Summary Artificial Intelligence &
Neurocognition articles
Leiden University // Suzanne de Vries // Year 3 Block 2
, 1
L1: Is a machine realization of truly human-like
intelligence achievable? McClelland
Are people still smarter than machines?
Is it still true that people are smarter than machines, and if so: why?
Natural cognitive tasks → in vision, computational approaches have made substantial gains and machines can
match human performance after training. Although it was tailored to the task, it is still impressive. Very
important limitation is narrowness of the focus in systems that have some degree of AI, there are things that it
does not take into account that humans would. For humans, any source of information can play a role in
inferences and plans we make when making decisions and planning actions.
Why are people still smarter than machines?
Few would doubt machines lack fluidity, adaptability, creativity, purposefulness and insightness humans have.
Why does AI have limitations? Computational power has not reached the exaflop of 1018 yet.
Computational theory
Marr’s three level taxonomy: fundamental nature and goals of computational model + algorithms and
implementations they use. What information is available in the environment, and how can it be optimally used?
There is a lot to understand about the relationship between stimulus variables and underlying reality. Difficult to
know how to frame a computational problem, example: to learn something about the relationship between
situations and consequences. Two views: either construe the learner’s goal as one of extracting a structured
statistical model of the environment, or to define the goal in terms of the problem of optimal prediction,
allowing the internal model to remain inchoate instead of explicit.
Algorithm and representation
How a computational mechanism can actually exploit information effectively → which algorithms and
representations? Exploring the computational basis of the characteristics of brain representations.
Architecture
Question of ‘the cognitive architecture’. Used in modeling human cognition and building artificial cognitive
systems. A common theme is to stress some sort of hybrid combination of explicit symbolic and implicit, more
connectionist, sub-symbolic components.
Nurturance, culture and education
An additional step that is needed is the understanding of the roles of nurturance, culture and education in
structuring human cognition and understand how it develops.
, 2
L1: Computing machinery and intelligence Turing
The imitation game
Can machines think? The game: a man (A), woman (B) and interrogator (C), interrogator should determine
which of A or B is the man and which the woman. A has to make sure that C makes the wrong identification,
and B should help C to make the right identification. What will happen when a machine takes the part of A?
Critique of the new problem
Advantage of drawing a fairly sharp line between the physical and intellectual capacities of a man. Interrogator
cannot demand practical demonstrations.
Criticised on the grounds that the odds are weighted too heavily against the machine. Assumed that its best
strategy is to try and provide answers that would naturally be given by a man.
The machines concerned in the game
Only permit digital computers. We are not asking whether all digital computers would do well in the game nor
whether the computers at present available would do well, but whether there are imaginable computers that
would do well.
Digital computers
Intended to carry out any operations which could be done by a human computer. Three parts: store, executive
unit and control. The store is a store of information in which the information is usually broken up into packets of
moderately small size, the executive unit is the part which carries out the individual operations involved in a
calculation. The control has to see whether the instructions are obeyed correctly and in the right order.
If one wants to make a machine mimic the behavior of the human computer in some complex operation, one has
to ask him how it is done, and then translate the answer into the form of an instruction table (programming).
Interesting is the digital computer with a random element, and infinite capacity computers.
The use of electricity cannot be of theoretical importance despite the similarity to the human mind, only very
superficial similarity.
Universality of digital computers
‘Discrete-state machines’: move by sudden jumps or clicks from one quite definite state to another. Described
by tables. It will seem that given the initial state of the machine and the input signals, it is always possible to
predict all future states. Reasonably accurate knowledge of the state at one moment yields reasonably accurate
knowledge any number of steps later. Possible number of states is enormously large, related to the storage
capacity. Provided it could be carried out sufficiently quickly, the digital computer could mimic the behavior of
any discrete-state machine. Must have adequate storage capacity and be sufficiently fast, must be programmed
afresh for each new machine it should mimic.
Contrary views on the main question
Can machines think?
Neurocognition articles
Leiden University // Suzanne de Vries // Year 3 Block 2
, 1
L1: Is a machine realization of truly human-like
intelligence achievable? McClelland
Are people still smarter than machines?
Is it still true that people are smarter than machines, and if so: why?
Natural cognitive tasks → in vision, computational approaches have made substantial gains and machines can
match human performance after training. Although it was tailored to the task, it is still impressive. Very
important limitation is narrowness of the focus in systems that have some degree of AI, there are things that it
does not take into account that humans would. For humans, any source of information can play a role in
inferences and plans we make when making decisions and planning actions.
Why are people still smarter than machines?
Few would doubt machines lack fluidity, adaptability, creativity, purposefulness and insightness humans have.
Why does AI have limitations? Computational power has not reached the exaflop of 1018 yet.
Computational theory
Marr’s three level taxonomy: fundamental nature and goals of computational model + algorithms and
implementations they use. What information is available in the environment, and how can it be optimally used?
There is a lot to understand about the relationship between stimulus variables and underlying reality. Difficult to
know how to frame a computational problem, example: to learn something about the relationship between
situations and consequences. Two views: either construe the learner’s goal as one of extracting a structured
statistical model of the environment, or to define the goal in terms of the problem of optimal prediction,
allowing the internal model to remain inchoate instead of explicit.
Algorithm and representation
How a computational mechanism can actually exploit information effectively → which algorithms and
representations? Exploring the computational basis of the characteristics of brain representations.
Architecture
Question of ‘the cognitive architecture’. Used in modeling human cognition and building artificial cognitive
systems. A common theme is to stress some sort of hybrid combination of explicit symbolic and implicit, more
connectionist, sub-symbolic components.
Nurturance, culture and education
An additional step that is needed is the understanding of the roles of nurturance, culture and education in
structuring human cognition and understand how it develops.
, 2
L1: Computing machinery and intelligence Turing
The imitation game
Can machines think? The game: a man (A), woman (B) and interrogator (C), interrogator should determine
which of A or B is the man and which the woman. A has to make sure that C makes the wrong identification,
and B should help C to make the right identification. What will happen when a machine takes the part of A?
Critique of the new problem
Advantage of drawing a fairly sharp line between the physical and intellectual capacities of a man. Interrogator
cannot demand practical demonstrations.
Criticised on the grounds that the odds are weighted too heavily against the machine. Assumed that its best
strategy is to try and provide answers that would naturally be given by a man.
The machines concerned in the game
Only permit digital computers. We are not asking whether all digital computers would do well in the game nor
whether the computers at present available would do well, but whether there are imaginable computers that
would do well.
Digital computers
Intended to carry out any operations which could be done by a human computer. Three parts: store, executive
unit and control. The store is a store of information in which the information is usually broken up into packets of
moderately small size, the executive unit is the part which carries out the individual operations involved in a
calculation. The control has to see whether the instructions are obeyed correctly and in the right order.
If one wants to make a machine mimic the behavior of the human computer in some complex operation, one has
to ask him how it is done, and then translate the answer into the form of an instruction table (programming).
Interesting is the digital computer with a random element, and infinite capacity computers.
The use of electricity cannot be of theoretical importance despite the similarity to the human mind, only very
superficial similarity.
Universality of digital computers
‘Discrete-state machines’: move by sudden jumps or clicks from one quite definite state to another. Described
by tables. It will seem that given the initial state of the machine and the input signals, it is always possible to
predict all future states. Reasonably accurate knowledge of the state at one moment yields reasonably accurate
knowledge any number of steps later. Possible number of states is enormously large, related to the storage
capacity. Provided it could be carried out sufficiently quickly, the digital computer could mimic the behavior of
any discrete-state machine. Must have adequate storage capacity and be sufficiently fast, must be programmed
afresh for each new machine it should mimic.
Contrary views on the main question
Can machines think?
