Here is the English translation of the provided Chinese investment research notes.
At a Glance
Nobel laureate in Economics and author of Thinking, Fast and Slow, Daniel Kahneman, discussed with host Lex Fridman his dual-system theory of thinking, the limitations of deep learning and AI, the complexity of human cognition, and the paradox of happiness. Kahneman argues that current deep learning is essentially a powerful "System 1" (pattern matching and prediction) that lacks the reasoning, causal understanding, and meaning representation capabilities of "System 2." He views this as a fundamental limitation that is unlikely to be overcome in the short term.
Thematic Sections
1. Deep Learning is a Powerful "System 1" but Lacks the Core Capabilities of "System 2"
Kahneman argues that the current success of deep learning is essentially a victory for "System 1"—fast, automatic pattern matching and prediction. It excels at solving highly constrained problems like Go and chess but lacks the key capabilities of human "System 2": reasoning, causal understanding, and meaning representation.
- Mechanism Breakdown: Kahneman points out that deep learning systems "match patterns and predict what's going to happen, so it's highly predictive." However, "it has no reasoning ability, and it has no causality or any way of representing meaning and interacting with the real world." This means a translation system can perform well, but "they really don't know what they're talking about."
- Historical Context and Comparison: Kahneman contrasts the progress of deep learning with human learning. He notes that humans (especially children) learn very quickly, "not needing a million examples, just two or three." Machines, however, require massive amounts of data, revealing a fundamental difference. He believes that to achieve fast learning, some form of "expectation" or "structure" must be built into the machine, a problem that remains unsolved.
- Extrapolation and Uncertainty: Kahneman explicitly states that he "agrees with most people" that "the capabilities of neural networks will hit a limit." He disagrees with Yann LeCun's view that current architectures can eventually achieve causal reasoning. He argues that to truly understand the world, AI needs a "perceptual system" and "contact with the real world," i.e., "grounding." Otherwise, it is just "a machine that doesn't know what it's talking about."
2. The Paradox of Human-Machine Collaboration: When Machines Are Smart Enough, Humans Become Redundant
Kahneman presents a profound paradox regarding human-machine collaboration: in a system, if a machine is advanced enough to truly help a human, it may soon no longer need the human at all. This view challenges the long-standing vision of "semi-autonomous systems" or "human-machine collaboration."
- Mechanism Breakdown: Using chess as an example, he notes that some once believed a "human-machine combination" would be unbeatable, but ultimately, systems like AlphaZero needed no human input. The core issue is that a machine needs the ability to "recognize when it is in a situation it cannot resolve" and "call a human." Kahneman believes "it's very hard to do that without understanding." To identify all possible failure scenarios, the machine would almost need to be smart enough to solve all problems itself.
- Data Chain and Analogy: He cites the evolution of chess as an analogy and asks, "How many problems are like chess? How many problems are not like chess?" His judgment is that "every problem might eventually be like chess," with the only difference being the length of the transition period.
- Falsification Condition: Kahneman's assertion would be falsified if a future system could reliably identify and hand off complex problems beyond its capability without possessing general understanding.
3. The "Dance" of Autonomous Driving: Complexity Far Beyond Intuition, Requires Understanding, Not Just Prediction
Kahneman believes the difficulty of autonomous driving (especially interaction with pedestrians) is severely underestimated because it involves a complex social interaction requiring "understanding," not just "prediction." This starkly contrasts with the public's intuition that "driving is simple."
- Mechanism Breakdown: Kahneman describes the "dance" between pedestrians and drivers: a pedestrian will look at the driver's eyes and then "look away before stepping into the street" to signal "I am committed to crossing." He argues that machines need to understand this "commitment" and the game-theoretic "game of chicken." He questions whether "prediction" alone (like AlphaGo's approach) is sufficient, as "prediction" can exist without "understanding," but "understanding" seems necessary for handling these subtle social signals.
- Data Chain and Comparison: He points out that while Go is "infinitely complex," it is "very constrained." The real world, however, "is much less constrained, with many more potential surprises." This is the root of the problem. He criticizes the public's intuition, which is to "assess the complexity of a problem based on how difficult it is for them to solve," which has almost no relation to how difficult it is for AI to solve.
- Extrapolation: Kahneman believes this difficulty with "human-machine collaboration" is a general phenomenon. "Almost every robot-human collaboration system is much harder than people realize." This implies a significant challenge for the real-world application of Artificial General Intelligence (AGI).
4. The Gulf Between the Experiencing Self and the Remembering Self: We Live for Memories, Not for Experiences
Kahneman elaborates on his famous paradox of happiness: there exists an "experiencing self" (living in the moment) and a "remembering self" (constructing a story about the experience), and our decisions and happiness are primarily driven by the latter. This leads to a profound contradiction: time is the currency of life, but in our evaluative memory, time is barely represented.
- Mechanism Breakdown: The remembering self is "schematic"; it constructs a story about an event where "time doesn't matter," but rather "the peak and the end of the event" matter. This leads to "duration neglect" and the "peak-end rule."
- Data Chain and Thought Experiment: Kahneman uses a thought experiment to illustrate this: "Suppose you are planning a vacation, but you are told that after the vacation you will take an amnesia drug and remember nothing... Would you still go on the same vacation?" He believes the answer is likely no, because "we go on vacation, to a large extent, to build memories, not to have experiences."
- Extrapolation and Uncertainty: Kahneman admits he cannot resolve this contradiction and consequently "gave up on happiness research." He notes that making the remembering self happy and making the experiencing self happy are different things. He observes that modern social media (like Instagram) has greatly "amplified" the remembering self, with people "living to take pictures," which is changing the way we experience the world.
5. The Root of Psychology's Replicability Crisis: The Researcher's Intuitive "Focusing Illusion"
Kahneman offers an original explanation for psychology's replicability crisis: the problem lies primarily with "between-subject experiments," because researchers' intuitions are systematically biased. He believes most psychological hypotheses are correct in direction, but the effect sizes are much weaker than researchers imagine.
- Mechanism Breakdown: Kahneman distinguishes between "within-subject experiments" (the same person experiences both conditions) and "between-subject experiments" (different people experience different conditions). He points out that researchers, as designers, naturally have a "within-subject" perspective and can intuitively feel the difference between the two conditions. However, in "between-subject" experiments, subjects only experience one condition, so the effect size is greatly diminished. The researcher's intuition is misled by this "focusing illusion," leading them to overestimate the power of the manipulation.
- Data Chain: He cites a striking example: "I recently heard that some friends funded 53 studies on behavior change... the goal was to precisely change how often people went to the gym... The success rate was zero. Not one of the 53 studies succeeded." He emphasizes that these were "the best people in the field," but they "were completely uncalibrated in their judgments."
- Extrapolation: The solution is to "reduce trust in intuition," pre-register experiments, and significantly increase sample sizes. He believes platforms like MTurk are changing psychology by making large-scale, high-statistical-power experiments cheap and feasible.
Position Moves
This section is a theoretical discussion and does not involve specific investable targets.
Judgments Worth Remembering
1. Deep learning is a powerful "System 1" but lacks the core capabilities of "System 2" (Daniel Kahneman): Current AI excels at pattern matching and prediction but lacks reasoning, causal understanding, and meaning representation, which is its fundamental limitation.
2. There is a paradox in human-machine collaboration: when machines are smart enough, humans become redundant (Daniel Kahneman): If a machine can recognize problems it cannot solve and call a human, it may already be smart enough to solve all problems itself.
3. The difficulty of autonomous driving is severely underestimated because it involves social interactions requiring "understanding" (Daniel Kahneman): The "dance" between pedestrians and drivers (eye contact, commitment signals) is far more complex than Go because the real world has fewer constraints and more surprises.
4. We live for memories, not for experiences (Daniel Kahneman): Decisions are driven by the "remembering self," which constructs stories and ignores duration, leading to the sacrifice of the "experiencing self's" well-being.
5. Psychology's replicability crisis stems from researchers' "focusing illusion" (Daniel Kahneman): Researchers overestimate the effect size of experimental manipulations in "between-subject designs" because their intuition comes from a "within-subject" perspective.
6. Most psychological hypotheses are correct in direction, but the effect sizes are extremely weak (Daniel Kahneman): An example is that 53 intervention studies designed by top teams to change gym-going behavior all failed.
7. Changing deeply held beliefs relies not on evidence, but on trusted leaders changing the narrative (Daniel Kahneman): People hold certain views more because they trust the group leader telling the story than because of their own assessment of the evidence.
8. The "why" question is unsolvable for humans (Daniel Kahneman): We can understand "how" (e.g., gravitational waves), but the "why" of life's meaning is beyond our comprehension.