AI Duel： Five Differences between Generative AI and Traditional AI

The hottest estimate in recent years is the AI track. But now everyone is basically talking about generative AI, not traditional AI like Alpha Go. What’s the difference between the two? Let’s analyze it from five aspects.

Artificial intelligence (AI) is a very broad topic. Today, the author will discuss the generative AI and traditional AI in detail. From the perspective of technology, application, development process, capability scope and future development potential, we can see their differences.

Traditional AI is based on rules and data.

Traditional AI, also known as "rule-driven AI" or "discriminant AI", mainly relies on preset rules and a large number of training data to work.

Its core idea is to train through a large number of data, so that the model can learn to extract features from the data, and then classify or predict according to these features. Traditional AI includes various machine learning algorithms, such as decision tree, support vector machine (SVM) and logistic regression.

For example, if you want traditional AI to recognize pictures of cats and dogs, you need to provide thousands of labeled pictures of cats and dogs. AI will classify these pictures by learning their features (such as pointed ears of cats and round noses of dogs).

Generative AI is based on generative model.

Generative AI is an AI that can generate new content. It can not only identify and classify data, but also create new data.

The core technologies of generative AI include generative countermeasure networks (GANs), variational encoders (VAEs) and large language models (such as GPT-3). By learning a large number of training data, these models can generate new content similar to the training data.

For example, if you let generative AI learn a lot of paintings, it can not only identify different styles of paintings, but also create new paintings by itself, imitating a certain style.

Application of traditional AI

Traditional AI is widely used in various fields that need accurate classification and prediction. The following are the main applications:

1. Medical diagnosis: Traditional AI can help doctors analyze medical images and identify early lesions, such as cancer and pneumonia.

2. Financial services: In banks and financial institutions, traditional AI is used for credit scoring, fraud detection and stock market forecasting.

3. Speech recognition: Traditional AI technology is used to develop voice assistants, such as Siri of Apple and Google Assistant of Google, which can recognize and execute voice commands of users.

Application of generative AI

Generative AI is used in many creative and generating tasks because of its unique creative ability. The following are the main applications:

1. Content generation: Generative AI can be used to create new text content, stories, articles and even news reports. For example, GPT-3 of OpenAI can generate high-quality articles according to the input prompts.

2. Artistic creation: Generative AI can create new music, paintings and videos. For example, GANs can generate new works of art and even imitate the styles of famous artists.

3. Game design: Generative AI can create new game characters, scenes and plots, and improve the diversity and interest of the game.

The development of traditional AI

The development of traditional AI can be traced back to 1950s. The Dartmouth conference in 1956 was considered as the symbol of the birth of AI, and scientists began to explore how to make machines intelligent.

1980s: Expert system was an early AI application, which used rule base and reasoning engine to simulate the decision-making process of human experts.

2000s: The rise of machine learning With the improvement of computing power and the explosive growth of data, machine learning algorithms have been widely used.

Development course of generative AI

The development of generative AI is relatively new, and the following are several important milestones:

2014: The proposal of GANs: Ian Goodfellow proposed the Generation Countermeasure Network (GANs), which is an important breakthrough of generative AI, and it can generate very realistic images and videos.

2017: Transformer model: The Transformer model proposed by Google laid the foundation for the development of generative AI, especially in the field of natural language processing.

2020: release of GPT-3: OpenAI released GPT-3, a language model with 175 billion parameters, which can generate high-quality text, marking an important milestone of generative AI.

The ability range of traditional AI

Traditional AI is good at solving problems in specific fields, such as classification and regression. Traditional AI can perform classification and regression tasks by learning historical data, such as image classification, speech recognition, and forecasting house prices.

Pattern recognition: It can recognize and extract patterns and features from data, and is widely used in image processing, voice processing and other fields.

Automated decision-making: Traditional AI can make decisions automatically based on preset rules, such as credit scoring and fraud detection.

Capability range of generative AI

The capabilities of generative AI are more extensive and flexible, and the main capabilities include:

Content generation: Generative AI can generate high-quality text, images, music and other content. For example, ChatGPT can write stories and answer questions; GANs can generate realistic face images.

Simulation and prediction: it can simulate complex systems and make predictions, such as weather forecast and market trend forecast.

Enhance creativity: Generative AI can help artists and designers create new works of art and provide creative inspiration.

It is important to note here that the development potential of traditional AI is not necessarily less than that of generative AI. The two are just like the application of different technical capabilities in different scenarios. As long as the choice is correct and appropriate, the problem can be solved well.

"In fact, an application, if it solves complex scenes and complex problems, must be a master of new and old technology paradigms. For example, the customer service robot of a financial enterprise requires extremely high controllability. The content generated by the big model can’t be seen directly, but must be manually configured. It can be RNN, Bert and other models for classification and vector retrieval, or it can be an expert system of the older generation. "

The above text comes from Yu Changhong, a teacher in the group chat (the head of Zhuiyi Technology Product Center and former product director of Tongcheng Yilong, WeChat official account: Hongguan AI).

Future development of traditional AI

Traditional AI is still developing, and the potential directions in the future include:

1. Higher accuracy and efficiency: With the improvement of algorithm and computing power, the accuracy and efficiency of traditional AI in classification and prediction tasks will be further improved.

2. Wider application: Traditional AI will be applied to more fields, such as intelligent manufacturing, smart cities, autonomous driving, etc., further changing our lifestyle.

3. Integration with other technologies: Traditional AI will be combined with technologies such as Internet of Things, big data and cloud computing to form a smarter system.

The future development of generative AI

Generative AI has great potential, and several potential directions in the future:

1. Stronger generating ability: Generative AI will be able to generate more complex and high-quality content, such as more realistic virtual worlds and more delicate works of art..

2. Cross-domain application: Generative AI will be applied to more fields, such as education, medical care, entertainment, etc., providing personalized and innovative solutions.

3. Cooperation with human beings: Generative AI will become an important assistant for human creative work, helping human beings to achieve more creative ideas and innovative breakthroughs.

Let’s understand the difference between generative AI and traditional AI through several cases:

Case 1: Writing Assistant

Traditional AI intelligent writing assistants, such as early text correction tools, mainly rely on preset grammar rules and dictionaries for their work. If you write a wrong word or use the wrong grammar, they will prompt you to make corrections. However, the ability of this tool is limited, and it can’t understand complex contexts and generate creative content.

Intelligent writing assistants of generative AI, such as GPT-4, can generate high-quality text content according to the context. For example, you can ask GPT-4 to help you write an article about environmental protection. It can not only understand your requirements, but also generate smooth and logical paragraphs, and even provide some innovative ideas and insights.

Case 2: Intelligent Customer Service

The intelligent customer service of traditional AI mainly answers users’ questions through preset question-and-answer pairs. If the user’s question is within the preset range, the intelligent customer service can answer it quickly and accurately. However, if the problem is beyond the preset range, the intelligent customer service will be powerless.

Intelligent customer service of generative AI, such as customer service system based on GPT-3, can understand and answer more complicated questions. Even if the user’s question is not within the preset range, the generative AI can generate appropriate answers according to a large amount of information learned. This makes the user experience more natural and smooth.

Case 3: Medical Diagnosis

The application of traditional AI in medical diagnosis is mainly through a large number of medical images and medical records to help doctors identify and diagnose diseases. For example, traditional AI can help doctors identify early lung cancer by learning a lot of X-rays.

The application of generative AI in medical care is more extensive and flexible. In addition to identifying and diagnosing diseases, it can also generate personalized treatment plans and even simulate the mechanism of action of drugs, helping doctors to make complex medical research and decisions.

In order to distinguish between generative AI and traditional AI. From the technical point of view, application point of view, development process, capability scope and future development potential, I hope to help you better understand the difference between the two AI technologies ~