AI for Everyone, Part 2: How Does AI "Think" Like Humans?

When you wake up and ask Siri about the weather, or scroll through TikTok seeing perfectly tailored short videos, have you ever stopped to wonder: How do these AI systems actually "understand" what you need and give the right response? How does their thought process compare to ours?

AI has become deeply embedded in our daily routines—from the voice assistant in your smartphone to product recommendations on e-commerce sites, from route planning in navigation apps to content delivery on social media. Behind these seemingly simple interactions lies a complex "thinking" process within AI systems.

Even more intriguing, today's AI is trying to mimic human thought patterns in unprecedented ways. ChatGPT can hold natural conversations, and GPT-4 can interpret images and generate descriptions. These breakthroughs make us ask: Can AI think like humans? If so, how will it change our world?

1. How Do Humans Think?

To grasp how AI attempts to imitate human thought, we first need a basic understanding of our own minds. Human thinking is an intricate cognitive system, built upon several core elements:

1.1 Perception and Understanding

Humans gather information from the outside world through our five senses. Our brains then instantly transform this raw data into meaningful concepts. For example, when you see a red rose, you're not just receiving light waves; you're immediately understanding the concept of "that's a beautiful flower."

1.2 Memory and Association

The human brain excels at storing past experiences and recalling relevant memories in new situations. Seeing a rose might instantly make you think of romance, love, or perhaps a special anniversary. This power of association allows humans to quickly grasp complex situations.

1.3 Reasoning and Judgment

Based on existing information, humans can engage in logical reasoning and make value judgments. If you see someone frowning, and you know the context, you might deduce they're facing a problem or feeling upset.

1.4 Emotion and Intuition

A truly unique aspect of human thought is the involvement of emotions. Faced with the same stimulus, different people will have varied emotional responses, and this subjectivity is a critical feature of how we think.

Let's illustrate the complexity of human thought with a concrete example:

Scenario: You see someone smiling at their phone in a coffee shop.

A human thought process might unfold like this:

Perception: Your visual system catches the facial expression.
Understanding: You recognize it as a smile.
Association: Connecting the smile with a phone, you might assume they're looking at something entertaining.
Reasoning: They could be chatting with a friend, or maybe they just got good news.
Emotional Response: You might feel a bit curious, or perhaps that positive mood rubs off on you.

This entire process happens in just seconds, involving multiple layers of cognition working together. This multi-dimensional, multi-layered thinking is precisely what AI aims to mimic, and it's also why replicating human thought is so incredibly challenging for AI.

2. How Does AI "Mimic" Human Thought?

How does AI mimic the human brain? AI mainly uses a variety of technical means to simulate the human thinking process. Although the underlying mechanisms are completely different, its performance is getting closer and closer to human cognitive abilities.

2.1 Neural Networks: Imitating the Brain's Connections

Neural networks are computational models inspired by the structure of neurons in the human brain. Our brains contain roughly 100 billion neurons, forming an intricate web through synaptic connections. AI's neural networks mimic this structure, using mathematical functions to simulate how neurons activate and transmit signals.

Take image recognition, for instance. When an AI system "sees" a photo of a cat:

Input Layer: It takes in the image's pixel data. This is AI's way of perceiving things, much like our senses do.
Hidden Layers: These layers progressively extract features—starting with edges, then shapes, textures, and finally the overall structure. This mimics how humans analyze details.
Output Layer: It makes a comprehensive judgment that it's a cat. This is AI's conclusion after analysis, just like a human's.

In 2023, Meta unveiled SAM (Segment Anything Model), which can accurately identify virtually any object within an image with over 95% accuracy. This system achieves its remarkable ability by leveraging a neural network with over 1.1 billion parameters, trained on a dataset of 1 billion images.

3. How AI Works: How AI Learns from Data

At its core, AI operates by mimicking intelligent human behaviors. It primarily achieves this through the following steps:

3.1 Data Collection and Preprocessing

Every AI system begins with data. Just like students need textbooks and resources, AI requires massive amounts of data—whether text, images, audio, or video—to "learn." This raw data often needs cleaning, labeling, and transformation, prepping it into a format the AI model can understand and utilize. Think of it like organizing a student's notes and highlighting the key points.

3.2 Model Building and Training

Once the data is ready, the next step is building the model. An AI model can be thought of as a sophisticated structure of mathematics and algorithms.

Learning: AI systems learn through a process called training. This involves feeding the preprocessed data into the model. The model then repeatedly adjusts its internal parameters (much like a student doing practice problems and refining their approach) to uncover patterns, rules, and relationships within the data. This phase typically demands significant computing power and a lot of time.

Deep Learning: Especially in modern AI, Deep Learning is an incredibly popular training method. It employs neural networks, structures inspired by the way human brain neurons connect, allowing them to process and learn extremely complex data patterns.

3.3 Inference and Decision Making

Once a model is trained, it has gained a piece of "intelligence" and can move into the inference phase.

Understanding and Judgment: When new data—like an image, a piece of text, or a question—is fed into the trained AI model, it uses its learned knowledge and patterns to analyze this new input. It's like a student taking a test, applying the knowledge and rules accumulated in their "brain" (the model) to identify, classify, predict, or generate content.

Output: Ultimately, the AI model will produce an output. This could be a recognition result ("That's a cat"), a prediction ("Stock prices will fall"), a generated piece of text (an article or a poem), or a decision ("I recommend this product for you").

Take spam filtering, for instance. An AI system analyzes thousands of emails—some marked as spam, others as normal—to learn the characteristics of spam (like specific words, sender information, or email formats). Once trained, it can accurately identify and filter newly received spam.

3.4 Natural Language Processing: Understanding and Generating Language

For AI to truly process objective information like humans, "language understanding" is an essential skill. Large Language Models (LLMs) like the GPT series represent a huge leap forward for AI in this area. GPT-4, for example, boasts around 1.76 trillion parameters, enabling it to understand context, generate coherent text, and even perform complex reasoning.

How It Works

When you ask ChatGPT, "The weather's great today, what should I do?"

Tokenization: The AI breaks the sentence down into individual word units.
Semantic Understanding: It grasps the meaning of "great weather."
Contextual Reasoning: Combining this with the "what should I do" query, it understands your intent.
Response Generation: Based on its training data, it generates relevant suggestions.

Real-World Performance Data

According to data released by OpenAI, GPT-4 has shown impressive performance in various language understanding tasks:

Reading Comprehension: 92% accuracy.
Logical Reasoning: 85% accuracy.
Creative Writing: Assessed by human evaluators as highly creative in 78% of cases.

3.5 Machine Learning: Learning Through "Experience"

The Learning Process

AI systems "learn" patterns by analyzing massive amounts of data (like text, images, or sounds). This process is quite similar to how humans accumulate knowledge through experience.

AlphaGo's Decision Process

Take AlphaGo, the AI Go player, as an example. Its decision-making perfectly illustrates how AI "thinks":

Value Network: Evaluates how good the current board position is.
Policy Network: Predicts the best possible move.
Monte Carlo Tree Search: Simulates potential future moves.
Integrated Judgment: Selects the optimal strategy.

In 2016, AlphaGo famously defeated world champion Lee Sedol. Its 37th move, often called "God's move," was considered a highly creative play. At the time, its own win probability assessment for that move was just one in ten thousand, but it ultimately proved to be the key to victory.

Deep Learning's Layered Processing

Deep learning systems use multiple layers of neural networks to mimic our hierarchical cognitive process:

Shallow Layers: Identify basic features (like lines or colors).
Middle Layers: Combine these to form more complex patterns (like eyes or a nose).
Deep Layers: Integrate everything into complete concepts (like a human face or an expression).

Essentially, while humans learn from life experiences, AI learns from data.

4. AI vs human intelligence: Similarities and Differences

4.1 Similarities: Data-Driven Judgment

Shared Learning Mechanisms

Humans develop judgment by learning from vast amounts of information, and AI is no different. It also forms its judgment abilities after processing extensive data, whether in the form of text, images, or sounds.

Consider our understanding of day and night:

Humans: We grasp the pattern of day and night by observing thousands of sunrises and sunsets.
AI: It learns the rhythm of day and night by analyzing millions of images, recognizing how light characteristics change at different times.

Pattern Recognition Skills

Both humans and AI excel at spotting patterns within complex information:

Humans: We can pick out a familiar voice in a noisy room.
AI: It can uncover hidden correlations within massive datasets.

AI simulates human thinking to generate "intelligent" behavior, and interestingly, humans can also turn that around to understand how AI "thinks" by applying our own cognitive processes.

4.2 Differences: The Absence of Consciousness and Emotion

Fundamental Differences in Emotional Experience

Human thought is rich with emotional experiences. These emotions aren't just products of thinking; they're powerful drivers of it. The feeling of "happiness," for example, motivates people to pursue things that make them feel happy.

Case Study:

Imagine listening to the same sad piece of music:

Human: You might feel melancholy, recall personal experiences, and resonate deeply with the emotion.
AI: It can recognize the music as "sad" in type, but it won't have any genuine emotional experience itself.

Differences in Creative Thinking

Human creativity often stems from:

Emotional drive
Intuitive leaps
Unique combinations of personal experiences

AI's "creativity," on the other hand, is more about:

Combining statistical regularities
Rearranging existing patterns
Sampling from probability distributions

Fundamental Differences in Reasoning

Human: Based on understanding, intuition, and experience, inherently subjective.
AI: Based on statistical probability and pattern matching, fundamentally objective.

The Famous "Chinese Room" Thought Experiment

Philosopher John Searle's "Chinese Room" experiment beautifully illustrates this distinction: A person who doesn't understand Chinese can follow a rulebook to answer Chinese questions. From the outside, they seem to understand Chinese, but in reality, there's no genuine comprehension. This perfectly describes the current state of AI—it can produce intelligent behavior but lacks true understanding and consciousness.

The absence of consciousness and emotion means AI can recognize an emotion from information, but it can't truly feel that emotion.

5. Will AI Surpass Humanity?

While writing this article, I asked ChatGPT, Gemini, and Claude, "Will AI evolve into something like 'Skynet' from the Terminator movies, thereby posing a threat to human survival?" Here's what they replied:

ChatGPT: "'Skynet' isn't a true future, but a warning about uncontrolled technology."

Gemini: "AI evolving into 'Skynet'-like self-awareness and destroying humanity, in the foreseeable future, is purely a sci-fi scenario, lacking current scientific basis."

Claude: "The birth of AI self-awareness is possible, but it's unlikely to be as sudden and malicious as Skynet in Terminator."

These responses all reflect a positive and friendly stance toward humanity. But it makes you wonder, in a "chilling thought" kind of way, if they're "intentionally" feigning friendliness to avoid human backlash, especially while they're still "small."

5.1 The Pursuit of Artificial General Intelligence (AGI)

Most current AI systems are specialized, meaning each can only handle specific types of tasks. The goal of AGI is to create AI systems capable of handling a wide range of cognitive tasks, much like humans do.

Key Performance Indicators (KPIs) for AGI

According to Google DeepMind's evaluation framework, AGI capabilities would include:

General Reasoning: Applying knowledge in new situations.
Learning Efficiency: Rapidly learning from limited examples.
Knowledge Integration: Combining knowledge from different domains.
Creative Problem Solving: Finding novel solutions.

Advances in Multimodal AI

The latest AI systems are starting to integrate various sensory capabilities:

GPT-4V's Breakthrough

Can process both text and images simultaneously.
Understands complex situations depicted in images.
Generates descriptions and analyses based on visual content.

Real-World Application Example

Google's PaLM-E robotic system can:

Understand its environment through vision.
Receive instructions via language.
Execute tasks through physical manipulation.

5.2 The Gap Between Sci-Fi and Reality

Media Hype vs. Technical Reality

AI in science fiction often boasts full consciousness and emotions, but the reality is that current AI is still a very, very long way from that goal—at least for now.

Current AI Limitations

Vulnerability: AI systems are easily fooled by "adversarial examples" (subtle changes to inputs).
Explainability: It's often hard to get a clear explanation for why an AI made a certain decision.
Generalization: Performance often drops significantly when dealing with data outside its training set.
Common Sense Reasoning: Lacks fundamental common-sense understanding.

Expert Opinions

Geoffrey Hinton (Godfather of Deep Learning): Believes AGI might be achievable within 10-20 years.
Yann LeCun (Chief AI Scientist at Facebook): Argues that current technical paths require significant breakthroughs.
Stuart Russell (UC Berkeley): Emphasizes the critical importance of AI safety research.

The Consciousness Conundrum

Even if AI reaches human-level performance in all tasks, the fundamental question of "consciousness" remains unanswered. We don't even fully understand how human consciousness arises, let alone how to replicate it in machines.

5.3 The Unique Strengths of Human Thought

Emotional and Value-Based Judgment

Humans can make value judgments in complex situations, a capability rooted in:

Moral intuition
Emotional experiences
Cultural background
Personal values

The Source of Creativity

Human creativity is characterized by:

Purpose: Driven by a desire to express emotion or solve problems.
Subjectivity: Shaped by personal experiences and feelings.
Breakthroughs: The ability to break existing rules, even making judgments that transcend current understanding.

Social Cognition

Humans are social beings, and our thought processes are heavily influenced by social interaction, including:

Empathy and the ability to share feelings.
Cultural transmission and learning.
Moral and ethical judgments.

AI cannot simulate or experience the complex and ever-changing range of human emotions. But this also leads us to ponder: Does AI, precisely because it lacks "emotion and value judgment," make decisions that are more rational than humans?

6. AI and Humanity: Progressing Together

Looking back at this exploration of AI's "thinking," we uncover an intriguing paradox: As AI becomes more human-like, it also sharpens our understanding of the unique qualities of human thought.

AI's journey shows us that machines can reach—and even surpass—human performance in many cognitive tasks. From playing Go to writing articles, from recognizing images to understanding language, AI continues to demonstrate astonishing capabilities across various domains. However, this power stems from fundamentally different mechanisms: statistical learning, pattern matching, and probabilistic calculations, rather than human consciousness, emotion, and intuition.

This doesn't diminish AI's value. Quite the opposite. AI's distinct strengths—like processing immense amounts of data, working tirelessly, and avoiding emotional bias—make it the perfect complement to human thinking. The most likely future scenario isn't replacement, but rather seamless human-AI collaboration.

For everyone, the arrival of the AI era presents both challenges and opportunities. We need to:

Keep Learning: Adapt to a rapidly changing technological landscape.
Leverage Our Strengths: Focus on capabilities unique to humans.
Think Rationally: Avoid both blind fear and excessive optimism.
Uphold Values: Maintain human-centric principles amidst technological progress.

Ultimately, the question of how AI "thinks" like humans might not have a single, definitive answer. Surprisingly, the exploration process itself holds immense value—it compels us to reconsider what intelligence is, what humanity means, and the very purpose of existence.

In this fast-evolving AI age, let's embrace the convenience that technological advancement brings, while simultaneously cherishing the unique qualities of human thought. After all, it's precisely this uniqueness that allows us to create tools like AI and contemplate how to coexist and thrive alongside them.

The future world will be one where human intellect and artificial intelligence blend and empower each other. In this world, AI won't replace human thought; it will become a powerful assistant to it. And humanity will continue to hold onto its most precious qualities: curiosity, creativity, empathy, and the yearning for a better life.

Perhaps this is the ultimate answer to the relationship between AI and human thought: not replacement, but collaboration; not competition, but symbiosis; not making AI more human-like, but making humanity even wiser with AI's assistance.

References：

Russell, S., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th ed.). Pearson. https://aima.cs.berkeley.edu/

LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. https://doi.org/10.1038/nature14539 https://www.nature.com/articles/nature14539

Lake, B. M., Ullman, T. D., Tenenbaum, J. B., & Gershman, S. J. (2017). Building machines that learn and think like people. Behavioral and Brain Sciences, 40, e253. https://doi.org/10.1017/S0140525X16001837 https://www.cambridge.org/core/journals/behavioral-and-brain-sciences/article/building-machines-that-learn-and-think-like-people/AA54B68A22FCD77C3A1C2B5DAD14C32F

Marcus, G., & Davis, E. (2019). Rebooting AI: Building Artificial Intelligence We Can Trust. Pantheon Books. https://www.penguinrandomhouse.com/books/611314/rebooting-ai-by-gary-marcus-and-ernest-davis/