Dhole Patil College of Engineering
HUMAN LEVEL ARTIFICIAL
Presented by – Rahul Chaurasia
T.E Computer Science
Div - B
R.No – T120604282
(Guide – Prof. Manisha Singh)
1. Definition of Artificial Intelligence
2. Goals of Artificial Intelligence
3. Today's Artificial Intelligence
4. Future Artificial Intelligence
6. HUMAN LEVEL MACHINE INTELLIGENCE (HLMI)
7. Assessment of Intelligence
8. Brain vs Hardware of a system
9. Ways towards Human level Artificial Intelligence
What is Artificial Intelligence(AI)?
John McCarthy, who coined the term in
1955, defines it as "the science and
engineering of making intelligent machines.
Artificial intelligence (AI) is
the intelligence exhibited by machines or software.
It is basically a study of how to create computers
and computer software that are capable of
intelligent behavior. It is "the study and design of
intelligent agents", in which an intelligent agent is
a system that perceives its environment and takes
actions that maximize its chances of success.
Deduction, reasoning, problem solving
Natural language processing (communication)
1) Social intelligence
3) General intelligence
Today's AI(Narrow AI)
1. Siri( Speech Interpretation
and Recognition Interface)
3. Self Driving cars
4. IBMs Watson
5. Autonomous Weapons
6. Facial Reorganization
7. Deep Blue - defeated the reigning world chess
champion Garry Kasparov in 1997
8. Proverb - solves crossword puzzles better than most
Future AI(Human level AI)
What are we looking for?
We are looking for a machine that can outperform
humans at multiple tasks and ideally at nearly every
Ways towards human level AI
Brain Inspired Computing
Quantum Weird stuff
We will consider
1) Machine Learning
Architecture of General Reinforcement Learning
Deep reinforcement learning model
Enhancing DRL with Predictive Model
2) Language Translation
3) Googles DeepDream
HUMAN LEVEL MACHINE INTELLIGENCE (HLMI)
Human level machine intelligence = Machine with a
A machine, M, has human level machine
intelligence if M has human-like capabilities to
Answer questions Remember
Architecture of general
Agent performs action which influences the
From the environment we get state
update(modifications to the state) and then we get a
certain type of reward.
Its task is to learn a policy over an action that will
maximize the reward over time. It's a trial and error
Optical Action Component
A component that finds an action that will maximize
the reward over time.
It is a predicted model. It predicts basically how the
world is going to carry on
Deep reinforcement learning
Example. Deep minds DQN (Google bought it for 400
Q) What Deep Minds DQN does?
Ans) It learns to play video game. Only input the DQN
has is the pixels on the screen and the reward(score).
It’s a reinforcement learning agent and it keeps on
trying different actions and improves the ability to play
the game better and better with time. And it does it by
We have the input to the agent which are the real
pixels on the screen.
Convolution Neural Network(CNN) – It feeds on the
data to other layers of neurons that ultimately feeds
the data to the function Q*.
Q* functions maps the action to their expected
rewards over time. Basically its gonna learn what the
best possible action can be to extract the best possible
reward in certain condition.
So basically DQN helps the system to learn the game
from scratch and can get to a human level ability at
Lets go back to the limitations on slide number 15.
Lets check the solution to the limitations.
It is basically an augmentation of basic architecture
with a predictive model.
Here Q* function doesn’t give the result directly but
rather also considers a predictive model which looks
ahead in time and predicts a result. Now we have two
results one is a basic result and other is the predicted
result. The best result is chosen and action is selected.
RNN – Recurrent Neural Networks.
They are good at learning sequential data.
Consider we are translating from English to French,
large data (English words) will be fed to the encoder
RNN. And this data will be paired with data(French
words) in Decoder RNN. It also predicts what the
future words are going to be based on the current or
prior words that’s why the name Thought Vectors.
DeepDream is a computer vision program created
by Google which uses a convolutional neural
network to find and enhance patterns in images
via algorithmic approach, thus creating a dreamlike
hallucinogenic appearance in the deliberately over-
Using the Model
1. Look ahead for threats and opportunities
2. Rehearse actions and plans
3. Search a tree of possibilities
4. Explore novel recombination's of behavioral
5. Think and Imagine
ASSESSMENT OF INTELLIGENCE
Every day experience in the use of automated
consumer service systems
The Turing Test (Turing 1950)
Machine IQ (MIQ) (Zadeh 1995)
THE CONCEPT OF MIQ
IQ and MIQ are not comprovable
A machine may have superhuman intelligence in some
respects and subhuman intelligence in other respects.
MIQ of a machine is relative to MIQ of other
machines in the same category, e.g., MIQ of Google
should be compared with MIQ of other search
Can we build hardware as complex as the
How complicated is our brain?
a neuron, or nerve cell, is the basic information processing unit
estimated to be on the order of 10 12 neurons in a human brain
many more synapses (10 14) connecting these neurons
cycle time: 10 -3 seconds (1 millisecond)
How complex can we make computers?
108 or more transistors per CPU
supercomputer: hundreds of CPUs, 1012 bits of RAM
cycle times: order of 10 - 9 seconds
YES: in the near future we can have computers with as many basic
processing elements as our brain, but with
far fewer interconnections (wires or synapses) than the brain
much faster updates than the brain
but building hardware is very different from making a computer behave
like a brain!
Prof. Murray Shanahan - Professor of Cognitive
Robotics in the Dept. of Computing at Imperial
College London, where he heads the Neuro dynamics