After reviewing books and consulting ChatGPT, I have completed this blog post, which is original content; please understand the difficulties of the creator. If you need to reprint, please include the source of this article so that I can continue to output original and high-quality content.
1. What is Artificial Intelligence#
Artificial intelligence, unlike natural life intelligence, is machine intelligence designed to imitate human intelligence.
Artificial intelligence can be divided into weak artificial intelligence and strong artificial intelligence. The difference between the two is that strong artificial intelligence has autonomous consciousness, while weak artificial intelligence only simulates certain aspects of human intelligence. Even though OpenAI's ChatGPT, Claude, and Google's Bard are developing rapidly, according to whether machine intelligence has autonomous consciousness, we are still in the era of weak artificial intelligence. (For the distinction between strong and weak artificial intelligence, refer to: Mo Hongwei: "Artificial Intelligence", 2020, People's Posts and Telecommunications Press, Chapter 1)
2. Three Schools of Thought on Artificial Intelligence#
Different thinkers and experts have different views on the development of artificial intelligence. There are three main schools of thought, and I will list their main representatives and their presence in modern technology companies.
- Symbolic AI
The symbolic AI school emphasizes the representation and processing of symbols and rules. It believes that intelligent behavior can be built through a system of symbols and logic composed of axioms and logical systems. It emphasizes the logic and reasoning between symbols and axioms. Representative experts of symbolic AI include Marvin Minsky, Allen Newell, and Herbert A. Simon, who are pioneers of symbolic AI. Marvin Minsky designed and built the first self-learning artificial neural network machine, while Allen Newell and Herbert A. Simon developed early symbolic AI systems such as Logic Theorist and General Problem Solver. Expert systems are the main achievements of symbolic AI.
In today's technology companies, early IBM was influenced by symbolic AI to some extent in computer development.
- Connectionism
The connectionism school emphasizes the imitation of human brain neurons and focuses on simulating the connections between neurons to achieve artificial intelligence using the connection mechanism of neural networks. It emphasizes the adjustment of weights during the learning process.
Representative experts of connectionism include David Rumelhart, Geoffrey Hinton, John Hopfield, Terrence Sejnowski, Herbert A. Simon, Fei-Fei Li, Andrew Ng, Allen Newell, Yann LeCun, etc. Hopfield's neural network model is the premise and foundation of deep learning. Among them, Hinton is one of the leaders of the connectionism movement and has become one of the most influential figures in the field of neural networks, pioneering deep learning.
- Behaviorism
The behaviorism school attempts to replicate and initiate human intelligence in machine intelligence starting from the simulation of animal "perception-action". They believe that intelligence depends on perception, behavior, and the adaptive ability of the machine subject to the external environment. Behaviorism focuses on observing and measuring observable behavior and emphasizes modeling external stimuli and responses. The main contribution of behaviorism is reflected in robot control systems.
The main experts of behaviorism include B.F. Skinner, who proposed a learning theory based on rewards and punishments. These theories constitute the theoretical basis of behaviorism in artificial intelligence.
Behaviorism is relatively small in the field of artificial intelligence. In some machine learning applications, reward-driven learning still has a certain influence. Currently, behaviorism is mainly manifested in practical technologies, such as robots used for firefighting and Boston Dynamics' robot dogs.
3. Connectionism and Symbolic AI in GPTs#
GPTs, represented by OpenAI's ChatGPT service, are gradually integrating into people's lives and work. These popular GPTs are mainly influenced by connectionism and symbolic AI, with a larger component of connectionism.
ChatGPT embodies the connectionist neural network thinking in its underlying structure and applies symbolic and semantic theory knowledge of symbolic AI through pre-training on large-scale model data. However, it is less influenced by behaviorism.
Currently, the development of artificial intelligence is still in the era of weak artificial intelligence. Although weak artificial intelligence seems to "emerge" autonomous consciousness, it has not been verified that artificial intelligence with true consciousness has appeared.
4. Three Attitudes Toward Artificial Intelligence#
In my understanding, different technology giants and influential people have different attitudes toward artificial intelligence. There are those who actively and even fanatically support it, those who follow steadily, and those who are pessimistic and worried. Therefore, these three types of people have formed the radical faction, the steady faction, and the pessimistic faction of artificial intelligence.
-
Radical Faction: OpenAI and Microsoft. When OpenAI's ChatGPT was released to the public in 2022, it was still an unfinished product, and users could be regarded as participants in large-scale data training. With the expansion of training data, ChatGPT's understanding, computation, and reasoning abilities have almost reached the level of human experts. Despite the existence of "illusion" in artificial intelligence, OpenAI is still actively promoting ChatGPT and other artificial intelligence. Microsoft, which heavily invested in OpenAI, saw the popularity of ChatGPT when the Bing search engine had a low market share. As a result, Microsoft has almost involved all aspects of the company in ChatGPT, such as Microsoft Copilot (formerly New Bing, which can be used on mobile phones with GPT-4).
-
Steady Faction: Google's Bard and Anthropic's Claude (invested by Amazon/Google). Google developed the Transformer model in 2017, which is the model that artificial intelligence such as ChatGPT should have. However, perhaps Google attaches importance to the ethical and legal issues of artificial intelligence and recognizes the risks that artificial intelligence development brings to humans, resulting in the temporary lag of ChatGPT and other similar artificial intelligence services compared to OpenAI. Google's emphasis on the risks of artificial intelligence is reflected in recent preparations by the Google AI team to draft a "robot constitution" in an attempt to establish laws for robots to prevent harm to humans. Previously, Google also attached great importance to ethical research on machines (the company's dismissal of Blake Lemoine, an ethical research expert who claimed that "LaMDA" has autonomous consciousness, caused controversy, and LaMDA should be a pioneer of ChatGPT, Bard, etc.).
Bard and Claude, although they also follow the development of GPT-based artificial intelligence, also pay attention to the potential risks of artificial intelligence. Therefore, they can be regarded as the steady faction of artificial intelligence.
- Pessimistic Faction: Elon Musk and his cultivated artificial intelligence XAI. Elon Musk was an early investor in OpenAI but later withdrew due to differences in views. Today, he attaches great importance to the risks brought by artificial intelligence and emphasizes that artificial intelligence should serve humanity without harming it.
5. Conclusion#
In conclusion, in the exploration of artificial intelligence, various schools of thought such as "symbolic AI," "connectionism," and "behaviorism" have emerged. Since the 1950s and 1960s, artificial intelligence seems to have integrated into various aspects of people's lives, work, and learning, forming different attitudes such as the radical faction, the steady faction, and the pessimistic faction.
However, human exploration in the field of artificial intelligence is still in its early stages. Whether weak artificial intelligence will "emerge" autonomous consciousness after large-scale data training is something that people need to pay attention to and study. The potential dangers of artificial intelligence also require attention and research.
Note: This article was originally published on January 8, 2024, and is archived here.
References: Zhihu 1, Zhihu 2, Science Popularization China.