— John McCarthy

➢ Nigel has a charging station he returns to when his battery drains.

▪ Nigel doesn’t move randomly; he forms a plan about how to clean.
Lecture 1 - Introduction: What is Artificial ▪ Nigel does this by investigating and building a map of the room.
Intelligence? ▪ Is Nigel an example of artificial intelligence? What do you think?

The Dartmouth Summer Research Project on

Artificial Intelligence: The big questions Artificial Intelligence (1956)
1. Can we build machines that think? The first AI “hypothesis” à “Every aspect of learning or any other feature
2. Can we build machines that learn?
3. Can we build machines that are more intelligent than us? of intelligence can in principle be so precisely defined that a machine
4. Can we build machines that are creative? can be made to simulate it”.
5. Can we build machines that have emotions?
6. Can we build machines that are conscious?
John McCarthy, Marvin Minsky, Claude Shanon, Ray Solomonnoff,
Alan Newell, Herbert Simon, Arthur Samuel
What is Artificial Intelligence?
Artificial intelligence is:
• “the science and engineering of making intelligent machines”
— John McCarthy.
• “the science of making machines do things that would require
intelligence if done by men”
— Marvin Minsky (1968)
• “The exciting new effort to make computers think... machines with
minds, in the full and literal sense”
— John Haugeland (1985)
• “The art of creating machines that perform functions that require Strong AI vs. Weak AI
intelligence when performed by people”
— Ray Kurzweil (1990)

What do we mean by “artificial”?

➢ Must “real” intelligence be made from biological stuff?
▪ This is known as carbon/protoplasm chauvinism.
▪ What if we work out how to engineer biological agents?
➢ Must “real” intelligence be the product of biological evolution?
▪ Should nature be the only source of “real” intelligence?
▪ What if we work out how to evolve biological agents?

So, we seem to mean two things:

➢ By artificial, we might mean non-biological. Here, it’s all about the kind of stuff
we use to build an agent. AI as Science vs. AI as Engineering
➢ By artificial, we might mean constructed by humans. Here, it’s all about the
origin of the agent, and who designed and built it.
What do we mean by “intelligence”?
➢ Few concepts in psychology have received more devoted attention and few
have resisted classification so thoroughly”
— A.S. Reber
➢ “Intelligence is the computational part of the ability to achieve goals in the
world. Varying kinds and degrees of intelligence occur in people, many
animals and some machines”
 Artificial General Intelligence vs. Narrow AI

The Rational vs. The Psychological

• Should the study of AI be guided by the psychological (our
understanding of how humans work), or the rational (our
understanding of logic, probability theory, etc.)?
• This distinction impacts both on the kind of systems we study, and
the kind of behavior we desire...

Most AI research doesn’t tackle the big AI questions directly.

Lecture 2: Cognition and Computation

 • The workers carrying out these calculations were called “computers”.
What is going on inside our heads?
• Throughout history, scientists have claimed that the activity going Can this work be automated?
on inside our heads is mechanical.
• During the Renaissance, it was thought that this mechanical activity In 1812 Charles Babbage imagined how these calculations could be carried out by a
resembled a clockwork device, and later on, a steam engine. machine, rather than “computers”.
“Mr. Babbage’s invention puts an engine in place of the computer”.
• Within the last century, the metaphor of the telephone exchange Next programmable computers – Zuse’s Z3, Colossus, ENIAC
has been invoked.
Computers resemble the mind/brain

• The computer is like the mind:

Alan Turing asked the question “Can machines think?” rather than “Is the
mind a computer?”
Turing considered the similarity “very superficial”

• The computer is like a brain:

The computer metaphor John von Neumann drew the analogy between computers and the
nervous system, based on the work of McCulloch & Pitts.
• The computer is the last metaphor; it need never be supplanted” • These analogies likened the computer to the mind/brain, rather than viewing
the mind/brain as a computer
(Johnson-Laird,1983)
• The brain is like the hardware, it’s a physical device.
• The mind is like the software because it requires the physical Cognitive revolution: The mind/brain as a computer
device to operate,
Crucial realization: Rather than carrying out exclusively numerical operations, Newell
but in itself is not material since it has no mass.
and Simon viewed computers as symbol manipulators.
• The computer metaphor states that the relationship between brain and
mind is the same as that between hardware and software. “The new mental chemistry”:
The atoms of this mental chemistry are symbols, [. . . ] combinable into larger and
more complex structures. The reactions of the mental chemistry [. . . ] employ el-
Why view thought/mental activity as elementary information processes: storage, retrieval, transmission, and comparison“
computational? (Simon, 1979)
”within 10 years, psychological theories will take the
form of computer programs“ (Simon, 1957)
The physical symbol system hypothesis

“The computer is a member of an important family of artifacts [. . .] physical symbols

systems. The hypothesis is that a physical symbol system has the necessary and
sufficient means for general intelligent action. (1976) “

1. The ability to manipulate symbols is necessary for intelligent action

(cognition is computational).
2. Manipulating symbols is sufficient for intelligent action (computers
can cognize).

The first computers Summary: Full circle

• The French government set about constructing tables of very accurate 1. Mental activity is the act of calculation (Leibniz).
numbers. 2. Creative, inspired human thought was beyond mere calculation (Romanticism).
3. Employ people to perform calculations, call them computers (De Prony).
• De Prony spearheaded this huge effort.
4. Replace these people with a machine, call it a computer (Babbage).
5. Devise programmable computers (WW2).
6. Notice how computers resemble the mind/brain (Turing).
7. Hypothesize that cognition is symbol manipulation, we are members of a broader
class (Newell and Simon).
8. Mental activity is the act of computation.

There are many ways of thinking about and formulating computation

➢ These above are all models of universal computation.

➢ Whatever one kind of machinary can compute, they all can.
➢ We can translate between these ways of formulating computation.
➢ Problem: Defining computation is arguably impossible.
➢ However: The theory of computation is a rich area, and relies on...

What is computation?
Important observations about Turing machines
1. Extending the design of the basic Turing Machine doesn’t make it any Example Problems and Their Complexity
more powerful.
2. Turing Machines are notional. Although it’s in principle possible, 1. Problem: Sorting two lists of numbers of length n using Bubblesort.
nobody builds physical Turing Machines. Complexity: O(n2). Remember, we ignore constants.
3. The Turing Machine is a universal model of computation: The class of 2. Problem: Determine if a given number n odd or even.
problems they can solve is identical to those solvable by a standard von Complexity: O(1). Requires just one operation.
Neumann architecture computer, a neural network, or any other model of 3. Problem: Find the max value of a list of n (unsorted) numbers.
universal computation. Complexity: O(n).
What are the limits of computation? 4. Problem: Determine if two-character strings of length n are equal.
Complexity: O(n).
Two issues need to be considered when asking if a problem to be 5. Problem: Determine if a string of length n a palindrome.
solvable by a computer: tractability and computability. Complexity: O(n).
1. Intractable problems are those that demand infeasible resources to
solve. Key resources are time (the solution takes too long to compute)
and space (finding the solution requires too much memory).
What kind of problems are computationally
2. Uncomputable problems are those that cannot be solved by any intractable?
computer. Some problems are probably uncomputable.
Example: The Traveling Salesperson Problem
Many problems we consider in artificial intelligence are computationally Given:
intractable, and some are uncomputable. 1. A set of n locations.
2. Connections between locations.
3. Distances between pairs of connected of locations.
Find:
4. Find the shortest tour, where a tour starts and ends at the same location, and
visits all n locations.
The traveling salesperson problem is computationally intractable.
A naive approach is to consider all possibilities, but this is O(n!)
See how fast it grows...
n n!
1 1
2 2
9 362,880

13 6,227,020,800

20 2,432,902,008,176,640,000

Some algorithms improve on this time complexity.

How do humans deal with computationally

 intractable problems? ➢ Actions available in initial state: (0, 1) (0, 2)
➢ Actions not available in initial state: (1, 0) (1, 2) (2, 0) (2, 1)

Traveling Salesperson Problem: Human Performance

Lots of research into human performance tells us:

Humans do really well at the Traveling Salesperson Problem.
Path lengths typically within 2-4% of shortest tours.
Humans typically solve the problems in time linear in n.

Practical 1 - Problem Representation On one side of a river, there are three gamblers and three robbers. They both want
to travel to the other side. However, there is only one boat that can hold up to two
people. If the robbers outnumber the gamblers, they will rob them. What sequence of
Example from Russel & Norvig (ch. 3): Knuth’s Conjecture actions would result in all the gamblers and robbers crossing to the other side of the
”[S]tarting with the number 4, a sequence of square root, floor, and river?
factorial can reach any desired positive integer.”
Some integers are easy to reach: √4 = 2 ➢ Give the problem definition
Others are hard to reach:
√√√√(4!)!c = 5 ➢ How can the initial state, actions, and goal state be represented to the agent?
What is the problem definition for this example? ➢ Can you make the representation as short as
possible?
States: Positive real numbers
➢ Which actions are available in the initial state?
• Initial state: 4
• Actions: Apply square root, floor, or factorial operation (factorial for integers
only)
• Transition model: As given by the mathematical definitions of the operations. ➢ States: 0-3 gamblers and robbers on the initial side, the rest on the goal side,
subject to the robbers not outnumbering the gamblers on either side and the
• Goal state: The desired positive integer boat on either side.
• Action cost: Each action costs 1 ➢ Initial state: Three gamblers, three robbers, and a boat on the initial side
➢ Actions: All actions that move 1 or 2 people and the boat to another side
Jugs of Water - Problem ➢ Transition model: The people moved, and the boat was subtracted from one
side and moved to the other.
You have three jugs. One currently holds 8L of water. The other two are empty but
can hold 5L and 3L respectively. How can you use the three jugs to divide the water ➢ Goal state: Three gamblers, three robbers, and a boat on the goal side
such that you have two jugs that each hold exactly 4L? ➢ Action cost: Each action costs 1
➢ Give the problem definition. ➢ Initial state: Only need to keep track of numbers on the initial side: (3, 3, 1)
➢ How can the initial state, actions, and goal state be represented to the agent? ➢ Actions: (gamblers, robbers)
➢ Which actions are available in the initial state? ➢ Why not boat and direction?
▪ States: Any division of the water over the jugs, subject to the jugs’ capacity ➢ Goal state: (0, 0, 0)
▪ Initial state: 8L in the first jug and 0L in the other two ➢ Actions available in initial state: (0, 1) (0, 2) (1, 1)
▪ Actions: Pour water from one jug into the other until one is full, or one is empty ➢ Actions not available in initial state: (1, 0) (2, 0)
▪ Transition model: The amount of water poured is subtracted from one and
added to the other jug.
▪ Goal state: 4L in the first two jugs, 0L in the last
▪ Action cost: Each action costs 1

➢ Initial state: (8, 0, 0)

➢ Actions: ’Name’ the jugs 0, 1, and 2 (or anything else). Any action can be
represented as pouring from one jug to the other: (from, to)
➢ Goal state: (4, 4, 0)
 breakthroughs and paradigm shifts.
3. Exploratory Creativity: Exploratory creativity involves exploring and expanding the
boundaries of creative space. It encompasses the process of discovering novel ideas,
concepts, or possibilities that were previously unexplored. It is often characterized by
curiosity, experimentation, and the willingness to push creative boundaries.

Why is Maggie Boden considered to be one

of the architects of Cognitive science?

➢ Early in Boden’s career, AI was largely unheard of, and cognitive science
didn’t exist. There was very little appreciation for the role that computational
thinking could play in the psychological sciences.

Discussion 1: Margaret Boden ➢ Boden was one of the first philosophers to recognize how AI could impact
fundamental questions in philosophy, such as the nature of mind,
consciousness, learning, etc.
➢ Singularity - a hypothetical future point at which AI systems attain
superhuman intelligence and the ability to rapidly self-improve, potentially
surpassing human capabilities and leading to uncertain societal ➢ For cognitive science to exist, an integration of multiple disciplines/concerns
consequences. was required. Boden, as a philosopher, was well-placed to guide such an
integration. She set up the School of Cognitive Science at Sussex University,
which was one of the first.
➢ Generative systems - Generative systems, as defined by Margaret Boden,
are computational systems or algorithms that have the capability to generate
➢ In addition, she established creativity as a concern for AI, the “computational
creative and novel outputs, such as artistic works, music, literature, or other basis for new ideas”. She focused on creativity because she considered it one
forms of creative expression. These systems are designed to mimic some of the hardest problems.
aspects of human creativity and can produce content that is not simply a
recombination of existing data but rather exhibits an element of originality and
innovation.
What, according to Maggie Boden, are the limits of
➢ Machine translation/thesaurus - Machine translation refers to the automated AI and how do these limits relate to the concept of
process of translating text or speech from one language to another using computation?
computer algorithms and software.
➢ In theory, there are no limits to understanding the mind/brain as a
computational
➢ Machine thinking - refers to the capacity of computational systems, device, “to say otherwise is to say that the human mind works by magic, and it
particularly artificial intelligence (AI) systems, to perform cognitive tasks and doesn’t”. [What view on AI is being assumed here?]
problem-solving activities that exhibit characteristics of human thought ➢ In practice, however, it might not be possible to replicate human cognition.
processes. Machine thinking encompasses the ability of AI systems to
process information, reason, learn, and make decisions, often relying on ➢ Mind is the most complex system out there, and language is a particularly
algorithms, heuristics, and computational techniques to mimic aspects of problematic capacity to understand.
human cognition. ➢ We know little about the limits. We have very little idea about what
consciousness is.
➢ Information retrieval - Information retrieval generally refers to the process of ➢ Computers speaking with humans in a natural way is a “fantasy” and a long
accessing and extracting relevant information from large collections of data or way off.
documents, typically using computer systems and algorithms. ➢ Most importantly, how we think about computation will change, as it has
done already. The concepts we have now will not be enough to understand
the mind/brain.
➢ Combinatorial vs. transformational vs. exploratory creativity –
➢ Singularity: More than a few decades away, at the very least.
1. Combinatorial Creativity: involves recombining existing ideas, concepts, or
elements in novel and meaningful ways. It focuses on generating new creations by
assembling or rearranging existing components. This form of creativity is often
associated with lateral thinking and finding fresh combinations among known
elements.

2. Transformational Creativity: Transformational creativity goes beyond mere

recombination. It involves fundamentally altering or transforming existing concepts or
ideas to create something entirely new and innovative. This type of creativity
emphasizes the capacity to change the very nature of concepts, leading to
 Artificial intelligence as representation and search
Pretty much all problems in AI require:
1. Representing the problem. What does the agent need to “think” about the
problem?
2. Searching for the solution. How should the agent explore the space of
possibilities?

Solving problems in certain worlds.

The 15-puzzle and 8-puzzle

The game consists of numbered tiles. Tiles can be moved by shifting them into the
empty position. From the starting configuration, the task is to reach the goal state
where the tiles are in order shown here.

Lecture 3: Classical Search How many board configurations, or states, are there?
16! = 16 × 15 × . . . × 1 = 20, 922, 789, 888, 000
9! = 9 × 8 × . . . × 1 = 362, 880
Categories of problem: Certainty vs. Uncertainty

▪ Certain world: Rule-governed, deterministic, we know the state of nature, we

can represent it, we know what actions there are, and what effect these
actions have.
▪ Uncertain world: Stochastic, we don’t know the state of nature, and our actions
can have multiple consequences.
 What can we say about a solution
reached by breadth-first search?
How might an agent solve the 8-puzzle? ➢ It will always find the shortest
path to the solution.
There are four possible operations: Move the blank up, left, down, right. ➢ We know this because we
consider one level of the state
space at a time.

Which other strategy might we use to search for a

solution?

Option 1: Breadth-first search

Avoiding cycles in depth-first search

▪ We use an open and closed list, as before.

▪ But now we add children to the front of the open list.

➢ Open list: States we need to explore from.

➢ Closed list: States we’ve already explored from. First six steps:
1. Open = (1), closed = ()
Close 1, and add children (2,18,32) to open.
First six steps: 2. Open = (2,18,32), closed = (1)
1. Open = (1), closed = () Close 2, and add child 3 to open.
Close 1, and add children (2,3,4) to open. 3. Open = (3,18,32), closed = (1,2)
2. Open = (2,3,4), closed = (1) Close 3, and add children (4,8) to open.
Close 2, and add child 5 to open. 4. Open = (4,8,18,32), closed = (1,2,3)
3. Open = (3,4,5), closed = (1,2) Close 4, and add child 5 to open.
Close 3, and add children (6,7,8) to open.
5. Open = (5,8,18,32), closed = (1,2,3,4)
4. Open = (4,5,6,7,8), closed = (1,2,3) Close 5, and add children (6,7) to open.
Close 4, and add child 9 to open. 6. Open = (6,7,8,18,32), closed = (1,2,3,4,5)
5. Open = (5,6,7,8,9), closed = (1,2,3,4) . . .
Close 5, and add children (10,11) to open.
6. Open = (6,7,8,9,10,11), closed = (1,2,3,4,5)
Breadth-first ➢ The methods we’ve discussed so far will not work.

➢ Guaranteed to find the shortest path.

➢ Problematic when the branching factor b is high. ➢ The combinatorial explosion refers to the extreme growth in the
complexity of many problems.
➢ Requires exponential space O(bn) where n is the number of moves (or depth
in tree).

Depth-first How can we overcome the combinatorial

explosion?
Consider the 8-puzzle again. Not all states are equally appealing:

➢ Preferable when solution paths are long. ➢ The number of states out of place differs.
➢ Shouldn’t we prefer states “closer” to the solution? In short, use knowledge of
the task to improve search.
➢ In short, use knowledge of the task to improve search.
➢ Only requires linear space O(b · n).

➢ Might miss short path solutions if they exist.

Heuristic search:
➢ Typically many possible heuristics, and skill required to choose one.
➢ This heuristic doesn’t take into account how far tiles need to move
➢ .Heuristics are shortcuts that ignore information.

Summary
Fine-tuning the heuristic
If two states have the same number of tiles out of place, we should prefer the state
➢ How can we solve problems like the 8-puzzle? with the
shortest path from the start state.
1. Devise some way of representing the problem. 1. Let g (n) be the distance from state n to the start state.
This allows us to represent different states of problems we need to 2. Let h(n) be the number of tiles out of place in state n.
consider. The final heuristic,
2. Devise operators which move us from one state to another. f (n) = g (n) + h(n)
3. Devise a search strategy for applying the operators, which hopefully
takes us from the
initial state to the goal state.
Which search strategy we use will depend on the problem.

The problem: The combinatorial explosion

Recall: for the 15-puzzle and the 8-puzzle:

16! = 20, 922, 789, 888,000

9! = 362, 880

➢ This is nothing: The search space for chess is ≈ 10120

Summary: Dealing with state-space complexity
Problem-solving as search:
➢ Approximately the number of nanoseconds since the big bang. ➢ Define a state representation, define operators, and a search
 strategy. Opening book: 4000 positions.
Extended book: 700,000 games with grandmaster choices.
➢ 8-puzzle: Breadth-first search found the solution after examining
46 states. Endgame book: Games with 5 or fewer pieces.

➢ 8-puzzle: Depth-first search found the solution after examining 31

states. Practical 2 - Classical Search
➢ Combating the combinatorial explosion:
➢ Most search spaces are huge. Gamblers and Robbers
➢ Heuristics can guide the search and reduce the search space. Problem
➢ Best-first search with the f heuristic found the solution after On one side of the river, there are three gamblers and three robbers. They all want to
examining only 14 states. travel to the other side. However, there is only one boat that can hold up to two
people. If the robbers outnumber the gamblers, they will rob them. What sequence of
actions would result in all the gamblers and robbers crossing to the other side of the
river?
• State representation: (G,R,b)
• Initial state: (3,3,1)
• Goal state: (0,0,0)
• Action representation: (G,R)
How can we solve this puzzle using Depth-First Search?

Chess: Machine vs. Human

• “Solving chess” is a controversial and open question.
• Progress has been measured relative to human opponents.

IBM Deep Blue: Basic information

1. Hardware:
30-node IBM RS/6000 computer.
Each node has 16 chess chips, enabling parallel search.
Evaluate between 100 million and 200 million positions per second.
2. Evaluation function:
How do we assign a score to a board position? Maze
8000 different features are used. Problem representation
Both hand tuned and automatically discovered.
3. Game databases:
 • State representation: (x,y)
• Initial state: (0,0)
• Goal state: (3,3)
• Action representation: (up), (right), (down), (left)
How can we solve this puzzle using Breadth-First Search?

➢ To apply the Breadth-First Search algorithm, we use the open and closed
lists:
Step Open list Closed list
0 (0,0)
➢ If we expand a state, the generated states are added to the end of the open
list.
➢ The rest of the algorithm is the same.
➢ Our search can be visualised in a search tree, as well as in the maze.

➢ To apply the Best-First Search algorithm, we use the open and

closed lists
➢ In heuristic search algorithms, we represent a node as the
combination of the state
and its (heuristic) cost:
➢ Step Open list Closed list
0 (A,366)
➢ If we expand a node, the generated nodes are added to the
open list
➢ After adding the nodes, we sort the open list in increasing cost.
➢ The rest of the algorithm is the same.

➢ Expanding the first (and only) node finds Zerind, Sibiu, and Timisoara
➢ We can look up their cost in the table and add them to the open list:

➢ Apply the Best-First Search Algorithm until you reach the goal. That is, until
 you expand the Bucharest node.

àContinue the search to Bucharest using the A* algorithm

Discussion 2: Can machines think?

Turing, Alan (1950). Computing Machinery and
Intelligence. Mind, LIX (236), 433–460

Solipsism - a philosophical concept that suggests that one's own mind or

consciousness is the only thing that can be known to exist, and the
existence of the external world and other minds is uncertain or
unknowable. It is a form of extreme skepticism and philosophical
idealism.
Romania
A* Search

1. ESP, or extrasensory perception, is a phenomenon in which individuals

➢ To apply the A* Search algorithm, we change the way we calculate the cost perceive information or events through means beyond the conventional
of a node senses (sight, hearing, touch, taste, and smell). It includes telepathy,
➢ In Best-First Search, we just used the heuristic cost clairvoyance, and precognition. ESP suggests the ability to acquire
➢ In A* Search, we add the actual cost so far to the heuristic cost knowledge or awareness without reliance on the known sensory organs.

➢ The cost of the initial node does not change, as the actual cost so far is 0:
2. Clairvoyance refers to the alleged ability to gain information about a
person, object, event, or location through extrasensory means. It is often
associated with the ability to "see" or perceive things that are distant in
space or time without using the physical senses. Clairvoyance implies an
intuitive or psychic awareness.

àExpanding Arad again finds Zerind, Sibiu, and Timisoara 3. Psychokinesis, often abbreviated as PK, is a purported psychic ability
àThe costs of these nodes are (S,140+253=393), (T,118+329=447), that involves influencing or manipulating physical objects or events using
(Z,75+374=449) only the power of the mind, without any physical interaction. It suggests
the capacity to move or affect objects through mental concentration or
willpower.
 not: “not of theoretical importance”. Computing is all that counts.
4. Telepathy is the claimed ability to transmit thoughts, feelings, or ➢ One can’t distinguish between different kinds of computers using
information directly from one person's mind to another's, without the use this test.
of verbal or written communication or any known sensory channels. It
implies a form of direct mind-to-mind communication.

Universal machine - refers to a computational model that can Objection 1: The theological objection
simulate the behavior of any other Turing machine. It's a concept
related to theoretical computer science and the idea that there can Thinking requires a soul, and only god can provide one. Animals and
be a single, versatile machine capable of emulating the machines don’t have a
functionality of any other machine. soul, so machines can’t think in this sense.
➢ This issue is a question of faith, which we can’t resolve
Discrete state machine - a mathematical model that consists of a scientifically.
finite number of states, transitions between those states, and an ➢ Not all people of faith will agree with this argument anyway.
initial state. Discrete state machines are used to represent and
analyze systems or processes with distinct, well-defined states and
➢ Issues like these are why Turing has replaced the question!
transitions. ➢ !!! He doesn’t accept this fact, he thinks that God decides which ones
have a soul and for God animals have souls. He can also improve the
thinking of a certain animal

Objection 2: The “heads in the sand” objection

Proposal Let’s not address the question because it’s implications are too
dangerous to contemplate.
Main question: What is Turing’s proposal? • It’s too late!
• Turing wants to address the question “can machines think?”. • Might the science of thinking machines help us to understand
• The problem is that the terms “machine” and “think” are too ourselves
vague. • Could thinking machines have a positive impact on humanity?
• As it stands, the question is “too meaningless to deserve • !!! High intellectual people tend to think that by making machines think
discussion”. would take people’s superiority and will make them less valuable. This
• The proposal is to replace the question. can also mean for men that they might lose general position to the
• He proposes the imitation game as a means to judge machines.
intellectual capacity.
• Pose the question: Could a machine pass this “Turing test”? Objection 3: The mathematical objection
• Turing predicts: By 2000 machines will pass with 30%
accuracy over 5mins. • It has been proven that some problems are uncomputable. This
• The meanings of “machine” and “think” will change, and we means that computers will be unable to answer some questions,
will likely accept that and this places a question mark over the question of whether they
machines can think. can think.
• Humans make errors too. In fact, to pass the test, shouldn’t the
Why adopt Turing’s notion of a “machine”? machine sometimes make errors.
• !!! The short answer to this argument is that although it is established that
What does Turing mean by “machine”, and why does he propose this there are limitations to the powers of any particular machine, it has only
specific kind of machine? been stated, without any sort of proof, that no such limitations apply to
the human intellect.
➢ He talks about a digital computer.
➢ Digital computers, with a von Neumann architecture: Store Objection 4: The argument from consciousness
(addressable memory), executive unit (that carries out operations),
and control (handles loops etc). Also, randomness. • No machine can feel it. To decide if a machine (or human) can feel
➢ Functionalism: Turing doesn’t care if the machine is electrical or requires us to be that machine (or human). This is solipsism.
 • The question of machine consciousness is an important one, but
how can we resolve it? Objection 8: Argument from the informality of
• Again, this is why Turing proposes the test. He believes questions behavior
about thinking can be approached without considering
consciousness.
We can never formulate all the rules needed to mimic human behaviour.
• !!! No one can really feel what the other person or the machine feels when
performing certain task. So in order to understand the feeling, you have to
be that particular person or machine. ➢ Some rules are general, and apply in an infinite set of
circumstances.

➢ !!! Humans don’t have such rules that they regulate their whole life, that’s why they
can’t be compared to machines. (Rules of conduct)
Objection 5: The argument from various disabilities
Machines will never be able to do “X”. E.g., have a sense of humour.
➢ Why not? What we refer to as “machines” are changing. Objection 9: Argument from extra-sensory
➢ If we can describe “X”, then why can’t a machine do it? perception (!)

➢ !!! Humans draw conclusions based on the experience they have with Phenomena like telepathy and clairvoyance are real, and tell us that the
machines: they are ugly, they are designed for limited purposes and can’t human mind doesn’t work according to the laws of physics. Could the
deal with other things that are outside of their requirements. interrogator in the Turing test use it to distinguish human from machine?
➢ Phenomena like telepathy and clairvoyance are not real. There is
no evidence.
Objection 6: Lady Lovelace’s objection
Learning machines
Machines can only ever do what we program them to do. They cannot
What is Turing telling us in this section of the article?
think for themselves.
➢ Turing’s argument from complexity: Machines often do the 1. He’s drawing us away from viewing “machines” as simple sets of
unexpected. rules, and toward
➢ Machines are unpredictable, and can beat their designers at viewing them as devices that can learn by themselves like humans.
various games. 2. Could we program a “child” machine, and let it learn for itself, like a
human?
3. After all, humans are not created in one shot, we are the product of
years of experience
and learning.
Objection 7: Argument from the continuity of the 4. If we equip a machine with the ability to think logically, perhaps it
could work things out
nervous system for itself.

Computers are discrete state machines, whereas the nervous system is

continuous. Computers therefore can’t mimic the mind/brain.
➢ Digital, discrete state machines can simulate continuous machines
to an arbitrary level of precision.
➢ And anyway, how could we tell between these two types of
machines in the imitation game?

➢ !!! It is true that a discrete-state machine must be different from a

continuous machine. But if we adhere to the conditions of the imitation
game, the interrogator will not be able to take any advantage of this
difference. The situation can be made clearer
if we consider some other simpler continuous machine. A differential
analyser will do very well.