the physical
level.
We need to replicate the same thing at logical level.
31. Learning Process
Not all that comes in front of our eyes, goes into our
brains.
Cause-Effect relation is used to get lessons.
Brain mostly learns in two ways, either by Interest or
by Repetition.
Things of interest get direct entry into the brain.
However, things that we don’t like have to be kept
long enough in the short term memory so as to
qualify for entering into the higher levels of memory.
So, we may have to define a computer’s points of
interest, hence controlling its learning process.
32. It is easier to plan if we Planning
can visualize our
problem.
Simulation is essential for
planning, and this can be
done in the thought
arena itself.
With the presence of
multiple TA’s we can
simulate the behavior of
not only non living things
but also of living things
like human etc.
33. Conclusion contd..
The following simplified model is proposed for designing an effective
TA 4 TA 3 TA 2 TA 1
Greed Fear STM
Motivators
Time Based Memories
AI Droid.
LTM
MTM
Emotional
Evaluators
Thought Arenas
Body Interface
Stereo Vision
Stereo Sound
Object Data Dictionary
Image
Processor
Audio
Processor
34. Conclusion
For practical AI to be
implemented, we need
to develop the entire
“body” rather than just
the “brain”.
Concepts like Dreams
and Imagination may
not be alien to them.
35. Conclusion (contd.)
Since there is no limit to
their learning capability,
AI systems may
eventually develop cyber
psyche, exhibiting
humane emotions like
trust, fear, happiness etc.
36. Conclusion (contd.)
Can be used in combat zones, or
patrolling difficult terrains, as found
on the border areas.
Possibility of being hacked and
misused, so encryption must be used
for internal data. But volume of data
may lead to degraded performance.
There is a possibility of commercial
gains too. We may eventually be able
to bring down the cost of such systems
using economies of scale, thus
allowing them to be used as
“Manpower” for households and
industrial usage.
Editor's Notes
In my image editing project, I had implemented
Our brain manipulates the images received through the eyes. In the Left image the middle character may be interpreted as either B or 13. The right image shows that we convert 2-dimensional images to 3-dimensional before interpreting. Similarly we do not consider the entire image at the same time. If we keep looking into the blue ball, we cannot tell what is written on top, even though we know there is something written there.