Phil of AI

What is computability?

Problems that can be solved by running them on a computer (p17)

What is AGI (artificial general intelligence)?

Systems that exhibit more human-like general intelligence (p30)

What is 'soft computing'?

'A motley assemblage of computational techniques designed to deal with imprecision, uncertainty, approximation, partial truths, and so on' (p 21)

What is an autonomous agent?

'A system situated in an environment that senses the environment and acts on it in pursuit of its own agenda' (p27)

What is an AI planner?

'A system that automatically devises a sequence of actions leading from an initial... state to a goal state' (p25)

What is strong AI?

'AI is the field devoted to building persons, period' We want to build actual thinking machines (p35)

What is machine learning?

'Algorithms that allow AI systems to learn' (p26)

What is weak AI?

'Engineering' AI devoted to passing various tests Contrast this against strong, which aims to build an artificial mind (p35)

What is functionalism?

'The internal functional organization of the system must be taken into account' According to functionalism, mental states are identified by what they do rather than by what they are made of (p35, Google: https://www.iep.utm.edu/functism/)

What is functionalism?

'The internal functional organization of the system must be taken into account' & According to functionalism, mental states are identified by what they do rather than by what they are made of & Objectors to functionalism generally charge that it classifies too many things as having mental states (think Chinese state/brain argument) (p35)

What is formalism?

'The notion that certain patterns of thought are valid by virtue of their syntactic form, independently of content' (p36)

What are the core aspects of each of the three main frameworks? Computational, connectionist, situated-emobied-dynamic (SED)

1) Computational: Emphasizes structure and content of representations, algorithms which manipulate these 2) Connectionist: Emphasizes network architecture, learning algorithm, training protocol, distributed representations 3) SED: internal and external forces in the brain-body environment. How these forces shape the trajectory/behaviour of an agent into a stable equilibrium with the environment. (p140)

What are the three core threads of Dynamics in SED?

1) Dynamical Systems Theory (DST): mathematical theory that can be applied to any system that changes over time. 2) Dynamical Framework: collection of concepts, intuitions, and metaphors which give a dynamical perspective of a system 3) Dynamical Hypothesis: how DST and framework can be combined as a counter to traditional computational hypothesis and cognitive science. (p134)

What is the frame problem?

1) Knowing which aspects of a situation are changed by a particular action 2) The problem of reasoning with incomplete knowledge (p93)

What are the 3 core aspects of embodiment?

1) Physical: an agent's physical body plays a role in its behaviour. Provide a capability for action 2) Biological: Specific biological facts about the agent must be taken into account. Neuroscience, evolution, etc... 3) Conceptual: Even when engaged in pure rational thinking our most abstract concepts are still ultimately grounded in our bodily experience and bodily-oriented metaphors (p132)

What is backpropagation?

A Neural Net is assigned random weights. Then trained and the weights are adjusted based on how close they represent the correct answer

PSSH (physical symbol system hypothesis)

A PSS is a machine that produces, through time, an evolving collection of symbol tokens which are related physically (i.e. one symbol next to another). These tokens can be created, modified, reproduce and be destroyed. & The PSSH is a theory that a physical symbol system has the necessary and sufficient means for general intelligent action. & Critiques - Can be thought to lack "symbol grounding" (how do symbols get their meanings?) - "computation as it is currently understood does not provide an appropriate model for intelligence

What is a semantic net?

A concept (node) gains meaning via its relationships (links) to other concepts (nodes) (p24)

What is GOFAI?

A label used to denote classical, symbolic, AI. While "AI" by itself refers to everything, like connectionism, evolutionary programming, evolutionary robotics. Goal is to build useful computer systems by doing or assisting with tasks humans want done. "GOFAI computation involves the construction and transformation of symbolic data structures." (p89)

What is a hueristic?

A rule of thumb that is not guaranteed to find the best solution but often finds a good enough one (p25)

What is a Mental Process?

A sequence of tokenings of mental representations which express the propositional content of corresponding thoughts

Holistic Context & Relevance

Ability to distinguish between essential and inessential Considered the major stumbling block of all AI and computational cognitive science (p47)

What is connectionism?

An AI approach to cognition in which multiple connections between nodes (equivalent to brain cells) form a massive interactive network in which many processes take place simultaneously and certain processes, operating in parallel, are grouped together in hierarchies that bring about results such as thought or action. & advantage include: massive parallelism, learning capabilities, fault tolerance. (from Google)

For situated, embodied, dynamic systems (SED); what is the notion of an extended mind?

An agent's environment plays an essential role in its behaviour. The agent can manipulate the environment by organizing it in a manner which helps it achieve its goal. In this manner the environment can act as an "extension" of the mind. (p131)

What was the perceptron?

An early neural net which, '... maps its input vector into a weighted sum subject to a threshold, yielding a yes or no answer.' (p19)

Proof Checking

Given a deduction which is said to derive a conclusion from some premises, decide whether the deduction is true. (p37)

What is the central question of CTM?

CTM supposes an inventory of 'primitives' which can be used as building blocks. How do these primitives inside our brain manage to be about objects and states outside our brain?

Block thought experiment?

Citizens of China are all arranged like neurons for a human body. They have radios to talk to each other. Can receive and send signals to simulate physical behaviour (raise an arm) & machine functionalism - 'consciousness can be determined by inputs, outputs, and causal mental states' (From Google) & According to 'machine functionalism' this China Brain would constitute a conscious mind since all inputs and outputs are properly related & Experiment show that 'functionalism' is too liberal. i.e. there is more to the mind than input, output and causality

What is Dynamical Systems Theory?

Cognition is dependent on continuous interaction of an agent with its surroundings (p80)

Cognitive Robotics & Developmental Robotics

Cognitive Robotics - The endowing of robots with cognitive capabilities Developmental robotics - Enable robots to learn continually (p29)

Advantages of symbolic model vs connectionist models?

Connectionist models include massive parallelism, learning capabilities, and fault tolerance. Symbolic models allow for crisp/explicit representation, easy of specifying symbolic processing steps and a resulting precision in processing. (p120)

What is Proof Checking?

Given a deduction which is said to derive a conclusion from some premises, decide whether the deduction is true. (p37)

Conjecture Generation

Presented with a body of information we can come up with interesting conjectures (opinions, conclusions). Then try and prove these conjectures (p37)

Informational Theories

Idea of 'covariance'. If x covaries (changes) systematically with y, then x carries information about y. 'A car's speedometer covaries systematically with the car's speed and thus carries information about it.' 'smoke means fire' so smoke has info about fire (p44)

Narrow AI vs Human-Level Intelligence

In the early days of AI many researchers aimed at creating human-level intelligence in their machines, the so-called "strong AI." Later, as the extraordinary difficulty of such an endeavor became more evident, almost all AI researchers built systems that operated intelligently within some relatively narrow domain such as chess or medicine. Only recently has there been a move back in the direction of systems capable of a more general, human-level intelligence that could be applied broadly across diverse domains. (from Google)

Evolutionary Theories

Intentional states are adaptations, the same way livers and thumbs are. The meaning of an intentional state is its purpose. Desire - "I want to eat those bananas" Purpose of desire - Obtain food/survive (p45)

What was Blockhead's way to cheat the Turing Test?

Make a lookup table of all plausible questions and answers

Dreyfus Critique of Strong AI?

Our ability to understand the world and other people is a non-declarative type of know-how skill. It is inarticulate, preconceptual and has a dimension that cannot be captured by a rule-based system. & emphasis on imagination, ambiguity tolerance, use of metaphor, fringe conciousness, gestalt perception. These are resistant to computational treatment (p47)

What is Conjecture Generation?

Presented with a body of information we can come up with interesting conjectures (opinions, conclusions). Then try and prove these conjectures (p37)

What is psychological reality or 'explicit vs implicit'?

Psychological reality is the fact that a set of rules might adequately describe a cognitive phenomenon. The rules might fit all available data. But it does not mean there is something in our heads with these rules encoded & Implicit rules describe behavioural regularities Explicit rules have encoded representations which produce behaviour.

Reasoning vs Knowledge

Reasoning - Think through everything well from solid premises and make a decision based on these Knowledge - Having a large knowledge base to work with for finding patterns and making inferences (Think expert systems & Data Mining) (class notes)

What is an expert system?

Reasoning is not all there is to intelligence. Real world problems demand the solver know something. 'A piece of software which uses databases of expert knowledge to offer advice or make decisions in such areas as medical diagnosis' (p20 | From Google)

Searle Chinese Experiment?

Searle in a room alone and does not know Chinese, outside are all Chinese speakers. Chinese speakers send questions inside, in Chinese. Searle consults a rule book with a lookup table that shows which Chinese answers to output based on the input. To Searle everything is just 'squiggle-squoggles' & Argument is that Searle represents everything a computer is. Searle is mindlessly moving around squiggles and, according to his argument, this is all a computer does.

Smart Software vs Cognitive Modeling

Smart Software - AI for business/engineering applications Cognitive Modeling - Using AI to have insight into the human mind (class notes)

Cognitive Revolution

Started in 1950's, overthrew behaviorism

What was the Cognitive Revolution?

Started in 1950's, overthrew behaviorism & 5 key ideas 1)"The mental world can be grounded in the physical world by the concepts of information, computation, and feedback." 2) "The mind cannot be a blank slate because blank slates don't do anything." 3) "An infinite range of behavior can be generated by finite combinatorial programs in the mind." 4) "Universal mental mechanisms can underlie superficial variation across cultures." 5) "The mind is a complex system composed of many interacting parts." & behaviourism - assumes that all behaviors are either reflexes produced by a response to certain stimuli in the environment, or a consequence of that individual's history

To Represent or Not

Such knowledge had to be represented somehow within the system, that is,the system had to somehow model its world. Such representation could take various forms, including rules. Later, a controversy arose as to how much of such modeling actually needed to be done. Some claimed that much could be accomplished without such internal modeling. (from Google) Think explicit symbolic representation for path-finding (wall here, split there) vs a dynamical systems robot which can follow a path by reacting to the environment according to basic rules (turn if you hit something, walk forward otherwise).

Symbolic AI vs Neural Nets (sub-symbolic)

Symbolic AI took the view that intelligence could be achieved by manipulating symbols within the computer according to rules. Neural nets, or connectionism as the cognitive scientists called it, instead attempted to create intelligent systems as networks of nodes each comprising a simplified model of a neuron. Basically, the difference was between a computer analogy and a brain analogy, between implementing AI systems as traditional computer programs and modeling them after nervous systems. (from Google)

Symbolic AI vs Neural Nets (sub-symbolic)

Symbolic AI took the view that intelligence could be achieved by manipulating symbols within the computer according to rules. Neural nets, or connectionism as the cognitive scientists called it, instead attempted to create intelligent systems as networks of nodes each comprising a simplified model of a neuron. Basically, the difference was between a computer analogy and a brain analogy, between implementing AI systems as traditional computer programs and modeling them after nervous systems. (from Google) & Symbolic - looks at symbols being explicitly represented Neural Nets - represent concepts distributed over a network as an activation pattern

Searle Chinese Experiment arguments?

Systems argument - Searle is a piece of a larger system and the system understands Chinese Robot argument - Although Searle is operating as a piece of a 'computer' he is not actually interacting with the world. i.e. he is not a 'strong AI'

What is the Total Turing Test?

TTT-passers must be robots able to operate in the world with behaviours in a way indistinguishable from humans (p35)

Brain in a vat vs Embodied AI

The early AI systems had humans entering input into the systems and acting on the output of the systems. Like a "brain in a vat" these systems could neither sense the world nor act on it. Later, AI researchers created embodied, or situated, AI systems that directly sensed their worlds and also acted on them directly. Real world robots are examples of embodied AI systems. (from Google)

CTM (computational theory of mind)

The essence of the mind is not to be found in the actual biology of the brain; but in the role that your mental state plays in your life. Especially with regard to stimuli (input) causing behaviour (output). & To have a belief is to have a mental representation of that belief. & Complex thought representations are built up recursively from simpler ones. & CTM holds that the mind is a computational system that is realized by neural activity in the brain. Computation can realized by neural networks or silicon chips so long as there is a series of outputs based on manipulations of inputs and internal states according to rules. (p44, Wikipedia)

What is AI?

The field devoted to building 'artifacts' capable of displaying behaviours we consider to be intelligent, in controlled and well-understood environments. (p34)

Conceptual-Role Theories

The meaning of a mentalese symbol is fixed by the role it plays in one's cognitive life and particularly by the relations that it bears to other symbols, perception, and action. Basically symbols in our head get their meaning from relations to other symbols in our head (45)

For situated, embodied, dynamic systems (SED); what is an affordance?

The possibility for action that an environment presents to an agent. Their significance is relative to the capacity of the agent. (p130)

What is the Frame Problem?

The problem of bringing to bear what is relevant to current tasks (p75)

What is the symbol grounding problem?

The problem of how words, and ultimately mental states, get their meaning (p134)