De Kai
AI Professor @ HKUST CSE / Berkeley ICSI / The Future Society
Table of Contents
- Grading scheme
- Project
- Syllabus
- Welcome!
- Chapter 2: Our artificial children
- Trolley problems everywhere!
- Chapter 17: Lessons from the history of AI
- Paradigms of AI ethics
- Preface, Afterword: The toxic AI cocktail
- AI and social disruption
- Chapter 12: Neginformation
- Willful algorithmic negligence
- Chapter 13: Algorithmic censorship
- Misinformation theory
- Project planning [0723]
- The Social Dilemma [0724]
- Chapter 8: Cognitive bias [0727]
- Deviation from rationality
- Final project
- Required texts
- Reference material
Do not index
This course examines leading frontiers of AI and NLP, building from Raising AI — the widely acclaimed new MIT Press book and forthcoming Penguin Random House audiobook. Large language models are impressive but still unable to truly understand and share the feelings and thoughts of the human-which is the very definition of empathy. Most of meaning actually lies in what's not said: the omitted context. True AI breakthroughs will have to emerge from understanding how context contributes to meaning-beyond the simplistic context windows of LLMs. We will design and prototype new empathetic AI understanding models to recognize and measure these deeper levels of meaning. Major societal applications include misinformation theory, information disorder, algorithmic bias, and mental well-being AI.
This is a rare opportunity to learn directly beyond the instructor's influential book which is leading the international pack and has been rapidly shifting the way Silicon Valley thinks since the book's release last year. It is the only book (out of thousands of AI books!) that JPMorgan awarded in its legendary Reading List each summer for over a quarter century — and JPMorgan has been giving thousands of copies to its most influential networks worldwide. It has won "must read" critical acclaim from Malcolm Gladwell, Adam Grant, Susan Cain, and Daniel Pink's book club, as well as from Literary Hub, Science magazine, Forbes, and many others, as well as a rare global TEDx Editor's Pick (out of 56,000 TEDx talks worldwide each year!)
Grading scheme
- 26% exercises, quizzes, assignments
- 20% midterm
- 25% class participation
- 29% final project
Project
Infrastructure: design choices are based on important criteria
AI model router: broadest coverage, local installation
API: prompt-result protocol
UX: browser-based for easy testing
Syllabus
Welcome!
The way AI eliminates empathy in our civilization is by amplifying neginformation through its algorithmic censorship choices, which dangerously triggers hundreds of unconscious cognitive biases in the human public. These are concepts introduced in Raising AI that we will explore. Here’s a video with a very famous Hollywood TV late night news comedy show host that just came out this week, illustrating the problem.
Chapter 2: Our artificial children
Trolley problems everywhere!
Overview and orientation to topics of fairness, accountability, and transparency in society, AI and machine learning, the impact of AI and automation upon labor and the job market (IEEE foundation of methodologies to guide ethical research and design) CILO-1, 5, 8
LECTURE 1
Provocation:
"The Trolley Problem", The Good Place, s02e05
Required reading:
- RAI ch2
- EAD p9-35, "From Principles to Practice", "General Principles"
Suggested materials:
The Good Place might be just a sitcom, but excellent introductory ethics books have been based on it.
Exercises:
- How should the AIs in self-driving cars make life-and-death decisions when suddenly faced with unexpected real world emergencies?
- Can AIs be trusted to make those decisions?
- Statistics show that self-driving AIs are far less likely to injure or kill people than human drivers. Is it more ethical to allow or to prohibit self-driving cars? (Notice that this dilemma is itself yet another trolley problem!)
- If a self-driving car is at fault in an accident, who is accountable? The owner of the car? The responsible human in the car? The manufacturer of the car? The maker of the AI in the car? Society at large? Nobody?
LECTURE 2
Provocation:
Required reading:
- Edmond Awad et al. (2018). “The Moral Machine experiment,” Nature 563(7729): 59–64.
Suggested materials:
The creator of the Moral Machine, Iyad Rahwan, is interviewed on my podcast.
- De Kai, host (2025). “What have machines learned about human ethics? MIT Moral Machine creator Iyad Rahwan and De Kai”. De Kai on AI, podcast, s01.
Exercises:
- Can you give logical rules to describe how a self-driving AI should make decisions?
- What criteria and objectives should a self-driving AI align to in its decision making?
- Are those culturally dependent?
- What is fairness?
- What happens to human taxi drivers and truck drivers?
Chapter 17: Lessons from the history of AI
Paradigms of AI ethics
Descriptive versus prescriptive and predictive ethics; relates classic philosophy of normative/comparative ethics and deontological/consequentialist/virtue ethics to the problem of AI ethics, and discusses why purely rule-based AI ethics will fail (IEEE goal of human rights; IEEE objective of legal frameworks) CILO-1, 2
Provocation:
Required reading:
- RAI ch17
- EAD p36-67, "Classical Ethics in A/IS"
- Fabio Morandín-Ahuerma (2023). “Twenty-three Asilomar Principles for Artificial Intelligence and the Future of Life”. Originally published in Spanish as “Veintitrés principios de Asilomar para la inteligencia artificial y el futuro de la vida. In: F. Morandín (ed.), Principios normativos para una ética de la inteligencia artificial 5–27. Concytep. By CC BY-NC-SA 4.0. https://osf.io/dgnq8/download/?format=pdf.
Suggested materials:
- Future of Life Institute (2017). “Asilomar AI Principles”. Beneficial AI. Asilomar, California. https://futureoflife.org/open-letter/ai-principles/ (retrieved 10 Feb 2025).
- Future of Life Institute (2019). Beneficial AGI. San Juan, Puerto Rico. https://futureoflife.org/event/beneficial-agi-2019/ (retrieved 10 Feb 2025).
- Christoph Salge and Daniel Polani (2017). “Empowerment As Replacement for the Three Laws of Robotics”. Frontiers in Robotics and AI 4, article 25. June.
Exercises:
- Suggest real-world examples of trolley problems where one or more of Asimov’s Laws of Robotics contradict each other.
- Suggest real-world examples of trolley problems where one or more of the Asilomar AI Principles contradict each other. https://docs.google.com/forms/d/1fbY_QAXZHz5MPRlRPRIvMKu1N0cwfiZntFimzwBMUOA/edit#responses
Preface, Afterword: The toxic AI cocktail
AI and social disruption
Deepfakes, chatbots, and drones: how AI democratizes weapons of mass destruction and disrupts civilization with information disorder and lethal autonomous weapons CILO-1, 5, 6
Provocation:
Required reading:
- RAI Preface, Afterword
- EAD p68-89, "Well-being"
Suggested materials:
PDF of the following is available at
- Caitlin Andrews. 2025. European Commission withdraws AI Liability Directive from consideration. https://iapp.org/news/a/european-commission-withdraws-ai-liability-directive-from-consideration (retrieved 12 Feb 2025).
- Saad Siddiqui, Kristy Loke, Stephen Clare, Marianne Lu, Aris Richardson, Lujain Ibrahim, Conor McGlynn, and Jeffrey Ding. 2025. Promising Topics for US–China Dialogues on AI Safety and Governance. Technical report, Oxford Martin School, University of Oxford; Safe AI Forum. https://www.oxfordmartin.ox.ac.uk/publications/promising-topics-for-us-china-dialogues-on-ai-safety-and-governance
Exercises: https://forms.gle/kwP1s5QwarPRjaip7
- Discuss how the emergence of AI might alter analyses of Carl Schmitt’s (1932) advocacy for making a “friend-enemy distinction” in The Concept of the Political.
- Contrast how a deontological rule-based AI ethics would look, assuming (a) Schmitt’s “friend-enemy distinction” should be made, versus assuming (b) Schmitt’s “friend-enemy distinction” should not be made.
- Contrast how a consequentialist AI ethics would look, assuming (a) Schmitt’s “friend-enemy distinction” should be made, versus assuming (b) Schmitt’s “friend-enemy distinction” should not be made.
- Contrast how a virtue AI ethics would look, assuming (a) Schmitt’s “friend-enemy distinction” should be made, versus assuming (b) Schmitt’s “friend-enemy distinction” should not be made.
Chapter 12: Neginformation
Willful algorithmic negligence
[Sound, informed judgment] Information disorder, misinformation, disinformation, malinformation, and neginformation; collective intelligence CILO-1, 3, 5
Provocation:
- TEDx Editor’s Pick 15 Apr 2025: “How partial truths are a threat to democracy: The dangers of ‘neginformation’” De Kai @ TEDxKlagenfurt
Required reading:
- RAI ch12
Suggested materials:
- “‘I can’t go toe to toe with social media.’ Top U.S. health official reflects, regrets.” @ Washington Post 12 Jan 2025
Exercises:
- How could the amount of neginformation in news stories be measured? (This is a difficult research question! It is well known that measuring recall is much harder than measuing precision.) Try to imagine some possible approaches.
Chapter 13: Algorithmic censorship
Misinformation theory
[Open minded diversity of opinion] Catering to the id: key challenges for social media, recommendation engines, and search engines CILO-1, 3, 5
Provocation:
- “Beware online ‘filter bubbles’” Eli Pariser @ TED, Mar 2011
- "The disastrous consequences of information disorder: AI is preying upon our unconscious cognitive biases" De Kai @ Boma COVID-19 Summit (Session 2)
Required reading:
- RAI ch13
Suggested materials:
- Eli Pariser (2011). The Filter Bubble: How the New Personalized Web Is Changing What We Read and How We Think. Penguin.
- “Echo chamber”. Wikipedia, retrieved 21 Apr 2025.
Exercises:
- Sometimes it’s suggested that people should be allowed to choose their own algorithmic censorship criteria. Given what we’ve studied about cognitive biases, what are the unintended consequences that could be dangerous?
- What percentage of the output given by a search engine or chatbot should give a human user exactly what they want (whether factually true or not), versus suggesting things the user may not have wanted but are more grounded logically and empirically?
Project planning [0723]
The Social Dilemma [0724]
Project specification
Chapter 8: Cognitive bias [0727]
Brandolini’s Law
Project specification
Deviation from rationality
LECTURE 1
Provocation:
- "Fundamental Attribution Error | Concepts Unwrapped" Ethics Unwrapped @ McCombs School of Business, University of Texas at Austin
Required reading:
- RAI ch8 (first half)
Suggested materials:
- Dugas, M. J., Hedayati, M., Karavidas, A., Buhr, K., Francis, K., & Phillips, N. A. (2005). Intolerance of Uncertainty and Information Processing: Evidence of Biased Recall and Interpretations. Cognitive Therapy and Research, 29(1), 57–70. https://doi.org/10.1007/s10608-005-1648-9
Exercises: https://forms.gle/SRxYSyiikESPRxdr6
LECTURE 2
Provocation:
- “Introduction to Behavioral Ethics | Concepts Unwrapped” Ethics Unwrapped @ McCombs School of Business, University of Texas at Austin
- “Confirmation Bias | Ethics Defined” Ethics Unwrapped @ McCombs School of Business, University of Texas at Austin
- “John Cleese Explains Dunning-Kruger Effect”, Poetry Eye, YouTube Shorts
- “Why we all fall victim to the Dunning-Kruger effect - BBC REEL”, BBC Global, Jun 2022
- “The Dunning Kruger Effect”, Sprouts, Mar 2021
- “The Irony of the Dunning-Kruger Effect”, Vallis | Video Essays, Oct 2021
Required reading:
- RAI ch8 (second half)
Suggested materials:
- “Why incompetent people think they're amazing - David Dunning” TED-Ed, Nov 2017
Exercises:
- Identify three of your own experiences where a cognitive bias caused you to make the wrong judgment, prediction, or decision.
Final project
Shared Notion:
Challenge:
Submission: Please zip your code and documentation into a single .zip folder.
Due: Aug 2026 at 11:59pm
Required texts
[RAI] Raising AI: An Essential Guide to Parenting Our Future, by De Kai. MIT Press. May 13, 2025. 978-0262049764.
[EAD] Ethically Aligned Design: A Vision for Prioritizing Human Well-being with Autonomous and Intelligent Systems (1st edition), The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems, by IEEE. 2019.
Reference material
Exercises:
Written by