Figure: pixabay
Original source: medium
Author: Prashant Guptah
"Robot Circle" Compilation: Well~ Astro Boy, Dora Aliang
Based on the goals we want to achieve and the criteria for measuring their success, we can use a variety of methods to create artificial intelligence. It covers an extremely wide range of things, from very complex systems such as autonomous driving and robots to common parts of our daily lives, such as face recognition, machine translation and email classification, which can be classified as the field of artificial intelligence.
After reading the following article, you may understand what is needed to truly create artificial intelligence.
"Do you really know what artificial intelligence is"
(https://hackernoon.com/so-you-think-you-know-what-is-artificial-intelligence-6928db640c42)
The path you take will depend on what your AI is aiming for and how complex and feasible you understand the various methods. In this article, we will discuss those approaches that are considered more feasible and universal for scientific development, namely the study of rational/agent design.
What is an agent?
• Perceive its surroundings through sensors
• Operate it through actuator
It will run back and forth in the cycle of perception, thinking and action. Take humans as an example. We perceive the environment through the human five senses (sensors) themselves, and then we think about it, and then use our body parts (actuators) to perform operations. Similarly, the machine agent senses the environment through the sensors we provide to it (can be a camera, a microphone, an infrared detector), then performs some calculations (thinking), and then uses a variety of motors/actuators to perform operations. Now, you should be clear that the world around you is filled with various agents, such as your phone, vacuum cleaner, smart refrigerator, thermostat, camera, and even yourself.
What is Intelligent Agent?
Intelligent Agent is an agent that, given what it perceives and what it has, prior knowledge , runs in a way that is expected to maximize its performance metrics.
performance metrics define the criteria for the success of an agent.
This type of agent is also called a rational agent. The rationality of an agent is measured by its performance metrics, its prior knowledge, the environment it perceives and its operations it can perform. The concept of
is the core of artificial intelligence.
Intelligent The above attributes of Intelligent Agent are usually attributed to the term PEAS (Performance, Environment, Actuators and Sensors), which represent performance, environment, actuators and sensors. So, taking an autonomous car as an example, it should have the following PEAS:
• Performance: safety, time, legal driving, and comfort.
•Environment: roads, other cars, pedestrians, road signs.
•actuator: steering, accelerator, brake, signal, speaker.
•Sensor: camera, sonar, GPS, speedometer, odometer, accelerometer, engine sensor, keyboard.
To meet the real-world usage, artificial intelligence itself needs to have a wide range of Intelligent Agents. This introduces the diversity of the types of agents we have and the environment. Next, let’s take a look together.
Environment type
If you want to design a rational agent, you must remember the environment type it will use, that is, the following types:
•Full observable and partial observable: If it is fully observable, the sensor of the agent can access the complete state of the environment at each point in time, otherwise it cannot. Chess, for example, is a completely observable environment, while poker is not.
• Determinism and randomness: The next state of the environment is entirely determined by the current state and the operations performed by the agent next. (If the environment is deterministic and the behavior of other agents is uncertain, then the environment is random). Random environments are random in nature and cannot be completely certain.For example, the 8-puzzle online puzzle game has a definite environment, but driverless cars don't.
• Static and dynamic: When the agent is deliberate, there is no change in the static environment. (The environment is semi-dynamic, and the environment itself does not change over time, but the performance score of the agent will change accordingly). On the other hand, the dynamic environment has changed. Backgammon has a static environment, while sweeping robot roomba has a dynamic environment.
•Discrete and Continuous: has a limited number of well-defined perceptions and behaviors that form a discrete environment. For example, checkers are an example of discrete environments, while autonomous vehicles need to operate in continuous environments.
• Single agent and Multi-agent : Only the agent that operates itself has a single agent environment. But if there are other agents included, then it is a multi-agent environment. Self-driving cars have multi-authority environments.
also has other types of environments, scenarios and orders, known and unknowns, which define the scope of the agent.
Types of agents
generally have 4 types of agents, which are distinguished according to the level of intelligence or the complexity of their ability to perform tasks. All types can improve performance over time and produce better operations. These can be summarized as learning agents.
Simple reflex agents
These selection operations are based only on the current state and ignore the perceptual history.
They can only work if the environment is completely observable, or the correct behavior is based on current perception.
Model-based reflective agents
agent tracks partial observable environment. These internal states depend on perceived history. The environment/world modeling is based on how it evolves independently from the agent and how agent behavior affects the world.
Goal-based agents
This is an improvement on model-based agents and is used when the current environment is not in short supply. The agent combines the provided target information with the environmental model to select actions to achieve the target.
Utility-based agents
improves target-based agents, which is not enough to help achieve the expected goals. We may need to consider the cost. For example, we might look for faster, safer, and cheaper trips to reach our destination. This is marked by a utility function. The utility agent will select the operation that maximizes the desired utility.
general agent, also known as learning agent, proposed by Alan Turing, is the most advanced method in the current artificial intelligence system.
All of the above agents can be summarized as learning agents to produce better operations.
Learning Agents
Learning Elements: Responsible for improving
performance elements: Responsible for selecting external behaviors, which is what we usually think of as of now.
Comment: How is the agent doing about the determined performance standards?
Problem Generator: Allows agents to explore.
internal state indicates
As the agent becomes more and more complex, the internal structure is becoming more and more complex. The way they store internal states also change. Due to its nature, simple reflective agents do not require storage states, but other types require them. The image below provides a high-level representation of the agent's state, arranged in order of representational capability growth (from left to right).
•Atomic representation: In this case, the state is stored as a black box, that is, there is no internal structure. For example, for Roomba (robot vacuum cleaner), the internal state is that the patch has been made with vacuum, and you don't have to know anything else.As shown in the figure, this representation is suitable for model and target-based agents and is used in various AI algorithms , such as search questions and adversarial games.
• Factored Representation: In this representation, the status is no longer a black box. It now has attribute value pairs, also known as variables that can contain values. For example, when finding a route, you have the GPS location and the amount of oil in the tank. This adds a constraint to the problem. As shown in the figure, this representation is suitable for target-based agents and is used in various AI algorithms such as constraint satisfaction and bayesian networks.
• Structured Representation: In this representation, we have a relationship between variable/factor states, which triggers logic in AI algorithms. For example, in natural language processing, a statement contains a reference to a person in a statement and whether the adjective in that statement describes the person. These statement relationships will determine whether the statement is sarcastic. This is a high-level artificial intelligence, algorithm for first-order logic, knowledge-based learning, and natural language understanding.
There are many more rational agents for artificial intelligence, and this is just an overview. As you know, the design research of rational agents is an important part of artificial intelligence because it has a wide range of applications in various fields. However, these agents cannot work independently, and they need an artificial intelligence algorithm to drive them. Most of these algorithms involve search. Soon, I will write about AI algorithms that drive rational agents and the use of machine learning in AI.
