How old is AI? Is it an omnipotent prodigy, or an “artificial intellectual retardation” full of loopholes? This article condenses the development of AI for more than 60 years into a growth story from fetus to adolescent: fetal age only hardcodes, infancy relies on feeding data to learn, childhood can read pictures and read, and today’s “teenagers” can write poetry and draw but also rebel against illusions.
Do you often hear two very different voices about AI?
- AI extreme worship: Some people think that AI is about to become a god, able to write novels, paint famous paintings, solve problems, and even worry that one day it will wake up and take over humanity.
- AI is extremely contemptuous: Another group of people think that AI is an “advanced search tool”, or simply ridicule it as an “artificial intellectual retardation”.
The root cause is everyone’s lack of understanding of the whole picture of AI:
- Fear stems from the unknown:If you don’t know what AI can do, what it can’t do, and where the boundaries are, it’s easy to think about the worst.
- Contempt stems from ignorance:I haven’t seen the powerful capabilities that AI has shown in specific fields and the momentum of its rapid progress.
Therefore, it is very important to understand the full picture of AI from a macro level.
With this macro perspective:
- Can eliminate blind emotions:Don’t be blindly afraid, don’t despise it, and have an objective and rational understanding of AI.
- Understanding Ability Boundaries:Know where AI can help a lot and where it may fall off the chain.
- Lay a good foundation:Pave the way for subsequent understanding of specific technologies. (In the next article, we will dismantle the “body structure” of AI!) ）
For product people, understanding the development stage and capability boundaries of AI is a key prerequisite for judging technical feasibility, designing reasonable human-computer interaction, and avoiding technical risks (such as hallucinations).
1. History of AI Development ≈ A condensed history of human growth
The best way to understand AI development may be hidden within ourselves.The evolutionary trajectory of AI from its birth to the present is much like the growth process of a person from fetus to adolescence.
B-end product manager’s ability model and learning improvement
The first challenge faced by B-end product managers is how to correctly analyze and diagnose business problems. This is also the most difficult part, product design knowledge is basically not helpful for this part of the work, if you want to do a good job in business analysis and diagnosis, you must have a solid …
View details >
Why is this analogy useful?
- Intuitive and easy to understand: we are all familiar with the stages of human growth.
- Reveal the essence: The improvement of AI capabilities has deep similarities with the way humans learn to cognition and interact with the world.
- Clear positioning: You can immediately understand how old AI is and what level it is now.
Below, we will use this “growth perspective” to break down the development stage and capabilities of AI step by step.
Please note that this analogy is intended to help us intuitively understand the evolutionary stages and characteristics of AI capabilities, but it is not rigorous at some levels, so if you have other ideas, please feel free to discuss them together.
2. The “life” stage and corresponding capabilities of AI development
Stage 1: Fetal stage (foundational stage: circa 1950s-1980s) – “hard-coded embryo”
Human analogy: Just like in the mother’s belly, the basic body structure (brain, limb prototype) has grown, but it is not yet able to independently perceive and learn from the outside world.
AI Capability Features:
Core driven: rules first. AI relies entirely on “hard-coded” rules written by programmers, and instructions dictate what to do when it dies.
What can you do? Deal with some problems with extremely clear rules and extremely narrow scope. For example, playing chess (with fixed rules) and playing simple logical reasoning games.
Where are the shortcomings?
- Extremely dependent on manual labor: Every time a little new knowledge is added, programmers have to manually write new rules.
- Zero learning ability: Playing chess today is like this, ten years later, it will not improve on its own.
- No flexibility: In the face of situations where the rules are not covered, it freezes immediately, and the knowledge is very narrow.
Stage 2: Infancy (Perceptual Learning Stage: circa 1990s-2010s) – “Learning to See and Listen for a Baby”
Human analogy: Babies begin to see with their eyes, listen with their ears, touch with their little hands, recognize their mother’s face, understand their own names, imitate adult babbling, and recognize simple patterns (for example, ringing has a sound).
AI Capability Features:
Core driven: data feeding. Machine learning, especially statistical learning, takes protagonists. AI began to explore patterns and patterns on its own from a large amount of data.
What can you do?
- Pattern recognition:Can recognize whether it is a cat in the photo (image classification), and can understand you say “turn on the light” (voice recognition).
- Simple prediction:Determine if an email is spam.
- Basic recommendation:Based on what you have bought, recommend a similar one (early e-commerce recommendation).
Where are the shortcomings?
- Data hunger + dependency tags:Do you want to learn to recognize cats? First give me tens of thousands of photos manually marked “This is a cat”! If you don’t mark the data, you will be blind. High cost and low efficiency.
- The understanding is very shallow:I know that there is a “cat” in the picture, but I don’t know what a cat is, why it meows, and what is the difference between it and a dog. I know it’s like this, but I don’t know why.
- Gullibility:If you change the picture a little (for example, add some noise that is not visible to the human eye), it may recognize the cat as a dog.
- Generalization difference:Only perform well in similar scenarios that have been trained, and change the angle, light, or background, you may be confused.
Stage 3: Childhood (Understanding and Interaction Stage: circa mid-2010s – present) – “Talking and Interactive Children”
Human analogy: 3-year-old children have explosive language skills, can understand more complex instructions (such as putting red blocks under the blue box), can have simple conversations, begin to think logically, and can interact and play with adults.
AI Capability Features:
Core Drive: Deep Learning + Big Data + Big Computing Power. Neural networks, especially the Transformer architecture, a new way of processing information, shine.
What can you do? (Ability explosion)
1) Natural Language Processing (NLP):
- The quality of machine translation has improved by leaps and bounds, and it is no longer a “word-to-word” mechanical translation;
- intelligent customer service robots can handle many common problems;
- Text generation is starting to appear, but it’s not stable enough.
2) Computer Vision (CV):
- High-precision recognition of objects, pedestrians, and vehicles has begun to become the eyes of autonomous driving.
- Face recognition is popular in mobile phone unlocking, security and other fields.
3) Interactive Capabilities: Intelligent voice assistants, such as Apple Siri and Xiaomi Xiaoai, have become popular.
Where are the shortcomings?
- Understanding Off-Surface:It mainly relies on statistical associations in massive texts, rather than real “understanding”. For example, we know that “cats eat fish” often appear together, but we may not understand the biological chain relationship behind it.
- Lack of common sense and deep reasoning:It is difficult to deal with problems that require common sense of life or complex logical chains. For example, the obviously unreasonable phenomenon of “elephant in the refrigerator”.
- The output is unstable, and the “hallucination” begins to appear:Sometimes the answers are good, sometimes they talk nonsense, and even make up content that seems reasonable but is actually wrong, showing signs of “hallucinations”.
Stage 4: Adolescence (Exploration and Creation Stage, Current Frontier Research) – “Rebellious Teenager Who Loves to Think and Create”
Human analogy: Adolescents develop abstract thinking, creativity, complex reasoning skills rapidly and begin to explore themselves (“Who am I?). ), explore the world (“Why is this so?”) “), try to create independently (write poetry, compose, invent), but may also be wild, impulsive, and rebellious against rules.
AI capability characteristics (represented by large language model LLMs/generative AI):
Core Driver: Massive unannotated data pre-training + fine-tuning/prompt engineering. “Emergent ability” appears, the model is large to a certain extent, and suddenly learns some tasks that have not been specially trained.
What can you do?
1) Generation and Creation:
- Write fluent articles, stories, and even poems, code, scripts.
- Generate realistic images (DALL-E, Midjourney) and music based on text descriptions.
2) Complex Understanding and Reasoning:
- Capable of handling longer conversations or documents (with longer contexts).
- Perform a certain level of logical reasoning, problem-solving, code debugging, and knowledge quizzing (such as ChatGPT, Claude, Gemini).
3) Multimodal fusion: Start trying to break through the boundaries of text, images, and sounds (for example, GPT-4V can read pictures and speak).
Where are the shortcomings?
- Logical rigor and deep reasoning are still insufficient:Solving complex math problems and making rigorous legal arguments is prone to errors or skipping steps.
- Factual accuracy is difficult to guarantee:The problem of “hallucination” is obvious, and it will confidently fabricate information, quotes, data that do not exist.
- Lack of true “understanding” and awareness:It can retell knowledge and relate information, but there is still a gap between human beings and deep understanding and integration of concepts. Knowing that it is so, partly knowing why it is so, but it is not “true knowledge”.
- Ethics, safety, and bias issues highlighted:It may output harmful and discriminatory content, and the risk of data privacy, copyright disputes, and misuse is huge.
Stage 5: Adulthood (Strong AI/AGI: Future Goals) – “Mature and All-Powerful Adult”
Human analogy: Adults have mature comprehensive abilities in cognition, learning, reasoning, creativity, emotional understanding, independent decision-making and solving various complex problems, and can live independently, adapt to new environments, and continue to learn and grow.
AI Capability Characteristics (Vision, Not yet realized):
Core goal: general intelligence. to reach or surpass the human level in a wide range of or even all tasks that human beings can do; can independently learn new skills, adapt to unknown environments, and take the initiative to innovate like humans.
Ability to imagine:
- True understanding: deeply grasp the essence and connection of concepts;
- Deep reasoning: Analyze and solve problems like a scientist;
- Cross-domain knowledge transfer: flexibly apply the learned knowledge to completely different fields;
- Set goals independently and solve them efficiently: proactively discover problems, plan paths, and achieve goals;
- Perhaps self-awareness, emotional understanding, and interaction (highly controversial)?
Key challenges (far from solved):
- How to achieve it?Can the current deep learning path lead to AGI? Is a new theory needed?
- How to ensure safety and control?With such a powerful system, what if the target conflicts with humanity?
- How is the ethical framework constructed?How are rights, responsibilities, and social impact defined?
3. The current situation of AI and our cognition
Overall, the development trajectory of AI is from relying entirely on artificial rules (fetuses), to learning to perceive patterns from data (babies), to mastering language understanding and basic interaction (children), and now it is showing strong potential for exploration and creativity (adolescents). With each step, the ability is significantly improved.
We are in the “adolescence” of AI.
AI at this stage:
- Amazing creativity: Writing and arranging music, with unlimited potential, really changed the way of working and creating.
- Great potential: the future is promising, and the development speed is super fast.
- But it is still “immature”: in terms of deep understanding, logical rigor, factual reliability, and sense of autonomy, it is still far from true maturity. It is like a smart teenager, talented but can also make mistakes, rebel, and need guidance.
Therefore, our most rational attitude towards AI is:
- Rejection of Apotheosis/Fear:You don’t have to think that it is about to rule the world just because it can write poetry and paint. It has no consciousness, boundaries in its capabilities, and relies heavily on data and human design.
- Abandon contempt:Don’t completely deny its value and the amazing achievements it has achieved just because it is still “nonsense”. It is far more efficient than humans at specific tasks.
- Understand the boundaries and use prudence:See clearly what “growth stage” it is in now, understand what it can and cannot do, and where it is prone to mistakes (such as beware of “hallucinations”).
Embrace the creativity of AI, face up to the limitations of its “adolescence”, and be a rational driver who makes good use of tools and distinguishes between truth and falsehood.
Now that we know how old AI is, how does it work? What does it rely on to “see”, what does it rely on to “think”, and what does it rely on to “move”?
Just as the human body needs bone support, muscle movement, brain thinking, nerve conduction, and sensory input to work together, the operation of AI, a “living form”, also relies on a set of complex “hardware organs” and “software systems” to coordinate precisely.
Next content: The mystery of AI’s “body”, how hardware, software and technology work together to “breed” intelligence
