Kitabı oku: «Incredible artificial intelligence Easy Diffusion 3.0. The Middle and High School Guide», sayfa 4
Term 3. New Artificial Intelligence
New Artificial Intelligence (Nouvelle AI) is an approach to Artificial Intelligence first proposed in the 1980s by Rodney Brooks, who was then working at the Massachusetts Institute of Technology’s Artificial Intelligence Laboratory. New AI differs from classical AI in that its goal is to create robots with insect-level intelligence. Some scientists believe that intelligence can arise independently in living beings from simple interactions between themselves and other beings in the real world16.
Working with Easy Diffusion
Let’s try to create an image of Artificial Intelligence with the level of intelligence of an insect.
To do this, perform the following steps:
– Let’s create a more complex query based on the definition above: «steampunk, cyborg worm, journey into the healing power of nature, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail.» The phrase literally translates to: «steampunk, cyborg worm, journey into the healing power of nature, photorealism, perfect composition, cinematic frame, intricate details, hyper-detailed.»
– We will use «DreamShaper» as a model.
– In the «Negative Prompt» field, add the phrase: «dark background, human face, woman face,» which will allow us to avoid getting dark pictures and people’s faces.
–
– To get a better-quality picture, set the «Image Quality» parameter to 85.
Click the «Make image» button and get a new result:
Please note that our Artificial Intelligence looks like some kind of mechanized animal – this is very symbolic. Perhaps in the future we will see some kind of symbiosis (or, in other words, the complex coexistence of different forms of life together) not only of man and machine, but also of various living beings, plants and machines.
Text-to-image AI software: Easy Diffusion
Prompt: steampunk, cyborg worm, journey into the healing power of nature, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail
Model: DreamShaper
Inference Steps: 25
Image Quality: 85
Image Author: Alexander Chesalov
In my case, the result obtained exceeded all my expectations. Easy Diffusion created a very unusual image of «Nouvelle AI».
What happened for you?
Term 4. Friendly Artificial Intelligence
Friendly Artificial Intelligence is a field of research whose goal is to create a positive and human-friendly system that can significantly improve the conditions of its existence and work. The intellectual abilities of such a system should be at the human level or even exceed them17.
Working with Easy Diffusion
To consolidate our skills, let’s do an exercise and try to create an image of friendly Artificial Intelligence.
But this time we’ll pay attention to the image rendering settings, which are located at the bottom of the Tools panel:
To do this, perform the following steps:
– Let’s make a request based on the above definition: «Friendly Artificial Intelligence, smiling cyborg face, steampunk, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail.» Please note that we use additional phrases to strengthen our request:
– «Smiling» to give positive emotions.
– «Steampunk» for image stylization.
– «Soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail» to give the image a more realistic look.
– We will use «DreamShaper» as a model.
– Set the «Image Quality» parameter to 75.
– And most importantly, activate all four options in the «Render Settings» section:
– a. «Show a live preview». This option will show us how the system creates the image step by step:
There is no particular need to use the «Show a live preview» option due to the fact that it only increases the time it takes to create a new image.
– «Fix incorrect faces and eyes.» This is a very useful option that corrects while creating an image of the face and eye. The default values are sufficient, but you can safely change its parameters and compare the rendering results of new images. If you have not activated this option and you have such a problem with your face or fingers, correct it with the «Fix Face» command or take another 25 steps with the «Draw another 25 steps» command.
– «Scale up by» This option allows you to immediately obtain an image in the maximum resolution for this system. Its use is only advisable when you are sure in advance that you will get a good version of the new image, or when you are using a powerful computer for your work. This option also increases the time it takes to create a new image. For scaling photos, it is better to choose the algorithm: «RealESRGAN_4x», for anime: «RealESRGAN_4x+Anime_6B».
– «Show only the corrected/upscaled image.»
Click the «Make image» button and get the following result:
Text-to-image AI software: Easy Diffusion
Prompt: Friendly Artificial Intelligence, smiling cyborg face, steampunk, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail
Model: DreamShaper
Image Quality: 95
Inference Steps: 25
Image Author: Alexander Chesalov
Friendly Artificial Intelligence turned out to be quite positive.
What happened for you?
Term 5. Human-Centered Artificial Intelligence
Human-centered Artificial Intelligence is a priority principle for the design and development of Artificial Intelligence, in which the goals, objectives and limitations of a person, a group of people or all of humanity are placed first in order to ensure that their interests and rights are respected.
In a global sense, human-centered Artificial Intelligence should work for the benefit of all humanity, the individual, societies, the environment and ecosystems, as well as their harmonious development and prevent harm to them18.
In order to create human-oriented Artificial Intelligence, it is necessary to use an interdisciplinary approach to studying the problems of interaction between humans, computers and Artificial Intelligence. To achieve positive results in this line of work, human-centered Artificial Intelligence must be fair and impartial, safe and ethical19.
Working with Easy Diffusion
Now let’s try to cope with a more difficult task and try to create an image of human-oriented Artificial Intelligence based on a request and an additional image.
To do this, perform the following steps:
• As an initial request, write the following: «Human-centric Artificial Intelligence, style of the artist Picasso, full length portrait, highly detailed good face, cinematic lighting.» Please note that we use an additional phrase to stylize the image: «style of the artist Picasso.»
• You need to download an image of Pablo Picasso’s painting «The Weeping Woman» from the Internet:
• Load the downloaded image using the Initial Image (img2img) option on the Toolbar.
• We will use “DreamShaper” as a model.
• Set the «Image Quality» parameter to 75.
Click the «Make image» button and get the following result:
Text-to-image AI software: Easy Diffusion
Prompt: Human-centric Artificial Intelligence, style of the artist Picasso, full length portrait, highly detailed good face, cinematic lighting
Model: DreamShaper
Inference Steps: 25
Image Author: Alexander Chesalov
Now let’s complicate our task and make a new one based on the resulting picture.
To do this, perform the following steps:
– Move the mouse cursor over our image and select the «Use as Input» command.
Click the «Enqueue next image» button and get the following result:
As we can see, the result is a rather unusual image of human-oriented Artificial Intelligence based not just on an artistic work of art, but on a masterpiece by the famous artist Pablo Picasso.
In the appendix to this book, you can see other versions of images created based on the painting «Crying Woman». I also suggest that you conduct your own creative experiments with Easy Diffusion, using any other works of art by other brilliant artists as a basis.
Text-to-image AI software: Easy Diffusion
Prompt: Human-centric Artificial Intelligence, style of the artist Picasso, full length portrait, highly detailed good face, cinematic lighting
Model: DreamShaper
Inference Steps: 25
Image Author: Alexander Chesalov
Term 6. Supplementary Artificial Intelligence
Supplementary Artificial Intelligence is an Artificial Intelligence system that helps a person make decisions based on additional information obtained from an analysis of a person’s interaction with the world around him. Ancillary or complementary Artificial Intelligence can be a useful addition to a human-centered Artificial Intelligence system. Often, assistive Artificial Intelligence is a system that is used by specialists to help solve specialized problems. For example, doctors use Artificial Intelligence as an auxiliary system in diagnosing cancer tumors or congenital heart defects.
In machine learning, there is a similar term called «assisted learning.» Assisted learning is an approach in which, through machine learning, a model detects the presence of objects that do not fit into any of the categories it has learned. The name «Assisted Learning» was chosen due to the introduction of an auxiliary class and is used to study unknown objects20.
Working with Easy Diffusion
Let’s try to create an image of assistive Artificial Intelligence.
To do this, perform the following steps:
– Let’s create a more complex query based on the definition above: «dog cyborg that is carrying a bag, cinematic lighting, highly detailed good face, cinematic lighting, white city background.»
– We will use «DreamShaper» as a model.
– To get a better-quality picture, set the «Image Quality» parameter to 95.
Click the «Make image» button and get a new result:
Text-to-image AI software: Easy Diffusion
Prompt: dog cyborg that is carrying a bag, cinematic lighting, highly detailed good face, cinematic lighting, white city background
Negative Prompt: black background
Model: DreamShaper
Inference Steps: 25
Качество изображения (Image Quality): 95
Image Author: Alexander Chesalov
A distinctive feature of this exercise was a not very complex query that did not contain any mention of Artificial Intelligence as such. However, we mentioned a cyborg in the description, which undoubtedly influenced the content of the image.
What happened to you?
Term 7. Artificial Intelligence of the Commonsense
Artificial Intelligence of the Commonsense is one of the areas of development of Artificial Intelligence, which is engaged in modeling a person’s ability to analyze various life situations and be guided in their actions by common sense21.
Common sense is the accumulated totality of generally accepted knowledge and views, ideas, forms of thinking – about the driving forces of nature and society, relationships between people; in addition to knowledge, includes values, beliefs, regulators of practical activities, moral and legal norms, elements of religious experience, artistic and aesthetic assessments of reality22.
Working with Easy Diffusion
Let’s try to create an image of sane Artificial Intelligence.
To do this, perform the following steps:
– Let’s make a request based on the above definition: «Sane Artificial Intelligence, robot solving human’s problems, cyborg, red light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail.»
– We will use «DreamShaper» as a model.
– To get a better-quality picture, set the «Image Quality» parameter to 75.
Click the «Make image» button and get a new result:
Text-to-image AI software: Easy Diffusion
Prompt: Sane Artificial Intelligence, robot solving human’s problems, cyborg, red light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail
Model: DreamShaper
Inference Steps: 25
Image Author: Diana Chesalova
What happened for you?
Term 8. Artificial general intelligence
Strong Artificial Intelligence is an applied Artificial Intelligence system, the technologies and algorithms of which can perform a significant number of tasks of data analysis, decision-making based on them and their implementation, providing an imitation of human intellectual abilities and explainability of the decision options offered to a person, reproducing and sometimes exceeding a wide range of cognitive and intellectual abilities of a person, including the interpretation of external data and influences and the extraction of meaning from them, the use of acquired knowledge for learning, planning and decision-making in conditions of uncertainty and the achievement of specific goals and objectives through flexible adaptation to changing conditions and interaction with the external environment23.
In other words, strong Artificial Intelligence is intelligence that is indistinguishable from human intelligence and is self-aware. He is able to see, hear, learn, solve problems, plan and improve himself, and most importantly, he has imagination and is capable of creativity.
Artificial General Intelligence and Artificial Intelligence at the human level (Human Level Machine Intelligence) are synonyms for strong Artificial Intelligence. These terms refer to the degree of development of Artificial Intelligence at the human level.
To date, strong or general Artificial Intelligence exists only as a theoretical concept. Some scientists suggest that its practical implementation will occur no earlier than 2045. Nevertheless, among a large number of scientists and practitioners there is a strong opinion that all even the most modern and advanced Artificial Intelligence systems that exist today are weak AI, and predictions that strong AI will be created by 2045 are a myth. The reason for this opinion is the complexity of the structure of our brain from the point of view of creating a supercomputer similar to it, which could at least somewhat approach the human brain in its computing power. Perhaps in the near future, a breakthrough will be made in the field of quantum computers and computing, which will bring us closer to the creation of strong Artificial Intelligence.
Working with Easy Diffusion
To consolidate our work skills, let’s do a simple exercise and try to create an image of strong Artificial Intelligence.
To do this, perform the following steps:
– Create your own queries based on the definition above. For example, we wrote the following:
– «Strong Artificial Intelligence, an athletic male robot that holds the planet Earth on its shoulders, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail, cyborg.»
– «Robot, cyborg, highly detailed, strong robot, male, day, smart robot, look at camera, head, perfect composition, in the space, Artificial General Intelligence.»
– «Artificial General Intelligence, robot lying under the tree, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail, cyborg.»
– «Development of Artificial Intelligence at the human level, Artificial General Intelligence, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail.»
– «Human Level Machine Intelligence, cyborg, robot, highly detailed, highly detailed face, robot face, human, man.»
– Use «DreamShaper» as a model.
– Set the «Image Quality» parameter to 75.
Click the «Make image» button.
We got the following results:
Text-to-image AI software: Easy Diffusion
Prompt: Strong Artificial Intelligence, an athletic male robot that holds the planet Earth on its shoulders, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail, cyborg
Model: DreamShaper
Inference Steps: 25
Image Author: Diana Chesalova
Text-to-image AI software: Easy Diffusion
Prompt: Robot, cyborg, highly detailed, strong robot, male, day, smart robot, look at camera, head, perfect composition, in the space, Artificial General Intelligence
Model: DreamShaper
Inference Steps: 25
Image Author: Vsevolod Chesalov
Text-to-image AI software: Easy Diffusion
Prompt: Artificial General Intelligence, robot lying under the tree, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail, cyborg
Model: DreamShaper
Inference Steps: 25
Image Author: Diana Chesalova
Text-to-image AI software: Easy Diffusion
Prompt: Strong Artificial Intelligence, Strong AI, Transhumanism, Bioconservatism, robot of the future, cyborg, highly detailed, highly detailed face, soft light, perfect composition, head, look at camera
Model: DreamShaper
Inference Steps: 25
Image Author: Vsevolod Chesalov
Text-to-image AI software: Easy Diffusion
Prompt: development of Artificial Intelligence at the human level, Artificial General Intelligence, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail
Model: DreamShaper
Inference Steps: 25
Image Author: Diana Chesalova
Text-to-image AI software: Easy Diffusion
Prompt: development of Artificial Intelligence at the human level, Artificial General Intelligence, soft light, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail
Model: DreamShaper
Inference Steps: 25
Image Author: Diana Chesalova
Text-to-image AI software: Easy Diffusion
Prompt: Human Level Machine Intelligence, cyborg, robot, highly detailed, highly detailed face, robot face, human, man
Model: DreamShaper
Inference Steps: 25
Image Author: Vsevolod Chesalov
Term 9. Explainable Artificial Intelligence
When we talk about Artificial Intelligence, we most likely imagine it as some kind of intangible or abstract perfect intelligent system that communicates with us in a pleasant voice and is able to give us an answer to any question that interests us.
However, Artificial Intelligence is a very specific computer program created by scientists, engineers and many other specialists from a wide variety of industries and fields of science and technology.
At the very moment when we realize the fact that we are communicating with the program, receiving information from it and we need to make some important decision for ourselves, we begin to think about the questions of whether we should unconditionally trust the information or advice of any kind. then the «smart» program and why exactly this program advises us to do this and not otherwise.
It is when we are faced with the question of «explainability» that the so-called «explainable Artificial Intelligence» comes to the fore.
Explainable Artificial Intelligence (XAI) is a set of rules and methods that allow users of an Artificial Intelligence system to understand why the machine learning algorithms of this system came to certain results and/or conclusions. Explainable Artificial Intelligence makes the operation of the AI system used transparent to its users, inherently opposing the «black box» principle of machine learning.
Questions about the explainability of the results obtained from an Artificial Intelligence system are actually of interest not only to you and me, as users of such systems, but also to the software engineers who create them.
When developers move from programming and experimenting to solve simple automation problems to developing serious software systems, for example, decision support systems, they must not only understand themselves, on the basis of which certain results are obtained, but also be ready to explain their origin to users the Artificial Intelligence system they are developing. This must also be done because each developer bears personal responsibility for the algorithms, models, programs, systems and other applications and services he creates that will subsequently be used by people. An incorrect decision based on the «inexplicable» results of an Artificial Intelligence system can lead to catastrophic consequences for both an individual person, a single company, a city, or even an entire country.
It often happens that the programmer training the model is not an expert in a particular subject area, and this is actually a big problem. For example, a programmer is not a doctor, which is, of course, obvious. As you know, many doctors have a wealth of knowledge, experience and intuition that can tell at first glance that one or another data obtained from the use of a particular model used in a decision support system cannot be trusted. The specialist will immediately note that the model makes an unreasonable forecast. Undoubtedly, programmers and doctors should work in close cooperation, but it very often happens that the programmer makes decisions about certain results of the work independently, which may entail the presence of a hidden error in the calculations and the accuracy of the data obtained. It is in these cases that software engineers must pay great attention to explainability issues when developing Artificial Intelligence systems.
Today, it is necessary to create additional Tools for understanding and explainability of the operation of an Artificial Intelligence system, which would be useful in making decisions about whether the data obtained with its help is trustworthy or not24,25.
For example, software engineers use the Lime library to explain the predictions of basic machine learning models and evaluate their usefulness in various classification and regression problems. Today, the library works with the analysis of text and table classifiers, and also works with image classifiers. Lime can explain any black box classifier with two or more classes26.
The need to implement explainable Artificial Intelligence both for the developer and for you and me, as users of Artificial Intelligence systems, becomes not only obvious, but also a completely important and responsible task.
Another important property of an Artificial Intelligence system, on a par with «explainability,» is its «transparency.»
Transparency of the operation of Artificial Intelligence systems is ensured by providing complete open information to users of Artificial Intelligence systems (that is, for you and me) about what products or services are provided directly or with the help of Artificial Intelligence systems. And also what management decisions are made by certain individuals, companies or organizations based on data obtained and provided by Artificial Intelligence systems. This information should indicate that the decisions made do not violate human rights (users of Artificial Intelligence systems) and do not expose them to any danger. A high level of transparency will provide the possibility of control by society, which, in turn, can help reduce the level of corruption, discrimination or the risk of one or another type of harm.
Thus, ensuring explainability and transparency of the operation of Artificial Intelligence systems is an essential condition not only during their design and development, but also during their operation in order to ensure respect for the privacy of citizens and the protection of human rights and freedoms throughout the entire life cycle of the created AI systems27,28,29.
The terms «transparency» and «explainability» are integral criteria for evaluating a «trusted Artificial Intelligence» system, the definition of which we will introduce you to in the next chapter.
Working with Easy Diffusion
To consolidate our work skills, let’s do the following exercise and try to create an image of explainable Artificial Intelligence.
To do this, perform the following steps:
– Create your own queries based on the definition above. For example, we wrote the following:
– «Explainable Artificial Intelligence, solving the simplest automation tasks to developing serious software systems, transparency, unmistakableness, robot, cyberpunk.»
– «Transparency, cyborg, look at camera, head, invisible.»
– We chose DreamShaper and Realistic Vision as models. Please note that these models are best suited for creating images of robots and cyborgs.
– Set the «Image Quality» parameter to 75.
Click the «Make image» button.
We got the following results:
Text-to-image AI software: Easy Diffusion
Prompt: Explainable Artificial Intelligence, solving the simplest automation tasks to developing serious software systems, transparency, unmistakableness, robot, cyberpunk
Model: DreamShaper
Inference Steps: 25
Image Author: Vsevolod Chesalov
Text-to-image AI software: Easy Diffusion
Prompt: Transparency, cyborg, look at camera, head, invisible
Model: Realistic Vision
Inference Steps: 25
Image Author: Vsevolod Chesalov
