Computational and Artificial Intelligence Flashcards
What is AI?
the synthesis and analysis of computational agents that act intelligently
What are the definitions of AI (Russell and Norvig, 2016)?
- systems that think like humans
- systems that think rationally
- systems that act like humans
- systems that act rationally
Explain the “cognitive modelling” approach
definition of AI: systems that think like humans
- requires greater understanding of human cognition
- need to know how humans think (through introspection, psychological experiments, and brain imaging) in order to model human mechanisms of “intelligent thought” in a machine
Explain the “Laws of Thought” approach
definition of AI: systems that think rationally
- if you can take any informal knowledge and turn it into formal notation, you should be able to solve any logical problem in logical notation
Explain the “Turing Test” approach
definition of AI: systems that act like humans
- can a machine convince a human that it is a human just by its actions?
Explain the “Rational Agent” approach
definition of AI: systems that act rationally
- if there is not a logically provably correct choice, correct logical inference will not work so machines need to select an action based on the best possible outcome
Explain thinking vs acting in AI systems
acting only gives the impression of thinking intelligently, whereas thinking requires a conscious choice and processing themselves
does it matter if the outcome is intelligent or just the actions? (e.g. self-driving car: is it enough to just notice pedestrians and not hit them or does it need to make a conscious effort to?)
Explain rational vs human in AI systems
AI decisions may be rational but not necessarily human-like (e.g. AlphaGo: rational moves for victory that humans would not choose)
What is strong AI?
a machine that actually thinks
What is weak AI?
a machine acting intelligently without thinking
What is computational intelligence?
alternative approach to AI that is adaptive, and changes on its own depending on its environment
finding best option in uncertain situations
Goals of AI/CI
- creating AI
- understanding natural intelligence
- creating intelligent solutions to problems
What is artificial general intelligence?
the ultimate goal of AI – machines that can perform any function, it does not need to know how to do a task specifically but can learn how to (intelligent, whilst also adaptive and flexible)
What are the 6 tests for intelligence?
- Turing Test
- Coffee Test
- Robot College Student Test
- Employment Test
- Flat Pack Furniture Test
- Mirror Test
Turing Test
convince a human that you are a human not a machine
standard for AI keeps changing, e.g. modern chatbots no longer considered strong AI
Coffee Test
enter a random home and use the facilities to make a coffee
intelligent traits: learning from experience, physical intelligence, flexibility/adaptability
Robot College Student Test
enrol at a university and obtain a degree
(not about the tasks, but having the flexibility to deal with changing situations; also requires awareness of circumstance)
Employment Test
perform as a well as a human in a job
general purpose learning systems; advocates for developing childlike machines that can learn and develop
Flat Pack Furniture Test
unpack and assemble an item of flat packed furniture
problem solving and flexibility
Mirror Test
machine should distinguish an object and its reflected image from a mirror
(originally designed for animals; testing self-recognition and self-awareness)
About AI tests
- debate of thinking vs acting
- prioritises features of: flexibility, adaptability, self-awareness, learning, development
- measuring intelligence through human features?
Argument against weak AI
can machines act intelligently?
Dijkstra
does the question of whether machines can actually think matter when its actions are still successful?
What is the Qualification Problem?
- human behaviour is too complex to capture as formal logic
- importance of environmentally and socially situated agents exhibiting embodied cognition, over disembodied programs
Argument against strong AI
can machines actually think?
Turing Test might only…
- passing Turing Test may only demonstrate simulation of thought, not actual thought
- argument from consciousness involves phenomenology (study of direct experience – “feel” emotions) and intentionality (beliefs and desires)
Searle’s Chinese Room experiment
non-Chinese speaking person convincing you they understand Chinese by looking it up from a book; they don’t actually “understand” Chinese; without understanding/intentionality, the machine isn’t thinking and doesn’t have a mind so strong AI hypothesis is false
Mind-Body Replacement
how can physical states also be mental states?
Brain replacement experiment
replacement of microchips instead of neurons, when is it no longer a brain and a machine? (i.e. Ship of Theseus)
What are some human biases in AI?
- age
- race
- gender
- status
- religion