“Should we really be measuring this by whether it can or cannot be done…?”
A chatbot released in 2016 was shut down after only a short time.
This book uses thought experiments to ask whether we can teach artificial intelligence to judge right from wrong or to understand ethics.
In conclusion, moving away from a binary, all-or-nothing approach allows for more flexible judgment.
It enables us to distinguish between strengths and weaknesses in different domains and how to handle things outside those frames.
In this note, while covering a wide range of themes such as book introductions, I mainly post about business.
If you are interested in business, please also take a look at the information for my membership, the ‘Mystery Business Laboratory’.
A perspective that brings the AI debate back from binary extremes to a matter of degree
The topic of artificial intelligence continues to be discussed in terms of a binary choice: whether it can or cannot do something.
Rather than whether it can or cannot, the granularity of where the difference lies changes the judgment
The material notes that artificial intelligence cannot do everything that humans can do in the same way.
On the other hand, it points out that there are fields where it demonstrates capabilities exceeding those of humans.
The book’s inquiry lies in using thought experiments to address subjects such as moral judgment, love, faith, and control.
It is introduced as a work where the author, a philosopher, interprets current artificial intelligence from a philosophical perspective.
In other words, it is not an explanation of technical specifications, but is positioned as an entry point for re-examining the definition of our relationship with it.
The example of the chatbot Tay, which was released in 2016 and shut down in a short time, is also cited.
Key point: Artificial intelligence is not something to be measured in binary terms, but something to be viewed in terms of domain differences.
Shifting ethics from a uniquely human problem to a design challenge for learning machines
The material organizes the idea that ethics has long been considered a problem unique to humans.
However, it points out that we have now entered a phase where we must teach ethics to artificial intelligence.
Previously, it was thought that as long as it operated on set programs, no responsibility would be incurred even if damage occurred.
This premise is shaken because the fundamental mechanism has become one that learns through data on its own.
It is stated that it becomes necessary to teach ethical principles regarding how it learns.
It is explained that Tay learned through machine learning from conversations with multiple users.
The document cites the repetition of praise for Hitler and discriminatory remarks as the reason for its shutdown.
Notes for moving away from binary thinking are easy to use if organized into the following four lines.
できる領域＝
できない領域＝
学習するデータ源＝
想定外のときの受け皿＝
If left blank, you can see where judgment is stalled based on the granularity of the material.
Preparing to treat it as an agent is a proactive task in anticipation of future development.
There is a point of discussion about treating artificial intelligence not just as an object of action, but as an agent that exerts influence on humans.
Whether or not to treat it as an agent significantly changes the scope of preparation required.
This is a bidirectional arrangement where, while humans provide consideration, consideration is also required from the artificial intelligence side.
This is positioned as a task that should be prepared for in order to anticipate future development.
It is explained that philosophical research centers on reading past texts and performing consistent interpretations.
The document states that there is no doubt that artificial intelligence will become a powerful weapon for that work.
On the other hand, the question of who verifies the validity of the interpretation is not explicitly stated in the document.
Therefore, at this point, it can be read that positioning it as an aid to interpretation work is realistic.
The axis of judgment is what happens when you step outside the frame.
Do not make the frame problem a defect unique to AI, but view it alongside human limitations.
The frame problem is introduced as a concept sometimes mentioned as a limitation of artificial intelligence.
It is explained as the inability to act if the frames that should be considered in advance are not set.
However, the document distinguishes this by noting that it is not a difficult problem unique to artificial intelligence.
The essence lies in the fact that while the information required for future actions is infinite, processing capacity is finite.
You can deal with it by limiting the frame, but it becomes vulnerable to information outside that scope.
It is pointed out that humans also unconsciously set frames according to their environment through education and learning.
It is suggested that how we accept self-driving cars should be considered with this problem in mind.
Information comparing the likelihood of accidents numerically is not included in this material.
The judgment of whether to treat something as an agent can be organized into the following criteria:
-
Is the learning path open to external data?
-
Is it designed with unexpected behavior in mind?
-
Is there a mechanism to leave the locus of responsibility with humans?
-
Who handles information outside the frame?
All of these are points indicated by the material, and the confirmation items in practice depend on each individual’s environment.
A book for those who want to confirm how to frame questions rather than expecting technical explanations.
This book focuses on organizing questions through thought experiments rather than implementation procedures or product comparisons.
It can be read as being positioned for those who want to refine how they formulate questions.
Since subjects such as good and evil, love, and faith are lined up, it must be read with the premise of high abstraction.
If you are looking for procedures that can be used immediately in the field of ethical design, the scope of this material is insufficient.
It serves as an entry point for those who want to rethink the acceptance of self-driving cars and conversational bots from the perspective of the frame problem.
Since the latest trends in technology change over time, cases after publication need to be checked separately.
The material focuses on introductory text and key points, and details such as chapter structure, volume, and price cannot be confirmed.
It is realistic to first check the overview and table of contents to see if the approach to thought experiments matches your reading style.
Recommendation Card for This Book
-
Who this book is for: Suitable for those who want to refine the granularity of arguments in situations where the limitations of artificial intelligence are often discussed in binary terms.
-
Who should not read this yet: Not suitable for those looking for immediate implementation steps, the latest product comparisons, or templates for ethical guidelines.
-
When to read it: Can be read at the stage of reviewing how to set premises before considering adoption policies for autonomous driving or conversational bots.
-
The role of this book: Not a technical guide, but a reference point for reordering how to frame questions about good and evil, learning, and responsibility.
-
What remains after reading: It is more flexible to view artificial intelligence as having different strengths and weaknesses depending on the domain, and to see the differences as a matter of degree.
-
Proximity to action: Within the scope of the material, no steps are provided that can be executed immediately after finishing the book.
-
Re-readability: The organization of the frame problem leaves room for reference when verifying the premises of a judgment.
-
What not to expect: Please do not expect guarantees of results or safety, numerical data on learning costs, or detailed chapter breakdowns.
-
Recommended way to read: Since there are many highly abstract thought experiments, it is best to first grasp the overall structure by checking the table of contents.
-
Final decision: It is sufficient to check the summary and table of contents to see if the approach to the questions aligns with your needs before deciding to proceed to the main text.
In Conclusion
The scale for measuring the capabilities of artificial intelligence relies more on understanding domain differences than on binary classifications.
First, check the summary and table of contents to see if the progression of the thought experiments suits you.
If the way the questions are arranged seems to fit, it is sufficient to check the contents of the book before making a decision.
Thank you for reading this far.
In this note, I write articles on various themes, including book reviews, but the main focus is on business topics.
In the membership ‘Mystery Business Lab,’ I research interesting businesses found online or abroad, and share the process of actually trying them out and what I learned from them.
If you are interested in business mechanisms, practice, and verification, please take a look at this guide. It summarizes my usual content and membership plans.
※Advertisement (Amazon Associates)
