Block 1 (1/2) - AI and Intelligent Agents

What are seven risks of AI in the real world

1. Autonomous weapons

2. Privacy weapons

3. Biases in decision-making

4. Employment of humans

5. Safety in safety-critical applications

6. Cybersecurity threats

7. Against humanity threats

What are four benefits of AI

1. Address global challenges: i.e. disease cures, climate change and resource shortages

2. Decrease repetitive work

3. Increase production of goods and services

4. Accelerate scientific research

What is Artificial Intelligence?

Software that imitates human capabilities.

1. Making decisions based on data and past experiences.

2. Recognising abnormal events.

3. Interpreting visual input.

4. Understanding written and spoken language.

5. Engaging in dialogues and conversations.

What are the four approaches to define AI?

1. Systems that think like humans. e.g. cognitive computing.

2. Systems that act like humans. e.g. Robotics.

3. Systems that think rationally. e.g. Knowledge based systems.

4. Systems that act rationally. e.g. Performance optimisation.

What is a system that can act humanly

A system that could pass off for a human to another human.

What is the Turing test

A famous test being the Turing test to see if a person could tell if they were talking to a human or a robot.

What is the problem with the Turing test

Turing test is not reproducible, constructive, or amenable to mathematical analysis.

What is a system that can think like a human?

A system that can mimic the human cognition artificially, (cognitive computing) using theories of the internal activities of the brain such as; Cognitive science which is predicting and testing behaviour of the human subjects (top down), Cognitive Neuroscience which is direct identification from neurological data (bottom-up).

What is Cognitive Science

Predicting and testing behaviour of human subjects (top-down).

What is Cognitive Neuroscience

Direct identification from neurological data (bottom-up). DATA FOCUSED

What is a system that can think rationally?

A system that has direct logical behaviour based on principles and rules.

They usually use knowledge-based systems, expert systems and decision-support systems. It can also bring mathematical and philosophical principles into AI.

What is the problem with systems that think rationally to solve problems

Not all intelligent behaviour relies on logical deliberation.

What is a system that can act rationally?

A system that maximizes goal achievement, by doing the right thing irrespective of simulated human thought processes.

They choose the action that maximises the expected value of performance given the measure and the percept sequence (series of events) thus far.

What is PEAS

1. Performance measure - What are the main priorities of performance?

2. Environment - Where will the system be used?

3. Actuators - What are the ways for an agent to output data and take action?

4. Sensors - What are the items used to scan surroundings and observe input data?

Give examples of PEAS for an automated taxi.

To design a rational agent for an automated taxi:

1. Performance: safety, destination, profits, legality, comfort.

2. Environment: streets/freeways, traffic, pedestrians, weather.

3. Actuators: steering, accelerator, brake, horn, speaker/display.

4. Sensors: video, accelerometers, gauges, engine sensors, keyboard, GPS.

Why is it important to know the environment types for a rational agent?

Understanding the environment type largely determines the agent design

What is the difference between fully observable and partially observable environments?

An environment is fully observable if the agent’s sensors give access to complete state of the environment at any given time.

In a partially observable environment, the agent does not have complete information about the current state of the environment.

What is the difference between single-agent and multi-agent environments?

A single agent environment contains only one agent. All other objects in the environment are not described as agents, but observe their effect on the performance of the main agent.

In a multi-agent environment, there are multiple autonomous entities or agents interacting and possibly competing or cooperating with each other.

What is the difference between Deterministic and Stochastic (Non-Deterministic) environments?

Deterministic environments have no uncertainty (can be determined every time), as the next state of the environment is only determined by:

1. The current state.

2. The action taken by the agent.

Stochastic environments have added uncertainty due to unknown factors or random elements of the environment, so the same outcome may not always happen with the same states.

What is the difference between Episodic and Sequential environments?

Episodic environments divide the agent’s experience into distinct episodes of interaction with a given percept (perceptual input) and action. The next episode does not depend on the actions in previous episodes.

Sequential environments mean that the agent's actions have consequences over time and the agent's decisions may affect future states and outcomes.

What is the difference between Static vs Dynamic environments?

An environment is static if it does not change while the agent is deliberating.

If it does change it is called dynamic.

What is the difference between Discrete and Continuous environments?

An environment is called discrete if:

1. There is a discrete number of states.

2. Time is measured in discrete steps.

3. There is a discrete set of possible perceptions and actions.

An environment is called continuous if:

1. There is a continuous number of states.

2. Time is measured continuously.

3. There is a continuous range of possible perceptions and actions.

What are the four types of AI agent and what can they all be turned into?

1. Simple reflex agents

2. Model-based reflex agents

3. Goal-based agents

4. Utility-based agents

They can all be turned into learning agents.

What are simple reflex agents

An agent that selects actions based on the current percept only (what the world is like right now).

E.g. auto braking in an autonomous car

What are model-based reflex agents?

An agent that maintains an internal state model based on percept history (full sequence of previous events).

e.g. autonomous cars keeping track of other cars around it to avoid collision.

What is a goal-based learning agent?

An agent that has an internal state model of percept history with a set goal to target.

e.g. the autonomous car that can reach a given destination

What is a utility-based agent?

A model-based agent that has a goal and also considers the best ways to achieve the goal considering performance.

e.g. autonomous car that is able to use the best path to the destination

What is a learning agent?

Any agent that can learn and modify itself as it goes.

e.g. autonomous car that learns how to adjust the braking pressure in different weather conditions.