Midterm 1 Flashcards
Covers intro, neuroscience, neural computation, and some of artificial intelligence. Material from readings and slides. Also the target papers Koederink, 2011 & Sacks, 2003.
What are the branches of AI and what do each of them mean?
- Knowledge representation - processing and representing facts about the world in some abstract way (rules of chess game in IF-THEN logic)
- Pattern recognition - extracting knowledge from images (face identification)
- Reasoning and inference/prediction - deriving conclusions from incomplete/noisy observations (medical diagnosis, weather forecasting)
- machine learning - Improving performance from experience (AlphaGo, google search engine)
- 5.Planning - planning complex sequence of actions to achieve a goal (robot navigation)
- natural language - production and interpretation of spoken and written language (chatgpt)
What is artificial intelligence?
the science and engineering of making computer programs that exhibit characteristics of human intelligence
What is the scientific aim of artificial intelligence?
To understand the requirements for and mechanisms of human, animal, machine, robotic intelligence
What is the engineering aim of artificial intelligence?
To apply scientific knowledge to build useful artifacts (machines & robots) capable to do things done by humans or animals
What is the interface theory of perception? What society holds conferences for this topic?
Theory that our perception operates like a species-specific desktop: we perceive the world in representations that do not represent the “truth” about the world as it actually is but rather, are useful “icons” to represent fitness-relevant information about the world. The Psychonomic Society.
What is a blind search?
a search algorithm that examines all possible paths to a solution without using any additional information about the problem.
What is intelligence?
The computational part of the ability to solve problems and achieve goals in the world in an efficient manner.
What branches of AI does the perceptual dog illusion demonstrate?
Pattern recognition, knowledge representation, reasoning & inference
How was “Artificial Intelligence” coined?
At a summer workshop at Dartmouth in 1956 led by John McCarthy
What was the general problem solver (GPS)? What general problem solving strategy was it based on?
A computer program by Newell and Simon in 1960-70s that solved simple problems/puzzles like the tower of Hanoi. Based on mean-ends analysis (work backward from the goal to decide on what actions will help you achieve in which goals are decomposed into subgoals in a recursive fashion). Influenced development of symbolic logic.
What is the weak vs strong AI view?
Weak AI view: The computer gives us powerful useful tools to aid human activities but can never exhibit human-level intelligence.
Strong AI view: A programmed computer can truly think and can therefore explain how the human mind works, helping us to create an artificial mind. AKA - artificial general intelligence
How have our approaches to AI changed?
symbolic systems approach: 1950-2000 traditional AI. CPU. Machine learning starting at 1980.
neural network approach: starting after geoffrey hinton breakthrough 2010s. Deep learning & then deep neural networks around 2011. GPU.
Who defined the notion of computation? What were the effects of this?
Alan Turing. Subsequently led to a definition of intelligence or intelligent agents such as humans, animals, and ultimately, all sorts of “automatic” machines (ATM, cellphone)
Who is the founder of modern computer science?
Alan Turing
What is the Turing machine (TM)? What are the three components of the machine?
gives a theoretical definition of computation (so intelligence); NOT an actual physical machine; an idealized mathematical abstraction of a digital computer.
- 1-dim tape of cells of unlimited length (written on each cell is a symbol from finite alphabet)
- read/write head
- control (action) table or program (if -> then actions)
What is the difference between a CPU and GPU
CPUs are general-purpose processors that can handle many tasks, while GPUs are specialized processors that excel at parallel processing
Von Neumann machines (VNM)
Modern computers are VNM. They can accomplish what TM do in a long sequence of steps in just one step.
Differences between VNM and TM
TM: only has one register, has linear access memory, had only a few hardware instructions, has a control table.
VNM: has several dozen registers, has random access (addressable memory), has several dozen instructions in CPU, has a program that controls CPU instead of a control table.
Turing’s definition of computation
It is turing computable (turing computable functions or problems) if it can be implemented on a turing machine; ie the solution can be found as a “halting” output written on the tape of a turing machine.
What is a universal turing machine and what are examples of them?
A single turing machine that can simulate any other turing machine. Programmable computers (PC, mac, von neumann machine) & DNA (biological computers)
What is the Church Turing Thesis? (CTT)
It states that a function on the natural numbers can be calculated by an effective method if and only if it is computable by a UTM
Assuming the CTT is true, we can define computation or computability as:
a problem is computable if it can be solved in a finite number of steps by a UTM, and uncomputable if no UTM can solve the problem.
What would invalidate the CTT?
If we found a non-UTM that solves what is thought to be an uncomputable problem.
What are examples of uncomputable problems?
The halting problem - based programs code and input, will it ever finish running?
The word problem for cancellation semigroups
What is the imitation game?
Three players. Player A tries to trick interragor C into concluding he is a woman. He succeeds C misidentifies him 30% of the time in a 5 minute conversation. Proves player A is ‘intelligent’ enough to fool player C.
What is the Turing Test
If a machine takes on the part of player A in the imitation game - can machines think?
- an attempt to define an operational definition of intelligence.
- behavior based test - directly probing at the inner workings of a UTM
- Turing described intelligent behavior as the ability to exhibit human-like conversational behavior, sufficient enough to fool an interrogator
What is the Chinese Room Argument?
a thought experiment to show that something (the chinese room) can pass the Turing Test without being genuinely intelligent. Against the Strong AI view.
Other than the turing test and chinese room argument, what are some forms of intelligence tests?
- CAPTCHA - reverse turing test, measures if the user acts like a human or computer by relying on the human visual system.
- Loebner Prize - competition that awards prizes to computers that are the most human like. Uses turing test. Judges have to decide if a computer or person behind a wall is the human.
- IBM deep blue first system to beat world champion in chess.
- IBM watson beats human jeopardy players
What was the first chatbot made?
ELIZA, the computer therapist (1966)
In what sense is intelligence a type of computation?
If it is a UTM that computes in an efficient (fast) manner so that it could pass the turing test.
How does the computation time of a UTM change it’s computablitity?
Finite –> computable. Fast = intelligent; slow = unintelligent.
Infinite –> Uncomputable/intractable
What has disproven the church turing thesis (CTT)?
Quantum computers since they are not a turing machine by design
What is cognitive science?
the interdisciplinary study of intelligence and intelligent systems, natural or artificial
5 foundations of cognitive science
Linguistics, psychology, philosophy, computer science, neuroscience
What is vertical integration?
Integration across levels (membranes, molecules, ions –> synapse –> microcircuits –> neuron –> centers and local circuits –> systems and pathways –> behavior)
What are the three levels of description (david marr)
Computational (defining problem)
Algorithmic (approach to solution)
Implementational (neurons/neural nets)
What is horizontal integration?
across domains (ie perception, memory, language, reasoning, etc)
What is the physical symbol system hypothesis?
a physical symbol system had the necessary and sufficient means for general intelligent action.
What’s an example of a physical symbol system
a theory in the field of artificial intelligence, proposed by Allen Newell and Herbert Simon, which states that a system capable of manipulating symbols (like a computer) has the necessary and sufficient means to exhibit intelligent behavior
a decision tree algorithm
What’s an example of a large behavior model (LBM)?
TRI’s (toyota) imitation learning based robots
Have unified theories of cognition been achieved yet?
no
What brain region is the “seat of intelligence”?
Cerebrum
What is in the cerebrum?
Cerebral cortex, limbic system (amygdala, hippocampus, cingulate gyrus), & basal ganglia
What is in the brain stem?
Midbrain, pons, medulla
Cerebral cortex function
sensory and cognition
limbic system function
emotion and learning
amygdala function
emotion
hippocampus function
learning
cingulate gyrus function
emotion
basal ganglia function
action control and selection
brain stem function
autonomous body functions
midbrain function
sleep, arousal
pons function
walking
medulla function
breathing, heart rate
What three structures are at the core of the brain?
cerebrum, brainstem, cerebellum
What are three main subcortical structures?
amygdala, hippocampus, basal ganglia
What are the six principles of brain organization?
Hierarchical organization, functional specialization, massively interconnected organization, layered organization, columnar organization, the neuron doctrine
What are the levels of brain organization (hierarchical)?
neurotransmitters, synapses, neurons, neural networks, systems, behavior
How many neurons do we have?
1,000 billion
How many synapses per neuron?
1,000
How much connectivity is there within 1mm^2?
only 3%, sparse
ratio of excitatory to inhibitory neurons:
90%/10%
What’s the neuronal processing time?
~ 1 msec
How many spikes per second?
50; ie internal clock speed of 50 Hz
What’s an asynchronous vs synchronous clock? What do humans have?
In a synchronous clock system, all operations are coordinated and rely upon a shared clock source. An asynchronous clock system means that components operate independently of one another, potentially with different clock sources. Our brain is an asynchronous clock; computers are synchronous.
What is the principal of re-entry?
Reentry in nervous systems is the ongoing bidirectional exchange of signals along reciprocal axonal fibers linking two or more brain areas.
Where is the most horizontal (layered) organization?
neocortex - exp. motor and sensory areas
What’s the best evidence of vertical (columnar) organization in the brain?
orientation columns in the visual cortex
What is the neuron doctrine?
The neuron is the fundamental structural and functional, (and computational) unit of the nervous system
What non-invasive method is sometimes used for brain-machine-interfaces (BMI)?
EEG
Where do we pick up signals for BMI?
motor cortex
What are some examples of uses for BMI?
move prosthetics, computer control, handwriting, neuromorphic artificial skin, brain to brain interface (BTBI), artificial touch via robot hands
what are the three stages of neuronal computation?
synaptic transmission, voltage summation, thresholding and generation of action potential
Steps of synaptic transmission:
presynaptic action potential at axon –> neurotransmitter release at synapse –> postsynaptic potential at dendrite (electrical voltage) (EPSP or IPSP)
What are the types of post synaptic potentials and what do they do?
Excitatory - depolarize the postsynaptic membrane
Inhibitory - hyperpolarize the membrane
What are graded IPSP and EPSP
Vary in magnitude depending on how much transmitter
was released by the sender neuron
Attenuated with distance within the receiver neuron
Temporal vs spatial summation in PSP
Spatial - ratio of EPSP to IPSP on multiple synapses
Temporal rate of EPSP in a single neuron
What is the equation for a single neuron computation?
Output signal = sum of the inputs and their weights - threshold of the neuron
{0,1} binary neuron
What are the six distinctive features of brain computing?
(1) Super-slow “clock” speed
(2) Unidirectional information flow
(3) Asychronous clock (no central co-ordinating clock; processing units are
activated only when needed, thereby reducing energy consumption)
(4) Parallel computation
(5) Integration of memory storage and information processing at synapse
(as opposed to the von Neumann machine in which over 90% of processing
time/energy is spent on shuttling data back and forth)
(6) Ultra-low power consumption
What is the most important factor in the brain’s ultra low power consumption? Second most important?
- Integration of memory storage and information processing at synapse.
- asynchronous clock
What are the EU and US brain initiatives and what are their goals?
EU - Human Brain Project: visualized some fibers of the hippocampus, wanted to make map of whole brain but unsuccessful
US - the BRAIN initiative: improving technologies to understand the brain
What is the farthest we’ve gotten with mapping a whole brain?
the fruit fly - 130,000 neurons
