Essential Topics in Artificial Intelligence - Book Review
Essential Topics in Artificial Intelligence by Anthony Rhem is a guide to responsible AI, the practice of developing and using AI with safeguards for fairness, privacy, transparency, and human oversight. Drawing on real-world examples, Rhem shows how organizations can put these principles into practice by setting policies, assigning accountability, and measuring progress.

The book “Essential Topics in Artificial Intelligence: An Examination of AI Ethics and Governance, Large Language Models, Artificial General Intelligence and Natural Language Processing” was written by Anthony Rhem and published in 2024. The book surveys the topics mentioned in its title but focuses on ethics and governance, his main areas of expertise.
Main topics covered in the book:
The book contains countless examples of risks and ethical considerations involved in implementing artificial intelligence in various contexts. It is recommended reading for anyone responsible for implementing AI in an organization who wants to learn firsthand how to act responsibly.
And the good news - despite it all, Rhem is optimistic about our future. So am I!
Ethics, Standards, and Governance
Ethics
The term ethics, as used in this book, refers to the full set of aspects required to ensure responsible artificial intelligence.
Note: this is a broad use of the term. In the literature, other definitions also exist in which "ethics" refers mainly to professional and cultural principles and standards, while other terms cover additional aspects of responsible AI. Nevertheless, there is considerable overlap between the aspects discussed and the approaches proposed for addressing them, and familiarity with the different terminology helps in understanding the context in which it is used.
The main aspects of ethics include:
Information Security Preventing information leakage from the organization, its exposure, or its becoming public information as a result of the use of artificial intelligence systems
Data Privacy - Preventing improper exposure and use of personal data and information about individuals
Fairness - including:
Preventing biases and discrimination arising from the use of data that is not properly curated or representative; these issues may manifest at various stages, from defining the data repositories, through data synthesis, selection of algorithms and models and processing procedures, up to quality assurance and validation
Preventing discrimination created by algorithms, whether as a result of error or as a result of amplifying existing human biases and discrimination
Ensuring fair access for various populations to artificial intelligence resources, its applications, and its outputs
Copyright - Preventing unauthorized use of information protected by copyright, as well as addressing the rights applicable to new information and content created by means of artificial intelligence
Environmental Responsibility - including:
Reducing the environmental impact resulting from the consumption of large-scale computing and energy resources
Acting in accordance with relevant regulation and ethical guidelines
Transparency and Explainability - The ability of humans to understand how the artificial intelligence system operates and why it recommended a particular course of action; this ability enables oversight, supports collaboration in decision-making processes, and helps build trust
Human Autonomy - Preserving humans' ability to control artificial intelligence systems and the manner in which they are used, even as the systems become more autonomous
Accountability- The ability to define and identify who bears responsibility for the system's actions and their outcomes, including in cases of error or harm.
Standards
Rhem presents throughout the book several key frameworks, principles, and standards that should be recognized in the context of responsible artificial intelligence:
The OECD principles for artificial intelligence
The Blueprint for an AI Bill of Rights of the White House
The AI Act of the European Union
Frameworks and principles of IEEE for artificial intelligence
Frameworks and standards of NIST for artificial intelligence
These frameworks address, to varying degrees, the ethical aspects described above. Rhem offers a mapping of the aspects each covers.
Also presented are frameworks dealing with privacy protection, foremost among them the -GDPR European regulations.
Governance
Alongside the principles and standards, organizational governance is required to ensure their practical implementation.
Rhem offers several recommendations for governance and the implementation of responsible artificial intelligence in organizations:
Begin AI implementation activity by formulating a policy for responsible artificial intelligence
Establish a team diverse in expertise, experience, and perspectives to develop, implement, and oversee AI products, strengthening critical thinking and decision-making.
Define metrics that will make it possible to examine progress in implementing the principles, guidelines, and standards of responsible artificial intelligence.
Recognize the inherent tension between the requirements of responsible AI and considerations of efficiency, economics, and the desire to maximize the use of artificial intelligence, and manage governance while balancing between them.
Adopt an approach of (Human-Centered AI).
Ethics from the Perspective of Knowledge and Data
Knowledge
The terms "data," "information," and "knowledge" are sometimes used interchangeably; however, in the context of this book, we will focus on the data used as input to artificial intelligence systems and on the data and information they produce as output, all being knowledge enablers:
Making knowledge accessible
Understanding existing knowledge
Learning new knowledge
Problem-solving
Knowledge-based decision-making
The improvement is multidimensional: not only in the ability to support the tasks themselves, but also in the dynamism of the knowledge (ensuring its currency), its accuracy, and its fit to the user’s needs and context. Rhem presents examples of these uses in autonomous vehicles, financial services, and healthcare.
However, realizing these benefits depends on implementing responsible AI principles grounded in ethics and a human-centered approach to action (Human-Centered AI). In accordance with OECD principles, artificial intelligence systems must support human well-being and inclusiveness (Inclusiveness), respect individuals' rights and values and treat them fairly, and operate safely, transparently, and responsibly.
From a knowledge management perspective, this means that responsibility is not limited to managing the knowledge itself, but also extends to ensuring its quality, reliability, currency, fair representation, and ethical use. The quality of the knowledge fed into artificial intelligence systems directly affects the quality of the decisions and outputs they produce; therefore, responsible knowledge management is a basic condition for responsible artificial intelligence.
Implementing knowledge management in organizations must account for these ethical values and ensure that AI systems manage the knowledge they use in a way that guarantees the ethics of the information and its use.
Data
If knowledge is the asset on which artificial intelligence systems rely, then data is the raw material from which this knowledge is built. Data quality directly affects output quality, decision quality, and the system's level of ethical responsibility. In this book, we distinguish between the input data used by artificial intelligence systems and the data and information they produce as output. At each stage, we must consider different ethical aspects.
Inputs:
Biases in the data and data repositories, and ensuring their fairness
Protecting the privacy of information
Outputs:
Transparency
Explainability
Privacy protection
Information security
Ensuring data ethics is not a one-time action but an ongoing process. Routinely check data quality, identify biases and gaps, assess the data's fitness for use, and, where necessary, curate it, re-select data sources, or validate the system's results.
Attention should also be paid to the use of (Synthetic Data) and to ensuring that it faithfully represents reality and does not create biases or misleading conclusions.
Technologies
Among the various AI technologies, Rhem dwells on key technologies in which the question of ethics is significant:
Large Language Models LLMs
Large language models, such as OpenAI's GPT series, are deep neural network models designed to understand and generate text based on patterns learned during training and the context provided during use. The architecture is based on neural networks with many computational layers, which is the source of the term “deep learning.” The input is divided into small units (tokens), representing words or parts of words, and each unit is mapped to a vector representation (embedding).
In the context of ethics, it is important to examine both the model's training and operating processes. In the training process, attention should be paid to data quality, the manner of its collection and processing, its curation, the balance between different groups, the selection of model parameters, learning loops, model accuracy, and extending use beyond the training conditions (scaling ), as well as adaptations for different audiences and personalization.
Main uses: text generation, natural language understanding, translation, and task performance.
Rhem presents uses and challenges specific to the content worlds of education, business, research, and entertainment.
Main challenges and limitations:
Cognitive Biases
Space and memory limitations
Statistical rather than conceptual understanding
The need for large available data repositories
Data quality
Computational costs
Explainability
Scalability
This technology has created a revolution in language understanding and processing, as described below, and it has numerous revolutionary uses across many areas of life and work.
Natural Language Processing NLP
Natural Language Processing (NLP) is a field of artificial intelligence that develops algorithms and models that enable computers to understand, analyze, process, and generate human language. This field is a key component in search engines, large language models, translation systems, chatbots, and a wide range of other AI applications. In recent years, language-processing capabilities have advanced significantly, overcoming many translation challenges and enabling nearly seamless communication between languages.
Key capabilities and techniques in natural language processing include dividing text into units (tokens), syntactic analysis and tagging, identification of entities in text (Named Entity Recognition), meaning comprehension (Semantic Analysis), identifying relationships between parts of text, sentiment analysis, automatic translation, text summarization, and question answering. Large language models rely on these capabilities and expand them.
Main challenges and limitations:
Ambiguity of words and expressions
Biases in data and models
Data quality, completeness, and availability
Data privacy protection
Information security
Natural language processing plays a growing role in identifying incorrect information, detecting contradictions, and improving the quality of information produced by language models, including helping to identify hallucinations and unsubstantiated answers. The field's rapid development drives progress in language models, search engines, translation systems, smart service systems, and many other AI applications.
Artificial General Intelligence AGI
Artificial General Intelligence (AGI) is a research goal in artificial intelligence, describing systems that can understand, learn, and perform a wide range of tasks at a level similar to humans. This involves transferring knowledge and experience across different fields and handling new, unplanned situations. Unlike dedicated artificial intelligence systems, which are designed to perform defined tasks, AGI aims to demonstrate general and flexible intelligence.
An AGI system is expected to include, among other things, the following capabilities:
Broad cognitive abilities
Abstract reasoning and analogical thinking
Learning, adaptation, and adjustment to changing situations
Transfer of knowledge and learning between different fields of knowledge (Transfer Learning)
Autonomous decision-making in complex situations.
Main challenges and limitations:
Developing a general intelligence capability similar to that of humans
Learning and transferring knowledge between different fields
Large-scale consumption of computing and energy resources
Ability to learn effectively even from limited amounts of information
Safety, control, and aligning the system's goals with human values (AI Alignment)
Philosophical and ethical issues related to values, consciousness, responsibility, and identity.
In recent years, there has been significant progress in the capabilities of artificial intelligence systems, particularly large language models. However, no consensus exists that artificial general intelligence has been achieved. Nevertheless, many see these developments as significant steps toward AGI. If and when this capability is achieved, it is expected to profoundly affect the economy, the labor market, science, society, and decision-making systems.
Continuing along this path may also lead to the development of superintelligence (Artificial Superintelligence - ASI) - systems whose performance significantly exceeds human capabilities in almost every cognitive domain. This scenario is accompanied by extensive professional and public discussion about the opportunities it creates, as well as the risks and challenges of developing systems that may surpass human capabilities.
AI Applications
Artificial intelligence is relevant and influences almost every area of our lives, both at work and in personal life. Nevertheless, some fields have especially broad applications and impact. Rhem chooses to delve into several key fields:
Medicine:
A significant portion of the AI examples focuses on medicine. This reflects both the enormous potential of artificial intelligence to improve medicine and the high risk of mistakes and errors. Key applications include medical diagnosis, treatment planning, medical record management, and monitoring patients' medical conditions. Artificial intelligence also assists learning and research, including the development of new drugs and medical devices. Alongside this, AI and robotics are increasingly integrated to assist with surgeries.
Manufacturing:
Manufacturing is among the most advanced fields for AI applications. These applications serve a variety of needs, including quality assurance, predictive maintenance and maintenance optimization, production chain and inventory management, integration of robotics and mechanization in production and assembly processes, energy management, and product customization for groups and customers.
Finance:
Artificial intelligence is integrated into many aspects of financial management, similar to other modern management fields. Alongside these, several applications particularly characterize the field, including credit rating, trading algorithms, fraud detection, financial advisory bots, personal financial management, financial trend forecasting, and risk management.
Commerce:
Artificial intelligence is integrated into personalized purchasing experiences, customer recommendation systems, supply chain optimization, product visualization (such as clothing and glasses) to promote online purchasing, payment automation, trend identification, store layout optimization, and customer behavior analysis.
These applications illustrate how far artificial intelligence has come from research laboratories to practical solutions in the field, changing the way we work, consume services, and go about our daily lives.
Collaboration between Human and AI
Collaboration between humans and artificial intelligence refers to the full set of interactions between them, including cooperation in performing tasks and knowledge-sharing for learning, decision-making, and creating new knowledge. This approach views artificial intelligence as a partner that empowers human capabilities, not merely a tool that performs tasks on one's behalf. The quality of collaboration between humans and AI is today considered one of the key factors in extracting value from artificial intelligence.
To achieve effective collaboration, attention must be paid, among other things, to the following aspects:
A natural and comfortable interface (UX) for the collaboration
Explainable artificial intelligence
Thoughtful planning of the division of responsibility and decision-making between human and AI
Collaboration can take many forms: voice or text interaction; joint work in real time or as a sequence of activities; interaction within a single session or across an entire project; and varying levels of human involvement and AI autonomy. As the technology advances, the range of collaboration patterns between humans and AI expands.
Main challenges and limitations:
Transparency and explainability.
Handling errors and uncertainty.
Learning from user feedback.
Balancing personalization with privacy protection.
Adaptation and learning over time, while maintaining users' trust and work habits.
Typical usages:
Collaboration between humans and artificial intelligence forms the basis for increasing creativity, creating new knowledge, improving decision-making, and raising productivity across many areas of life and work. It enables combining human capabilities - including judgment, experience, values, and contextual understanding - with the capabilities of artificial intelligence in processing information, identifying patterns, and generating insights. Alongside the opportunities, attention must also be paid to copyright, intellectual property, and responsibility for the knowledge used and the new knowledge created within the collaboration.
Optimal collaboration requires careful planning for how roles are divided between humans and artificial intelligence. Some tasks will shift to full or partial AI responsibility, while others will retain a central human role or require close collaboration. Choosing the collaboration model appropriate for each task is expected to significantly affect work processes, role structures, and the future work environment, as the next chapter will detail.
AI in the Future Work Environment
One question troubling all of us concerns our personal future and our place in the future labor market. Toward the end of the book, Rhem devotes a chapter to discussing this important topic.
The underlying assumption is that artificial intelligence will change the face of the world of work in several respects:
Reducing and even eliminating jobs involving repetitive work, such as some jobs in the manufacturing world, where AI can mechanize processes that until now were performed by humans
Creating new jobs in professions related to the implementation of artificial intelligence, including Data Science, Machine Learning Engineering, AI Ethics, and maintenance of AI
Creating new opportunities in ways of working, particularly by expanding remote-work options.
A need for continuous learning (see the book by Alex Bennet, Infinite Learning ) and constant upgrading of capabilities
Rhem lists a variety of occupational fields and, for each, provides examples of new roles that will be created and places where AI may replace humans.
Because of the complexity and sensitivity of this issue, he interviews and quotes several AI experts and presents their perspectives.
Points worth noting:
According to Annie Green, our challenge often stems from the need to balance AI capabilities with human involvement (see Green's book, Diary of HI and AI).
Adams and Horton emphasize that artificial intelligence is meant to empower human potential.
The future awaiting us is fascinating. Let's step into it wisely and responsibly.





Comments