AI Intro & Bakgrund Flashcards
What does solving a problem mean in the context of AI?
Finding actions that effectively solve the problem, which may involve identifying the problem, having knowledge of standard problems and solutions, and applying the appropriate solution.
Can AI replace humans in problem solving?
Yes, if the process can be described and the related knowledge, such as standard problems and solutions, is available.
Is a washing machine considered an AI system or intelligent system?
No, a typical washing machine does not possess the ability to learn, reason, or adapt to new situations autonomously.
En tvättmaskin anses inte vara ett AI-system, men den kan betraktas som ett intelligent system beroende på dess funktioner.
What criteria can be used to judge whether a machine is intelligent?
Problem-solving ability
Learning capability
Adaptability
Creativity
Natural language understanding and generation
Perception and understanding
Autonomy
Ethical and moral reasoning
What are the four views of AI?
Thinking humanly
Thinking rationally
Acting humanly
Acting rationally
What is “Thinking humanly” in AI?
Cognitive modeling, which involves applying the scientific method to study human cognition and developing scientific theories of internal brain activities.
What are the challenges in “Thinking rationally” or “laws of thought”?
In other words “right thinking” give problems:
Uncertainty: Not all facts are certain.
Resource limitations: Solving a problem in principle vs. in practice under constraints like time and computation.
What does “Acting rationally” mean in AI?
It involves doing the right thing to maximize goal achievement based on the available information. This can include rational behavior without necessarily involving thinking.
What does “Acting humanly” mean in AI?
The ability of artificial intelligence systems to mimic human behavior or actions. A machine can show behavior as a human.
What is the Turing Test?
A method to determine whether a machine can demonstrate human intelligence by engaging in a conversation with a human without being detected as a machine.
Who pioneered the first AI program and what did it include?
Alan Turing, including the notion of computation (Turing machine), the programmable computer (universal Turing machine), the first chess program, and the Turing test.
What are the steps involved in the Turing Test procedure?
Setup: Three participants (interrogator, human, machine) with isolated communication.
Objective: Machine convinces the interrogator it is human.
Procedure: Interrogator engages in conversations with both participants.
Testing: Machine’s success is based on generating human-like responses.
Passing the Test: If the machine consistently convinces the interrogator, it passes the test.
Does passing the Turing Test mean a machine is truly intelligent?
No, it demonstrates the machine’s ability to simulate human-like conversational behavior effectively, but it doesn’t necessarily mean the machine possesses true intelligence or consciousness.
What tasks must be solved to create an AI system? (Replace a “profession”
- Decision making, problem solving, language handling, learning, image
processing, predicting, and cognition. - Define structure of expertise and expert knowledge
- How an intelligent agent (a doctor/lawyer/etc) works, and acts with others
(multi-agent system) - How is autonomy? That the agent works on his own.
- How does the problem (customer/situation/artifact/…) perceive? (Perception)
- How does thinking work when solving the problem (Cognition)
- How one learns (Machine learning)
- How to communicate in speech (natural language processing)
What are the different types of Artificial Intelligence?
Weak AI
General AI
Super AI
What is Weak AI?
AI systems that behave like humans but do not provide insights into how the brain works, such as IBM’s Deep Blue chess play. Single task-based Algorithms, Dedicated for one task, Driven by industry, Practical
What is General AI?
AI that simulates the human brain, performing tasks as well as or better than humans and providing insights into brain functions.
What is Super AI?
AI that exceeds human intelligence, with capabilities for judgement, learning, planning, reasoning, and communication, which remains a hypothetical concept.
What are the key achievements of AI?
Facilitating and replacing human decision-making
Robots
Automatic process control
Limited spoken language understanding
Text, human, object, and emotion recognition
Smarter search engines
Observing and understanding human emotions
Solving mathematical problems
How does AI research and engineering work?
AI research and engineering involve:
Problem Definition
Data Collection and Preparation
Algorithm Selection and Development
Model Training
Evaluation and Validation
Iterative Improvement
Deployment and Integration
Monitoring and Maintenance
What disciplines are involved in AI research and engineering?
Computer science, mathematics, cognitive psychology, neuroscience, and more.
What are the key historical milestones in AI development?
1943: McCulloch & Pitts: Boolean circuit model of brain
1950: Turing’s “Computing Machinery and Intelligence”
1956: Dartmouth meeting: “Artificial Intelligence” adopted
1950s: Early AI programs, including Samuel’s checkers program and Newell & Simon’s Logic Theorist
1965: Robinson’s complete algorithm for logical reasoning
1969-79: Early development of knowledge-based systems
1980: AI becomes an industry
1986: Neural networks return to popularity
1987: AI becomes a science
1995: The emergence of intelligent agents
What does AI do according to most AI research?
Select a specific problem to solve: study the problem, represent necessary knowledge, acquire and codify that knowledge, and build a problem-solving system.
Select a category of problem or cognitive activity: theorize a solution, build experimental systems, and modify as needed.
What are intelligent agents and multi-agent systems in AI?
Intelligent agents are systems that perceive and act in an environment, and multi-agent systems involve multiple interacting intelligent agents.
What are AI researchers focusing on instead of traditional “AI”?
Intelligent agents and multi-agent systems
Ontologies
Machine learning and data mining
Adaptive and perceptual systems
Robotics and path planning
Search engines, filtering, and recommendation systems
When we say that AI researchers are focusing on specific areas and applications rather than “traditional AI,” we mean that they are working on practical, specialized projects instead of trying to create a general, all-purpose artificial intelligence that can do anything a human can do.
What are some approaches used in AI research?
Knowledge-based
Ontologies
Probabilistic (Bayesian Nets)
Neural Networks
Fuzzy Logic
Genetic Algorithms
What are some areas of current AI research interest?
Natural Language Understanding/Information Retrieval
Speech Recognition
Planning/Design
Diagnosis/Interpretation
Sensor Interpretation
Perception
Visual Understanding
Robotics
What are the key milestones mentioned in the AI timeline?
Teoretiska genombrott (30-40 talet)
Neural födelse (40-60 talet)
Klassisk era (50-60 talet)
Första vintern (70-talet)
En AI-industri föds och neural återkomst (80 talet)
Andra vintern (~1990-2010)
Nu (2010 -)
Vad hände under teoretiska genombrott (30-40 talet)
Alan Turing (1912-1954)
Turing-maskin (1936)
Turing-test (1950)
What is “Beräkningsbarhet” (Computability)?
Definition: A process where we proceed from initially given objects (inputs) according to a fixed set of rules (algorithm) to arrive at a final result (output).
Key Concepts:
An operation is computable if it can be performed by a finite, mechanical procedure.
Algorithm: A mechanical step-by-step instruction on how to achieve a certain goal from a given starting position.
Example: Python function determining age eligibility.
What are decision problems (“Beslutsproblem”)?
Problems that can be posed as yes/no questions.
Example questions:
Given two numbers, x and y, is x evenly divisible by y?
Is a given number x a prime number?
In logical terms: Is there a mechanical way to determine if a certain conclusion follows from given premises?
What were the challenges identified with decision problems (beräkningsproblem) in the 1920s?
Some decision problems were suspected to be unsolvable.
Issues:
A proposed algorithm for solving a certain problem is easy to verify.
Proving that no algorithm can solve it is infinitely harder.
This requires a general theory of what algorithms can achieve in general.
This search for a general theory of what could be solved with algorithms interested many logicians.
What did Turing’s approach to decision problems help illustrate?
The notion that not all problems can be solved by algorithms.
The concept of computational limits and the fundamental principles of computability.
