EXAM SUMMARY MAN AND
MACHINE 2020
Problem 1: outsourced intelligence
Cognitive science
Cognitive science= the science of explaining how people accomplish various kinds of thinking, it describes
and explains how, as well as explaining cases where thinking works poorly. Many cognitive scientists view
thinking as a kind of computation. Knowledge in the minds consists of mental representations; rules,
concepts, analogies, images and more, which are not acquired for the sake of accumulation but for solving
different kinds of problems. People perform mental procedures that operate on these representations to
produce thought and action. Cognitive modelling offers precision, transparency and heuristic value
Beginnings
1. Plato→ virtue
2. Aristotle, Locke and Hume→ empiricism; experience is what matters
3. Descartes, Leibniz→ rationalism; thinking and reasoning
4. Kant→ both rationalism and empiricism; experience and innate capacity
5. Wundt→ one of the first to systematically study mental operations
6. Watson→ behaviourism; there is no mind only behaviour
7. George Miller→ human thinking has limited capacity, they recode information into chunks
8. McCarthy, Minsky, Newell and Simon→’founding fathers’ AI
9. Chomsky→ people understand the grammar of a language via mental grammars consisting of
rules
− Founders of cognitive science
60’s, 70’s and 80’s→ power of rules, schemas, scripts, mental imagery. 80’s→ rise of connectionist
theories, 90’s→ increase in the use of brain-scanning technology.
The methods of cognitive science
Cognitive psychologists theorize about computational models, the primary method of testing is
experimentation w/ human participants, to investigate; how people make mistakes during deductive
reasoning, how they form and apply concepts, the speed of thinking w/ mental images and their
performance on problem-solving w/ analogies tasks.
These psychological experiments need to be interpreted w/I a theoretical framework that postulates the
mental procedures and representations. This can be done by forming and testing computational models,
,which is the central method of AI. Neuroscientists perform controlled experiments w/ the use of scanning
devices and provide additional evidence about brain functioning.
Cognitive science in its weakest form is merely the sum of psychology, AI, linguistics, neuroscience,
anthropology and philosophy.
The computational-representational understanding of the mind-
CRUM
The central hypothesis to CRUM is that thinking can best be understood in terms of representational
structures in the mind and computational procedures that operate on those structures. it might be wrong
but has nonetheless made considerable progress in understanding the mind.
The computer-brain analogy has provided a powerful way of approaching the mind: computer programs
consist of data structures and algorithms which can be defined to operate on various kinds of structures.
program Mind
Data structures + algorithms= running programs Mental representations + computational
procedures= thinking
Connectionist theories have proposed that neurons and their connections are the data structures, the
neurons firing and spreading activation is the algorithm.
− Crum then works w/ a 3-way analogy
mind
computer brain
The computers most of us work w/ are serial processors; one instruction at a time, but the brain is
capable of many operations at once and is therefore a parallel processor.
Theories models and programs
Computer models are useful for theoretical investigation of mental processes, comprehension of the
model requires noting the distinctions and connections among 4 elements:
1. Theory: postulates the representational structures and a set of processes
2. Model: precises the theory by interpreting by analogy
3. Program: tests the model by implementation in the program
4. Platform: the program can run on a variety of hardware platforms
The analogy between mind and computer is useful at all 3 stages of the development of cognitive theory;
discovery, modification and evaluation.
Evaluating approaches to mental representations
1. Representational power: how much information a particular kind of representation can express
a. Varies greatly per theory
, 2. Computational power: how much and how efficient
a. Problem-solving; explaining how people reason to accomplish goals
i. Planning
ii. Decision-making
b. Learning; sufficient power to explain how people learn
c. Language→ 3 aspects
i. Comprehension
ii. Production
iii. Learning of language
3. Psychological plausibility: also accounting for the quantitative results of psychological
experiments
4. Neurological plausibility: consistent w/ results of neuroscientific experiments
5. Practical applicability: there are many desirable practical results
There is not yet a single approach that fully satisfies all criteria.
Artificial intelligence and prediction
Artificial intelligence= the simulation of human intelligence in machines that are programmed to think like
humans and mimic their actions. The term can also be applied to any machine that exhibits traits
associated w/ the human mind such as learning and problem-solving.
Ai present an opportunity to make something that has been rather expensive abundant and cheap.
Artificial general intelligence
Artificial general intelligence is the hypothetical intelligence of a machine that has the capacity to
understand or learn any intellectual task that a human being can. Achieving this type of intelligence is the
goal of some AI-researchers, but the nightmare of others. Cognitive scientists hope to one day achieve
this intelligence by means of cognitive science, one method is CRUM, or imitating evolution.
The Turing test
The Turing test is a test of a machine’s ability to exhibit intelligent behaviour equivalent to or
indistinguishable from that of a human. A human judge judges a conversation between a person and a
machine, but the evaluator does not know which is which. If the evaluator cannot reliably tell which one is
the machine it has passed the Turin test→ Turing-point: when a machine exceeds human intelligence.
Machine learning
Machine learning= the ability of computers to learn w/o being explicitly programmed. A scientific study of
algorithms and statistical models that computer systems use to perform a specific task w/o using explicit
instructions, relying on patterns and inference instead.
In the real-world complexity increases exponentially and in environments like these machine-learning is
most useful→ the machine notices a correlation in example data and then references information from
past experience to predict something. An AI is fed 1000 pictures of apples, it is then shown a picture of an
apple and asked whether it contains an apple or not. The machine references its example data, notices a
, correlation between all the pictures of apples (round, red or green, etc..) and then decides that the picture
contains an apple.
Deep learning: the computer learns a way to relate data and categorize it
Artificial neuronal networks: mathematical, computational and technological models that mimic the logic
and learning functions of neurons in the brain.
Machine learning can be:
− Supervised: labelled data processing→ learning a function that maps an input to an output based
on example input-output pairs
− Unsupervised: unlabelled data processing→ learning an algorithm or function by inferring
patterns from a dataset w/o reference to known or labelled outcomes
− Causal: captures causal relationships of data
− Active: an algorithm can ask a user to label new data points w/ desired outputs, also called
optimal experimental design
o Is supervised
A major pitfall in machine learning (and AI in general) is that one has to be careful not to overfit or
underfit the data→if we ask a machine to solve a mathematical error, and it has very much power in the
physical world (just imagine for a sec) it could turn every surrounding computer into a calculator to help it
solve the problem. This would have a negative consequence for humans, so it is important to specify
exactly what it’s goals are and what is and isn’t allowed to do to get to those goals.
AI vs. Humans
Advantages of current AI over humans:
− It can find complexity in data; relevance, correlation, etc
− It can handle uncertainty
− It can make decisions faster
Advantages of humans over current AI:
− Humans have compassion and empathy as well as creativity (debate, see other task)
− It can respond to different needs
− Emotions
− Judgment
− Ethical
Levels and types of artificial intelligence
Types of AI
− Weak/ narrow: AI that can only perform simple tasks. Automation of time-consuming task→
Chinese room experiment
o Specific intelligence→ one task
