-- Key technologies and milestones from Turing test to DeepSeek

DeepSeek

DeepSeek is a Chinese artificial intelligence start-up company, focusing on the research and development of large language models (LLM). The flagship product DeepSeek-R1 model has excellent reasoning ability in fields such as mathematics and coding, and its performance is comparable to that of the OpenAI model. DeepSeek adheres to the open source strategy, promotes technology sharing and innovation, and actively promotes the development of multi-modal large language models.

Time: The company was established on July 17, 2023 and has released multiple versions of models since January, 2024. At the end of January 2025, the open-source multi-modal AI model Janus-Pro was released.

Characters: The founders are Liang Wenfeng, the founder of Magic Cube quantification, and DeepSeek R & D team.

Papers: "DeepSeek LLM: Scaling Open-Source Language Models with longterrains"(arXiv, January 5, 2024) and other papers.

Impact: The rise of DeepSeek has a profound impact on the global artificial intelligence industry structure. DeepSeek-R1 achieve high performance at low cost, triggering a global upsurge of open source reproduction and promoting AI innovation and competition. DeepSeek has shown strong performance in natural language processing, multi-modal and other fields, and its core capabilities such as mathematical reasoning, code generation, and multi-round dialogue have reached the international leading level. DeepSeek-V3 is outstanding in knowledge tasks, training efficiency and reasoning speed are greatly improved, and training costs are significantly reduced. In addition, Janus-Pro performs well in multi-modal understanding and generation tasks, beating OpenAI's DALL-E 3 and Stable Diffusion. DeepSeek innovation is reflected in the optimization of model architecture and training methods. For example, Janus-Pro adopts the three-stage training method. DeepSeek mobile applications quickly topped the US iPhone download list, demonstrating its technical strength and market influence.

BERT: pre-training of deep bidirectional Transformer

BERT(Bidirectional Encoder Representations from Transformers) is a Transformer-based deep Bidirectional pre-training Language Model. Through Masked Language Model(MLM) and Next Sentence Prediction(NSP) tasks, learn deep two-way language representation and make breakthrough progress in many natural language processing tasks.

Time: October 2018

characters: Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova(Google AI Language team)

paper: BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (arXiv, October 11, 2018)

impact: BERT's proposal is an important milestone in the field of natural language processing. For the first time, it realizes deep two-way pre-training of joint adjustment of left and right contexts at all levels, thus capturing semantic information of words and sentences more effectively. BERT refreshed 11 records in the GLUE benchmark test, proving its strong language understanding ability. BERT's pre-training-fine tuning paradigm and bidirectional Transformer architecture have had a profound impact on the subsequent development of language models, such as RoBERTa, ALBERT, DistilBERT, etc. BERT and its derivative models have been widely used in information retrieval, question answering system, machine translation and other fields, greatly promoting the progress and application of natural language processing technology.

Sophia

Sophia is a social humanoid robot developed by Hanson Robotics, which is famous for its humanoid appearance and advanced interaction ability. She is the first non-human to win the title of the United Nations, triggering extensive discussions around the world on the role of artificial intelligence in sustainable development.

Time: february 14th, 2016: sofia first activated; October 25th, 2017: acquired saudi arabian citizenship; November 21th, 2017: appointed as the innovation champion of the united nations development programme in the asia-pacific region.

Characters: David Hanson (founder), Ben Goertzel (Chief Scientist), Hanson Robotics research and development team.

Paper: there is no academic paper with Sofia as the theme directly, and its subsystem "Open Arms" related papers are submitted to NeurIPS 2022 conference.

Impact: Sophia's appointment as the innovation champion of the United Nations Development Programme has stimulated extensive global discussions on the role of artificial intelligence and robots in sustainable development. She participated in various high-level activities and interacted with people from all walks of life around the world, demonstrating the potential of artificial intelligence in meeting global challenges. Sofia's public exposure not only improved the public's understanding of artificial intelligence technology, but also promoted the development of human-computer interaction technology, and triggered extensive discussions on artificial intelligence ethics and social impact. Although Sophia's technical ability is still controversial, her role in promoting public cognition and social discussions cannot be ignored.

AlphaGo

AlphaGo is a weiqi artificial intelligence program developed by DeepMind. Through deep neural network and Monte Carlo tree search algorithm, it has reached and surpassed the top level of human beings in weiqi game and has become an important milestone in the field of artificial intelligence.

Time: January 2016 (paper publication time), March 2016 (man-machine war time)

characters: David Silver, Demis Hassabis, Shane Legg, Mustafa Suleyman and DeepMind team

paper: Mastering the Game of Go with Deep Neural Networks and Tree Search, published in the journal Nature on January 28, 2016

impact: AlphaGo's success demonstrates the great potential of deep learning and reinforcement learning in solving complex strategy problems and has a profound impact on the field of artificial intelligence. It has not only made a breakthrough in the field of go, but also provided new ideas and methods for the application of artificial intelligence in other complex decision-making problems. The emergence of AlphaGo has stimulated people's attention and research boom on artificial intelligence, promoted the rapid development of related technologies, and had a positive impact on robot control, autonomous driving, resource management and other fields.

Federated Learning

federated learning is a machine learning technology that trains on distributed datasets. It enables effective learning without centralized data.

Time: February, 2016

characters: H. Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hammond, Blaise Agüera y Arcas(Google UED)

paper: "Communication-Efficient Learning of Deep Networks from Decentralized Data" (arXiv, February 2016; AISTATS, 2017).

Impact: Federated learning has a profound impact on artificial intelligence and machine learning by solving privacy protection issues and supporting collaborative learning across distributed datasets. This technology can conduct model training without exchanging raw data, which is especially critical in industries such as medical treatment and finance that are crucial to data privacy. In addition, federated learning has a positive impact on the development of edge computing and device-side artificial intelligence because it allows model training on user devices without the need to transmit sensitive data to centralized servers. With the increasingly strict laws and regulations on data protection and the increasing attention to data privacy, Federal learning has a wider application prospect.

OpenAI

OpenAI has made remarkable progress in many fields of artificial intelligence, especially in Natural Language Processing (NLP) and Reinforcement Learning.

Time: established in December 2015; GPT-3 was released in June 2020, GPT-3.5(ChatGPT) was launched in late 2022, and GPT-4 was released in March 2023.

Character: OpenAI was founded by Elon Musk, Greg Brockman, Ilya Sutskever and others.

Paper: for example, Language Models are Few-Shot Learners (related to GPT-3), published in arXiv in 2020.

Impact: OpenAI's GPT series models have laid the foundation for pre-training models in the field of natural language processing, greatly improving language generation and understanding capabilities, it also promotes the rapid development of language models based on Transformer architecture. The release of ChatGPT has attracted worldwide attention to generative AI and accelerated the process of integrating natural language processing technology into actual business scenarios in various industries, it expands the application of artificial intelligence in many fields. In addition, OpenAI's breakthroughs in reinforcement learning and other fields provide examples for the application of artificial intelligence in complex decision-making scenarios, and promote the research and development of robot control, autonomous driving and other technologies.

ResNet

resNet(Residual Networks) introduces the deep Residual learning framework, solves the problem of gradient disappearance in deep neural network training, and realizes a deeper and more accurate image recognition model.

Time: December 2015

characters: Kaiming He, Xiangyu Zhang, Shaoqing Ren and Jian Sun (Microsoft Asia Research Institute, Microsoft Research Asia)

paper: Deep Residual Learning for Image Recognition, published in CVPR in 2016

impact: ResNet revolutionarily solves the training problem of deep neural networks, making it possible to build networks with more than 100 layers. It has achieved great success in image recognition tasks such as ImageNet and has deeply affected the subsequent in-depth learning research. The idea of residual connection is widely used in various tasks, such as target detection and semantic segmentation, and has become an important part of modern deep learning models. The emergence of ResNet is an important milestone in the development of deep learning, marking that deep learning has entered a "deeper" era.

TensorFlow

TensorFlow is an open source machine learning framework developed by Google Brain. It provides a comprehensive platform for expressing machine learning algorithms and can efficiently execute these algorithms in various heterogeneous computing environments from mobile devices to large-scale distributed systems. TensorFlow uses data flow graphs to represent computing, which makes the deployment of machine learning models highly flexible and scalable.

Time: November 2015 (first open source)

people: The Development of TensorFlow is led by Google's brain team. Its main members include Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, etc.

Paper: "TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems",(arXiv, March 14, 2016).

Impact: TensorFlow provides a universal and scalable platform for Machine Learning Research and Application Development, greatly promoting the development of artificial intelligence. Its open source nature has been widely adopted by academia and industry, promoting research and application in computer vision, natural language processing, robots and other fields. TensorFlow's flexibility supports deployment across various hardware platforms, which makes it play an important role in the development and deployment of artificial intelligence models.

Deep learning (DL)

this paper comprehensively reviews the development process, core technologies and application prospects of Deep Learning, and expounds that Deep Learning is a kind of Multilayer Neural Networks learning complex patterns and Feature Representation methods from data breakthrough progress in other fields.

Time: May 2015

characters: Yann LeCun, yoshibengio, Geoffrey Hinton

Paper: Deep learning, published in Nature, May 2015, Volume 521, No. 7553.

Impact: This paper is a milestone document in the field of deep learning, providing a comprehensive knowledge framework for deep learning for academia and industry, and promoting deep learning in computer vision, natural language processing, widely used in speech recognition and other fields. It not only provides an important theoretical basis for subsequent research, but also stimulates a large number of technological innovations, such as Convolutional Neural Networks,CNN), Recurrent Neural Networks,RNN), etc., have profoundly affected the development process in the field of artificial intelligence.

Generate an anti-Network (GAN)

contextual Adversarial Networks (GANs) is a new framework for generating models through confrontation process training. It consists of two neural Networks: Generator and Discriminator. The generator learns to generate realistic data samples, while the discriminator learns to distinguish real data from generated data.

Time: June 2014

characters: Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, yoshibengio

paper: Affective Adversarial Networks, published in Advances in Neural Information Processing Systems (NeurIPS) 2014.

Impact: GANs has completely changed the field of model generation and has achieved remarkable results in image synthesis, style migration, data enhancement and other fields. The emergence of GANs has promoted the progress of Deep Learning in Unsupervised Learning and stimulated a large number of follow-up studies, such as DCGANs, WGANs, StyleGANs, etc. These studies have continuously improved the performance and application scope of GANs, injecting new vitality into the development of artificial intelligence in the fields of image generation, video generation, natural language processing, etc. GANs also shows great potential in the fields of artistic creation, drug discovery, game development and so on, which has a profound impact on the future development of artificial intelligence.

Dropout regularization technology

in 2014, niansrivastava et al. proposed Dropout, a regularization technique used to alleviate the over-fitting of Deep Neural Networks (DNN). This method randomly discards some neurons in the network and their connections during the training process, effectively preventing excessive cooperative adaptation between neurons and improving the generalization ability of the model.

Time: 2014

characters: niansrivastava, Geoffrey Hinton,Alex Krizhevsky,Ilya Sutskever,Ruslan Salakhutdinov

paper: Dropout: A Simple Way to Prevent Neural Networks from Overfitting, published in Journal of Machine Learning Research,2014.

Impact: Dropout, as an effective regularization method, has a profound impact on the field of artificial intelligence. It significantly improves the performance of deep neural network in image recognition, natural language processing and other tasks, and becomes one of the standard technologies in deep learning. The proposal of Dropout enables researchers to train deeper and more complex network structures, which promotes the rapid development and wide application of deep learning.

Variational self-encoder (ADATA)

Variational Autoencoders (ghts) is a Deep Inference Model based on the Variational Inference. By learning the Latent Representation of data to generate new data.

Time: December 2013

characters: Diederik P. Kingma and Max Welling

paper: Auto-Encoding Variational Bayes, published in arXiv(arXiv:1312.6114).

Impact: The proposal of the variational self-encoder (ADATA) is an important milestone in the field of deep generation models. It combines variation inference with neural networks and provides an effective method for dealing with probability models with continuous potential variables. GenBank has been widely used in many fields, including:

image Generation: ADATA can generate high-quality and diversified images, such as faces and landscapes.

Natural language processing: ADATA can be used to generate text and learn text representation.

Drug discovery: GenBank can be used to generate molecules with specific properties and accelerate the drug research and development process.

Recommend system: ADATA can be used to learn the potential representation of users and articles for personalized recommend.

In addition, the proposal of VAE also promotes the development of other generation models, such as Generative Adversarial Networks (GANs).

Word vector model (Word2Vec)

in this study, a word vector model (Word2Vec) is proposed, which is an efficient method for learning the distributed representation of words and can capture the semantic relationship between words.

Time: October 2013

characters: Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg Corrado and Jeffrey Dean (Google team)

paper: "Distributed Representations of Words and Phrases and their Compositionality", published in Advances in Neural Information Processing Systems (NeurIPS)2013.

Impact: This study has created a new paradigm in the field of Natural Language Processing (NLP). Word vectors generated by Word2Vec model can effectively capture the semantic and syntactic relationships of words, and bring significant performance improvement for many NLP tasks (such as machine translation, sentiment analysis, text classification, etc.). With its simplicity and high efficiency, Word2Vec quickly became the basic tool in NLP field, which promoted the development of subsequent word vector representation methods and deep learning models, and had a profound impact on the birth of Transformer, BERT and other models.

Representation Learning

this study comprehensively reviews Representation Learning, emphasizes the importance of Learning effective data Representation for improving the performance of machine Learning algorithms, and puts forward new research directions.

Time: August 2013

characters: yoomed Bengio, Aaron Courville, Pascal Vincent (University of Montreal)

paper: Representation Learning: A Review and New Perspectives, published in IEEE Transactions on Pattern Analysis and Machine Intelligence.

Impact: This paper lays a theoretical foundation for the field of representation learning and deepens people's understanding of the importance of data representation in machine learning. It systematically reviews a variety of Representation learning methods and puts forward new research directions, inspiring a large number of subsequent studies. This work provides an important promotion for the development of deep learning, has a profound impact on computer vision, natural language processing and other fields, and has promoted the transformation from traditional feature engineering to automatic feature learning.

AlexNet

AlexNet is a Deep Convolutional Neural Network (CNN) architecture, which won a breakthrough in the ImageNet Large-Scale Visual Recognition Challenge (ILSVRC) in 2012, it significantly improves the accuracy of image classification and marks the rise of Deep Learning in the field of Computer Vision.

Time: September 2012

characters: Alex Krizhevsky,Ilya Sutskever,Geoffrey Hinton (University of Toronto, University of Toronto)

paper: ImageNet Classification with Deep Convolutional Neural Networks, published in the 2012 Neural Information Processing Systems (NIPS).

Impact: AlexNet's success has proved the powerful ability of deep Convolution Neural Network in large-scale image recognition tasks, which has triggered extensive attention and research upsurge in deep learning in academia and industry. It not only provides an important reference for the development and optimization of subsequent deep learning models, but also promotes the rapid development and wide application of computer vision technology in the fields of image classification, target detection, image segmentation, etc. The ReLU activation function and Dropout regularization technologies adopted by AlexNet have also become the standard for subsequent deep learning models. In addition, AlexNet's training process shows the important role of GPU in deep learning and promotes the popularization of GPU in the field of deep learning.

Google Knowledge Map

Google Knowledge Graph is a semantic network designed to improve search results by understanding entities and their relationships.

Time: May 16, 2012

person: Google search team led by Amit Singhal (then senior vice president of Google engineering)

paper: "Introducing the Knowledge Graph: Things, Not Strings"(Google official blog, May 16, 2012)

impact: Google Knowledge Map significantly improves the accuracy and relevance of search by understanding the context and relationship between entities. It has changed the paradigm of search engines from keyword matching to semantic understanding, making search results more comprehensive and intuitive and presented in the form of information cards. This technology not only optimizes the user search experience, but also promotes the development of artificial intelligence fields such as semantic search and knowledge representation, and has a profound impact on subsequent applications such as intelligent Q & A and recommend systems.

IBM Watson Zhisheng "the edge of danger"

IBM Watson is an artificial intelligence Q & A system, which was published in February 2011 in "the edge of danger" (beautardy!) The intelligence contest program defeated two human champions, demonstrating their strong abilities in natural language processing and knowledge Q & A.

Time: February, 2011

person: IBM DeepQA team, in charge of David Ferrucci

paper: "Building Watson: An Overview of the DeepQA Project", published in the autumn of 2010 in the journal AI Magazine.

Impact: Watson's victory proves the great potential of artificial intelligence in natural language processing and question answering systems. It demonstrates the ability of machines to understand and respond to complex human language queries, paving the way for the application of artificial intelligence in the fields of medical treatment, customer service and information retrieval. Watson's success in "the edge of danger" is an important milestone, proving that machines can compete with human beings in tasks requiring extensive knowledge and rapid information processing. This achievement rekindled people's interest in the research and application of artificial intelligence and promoted the current development wave of artificial intelligence. Since then, IBM has expanded Watson's capabilities to various industries, including health care, to assist medical diagnosis and treatment advice, this further proves the potential of artificial intelligence to enhance human professional knowledge in complex decision-making processes.

Google driverless car project

google launched the driverless car project in 2009, aiming at developing vehicle technologies that can realize self-driving in complex traffic environments. Through advanced sensors, software and artificial intelligence technology, the project has realized autonomous navigation of vehicles under real road conditions.

Time: January 17, 2009

Person: The project is in the charge of the engineer team of Google X laboratory, and Sebastian Thrun is an important promoter of the project. In addition, Anthony Levandowski, John Krafcik and others have also made important contributions to the project. In December 2016, the project was independent from Google and Waymo was established.

Paper: the results and progress of relevant research are mainly published in the form of news reports and project reports, rather than non-traditional academic papers. For example, the 2010 New York Times report "Google cars drive themselves in traffic" reported on the project.

Impact: Google's driverless car project is an important milestone in the development of self-driving technology. It not only promotes the application of artificial intelligence in the field of transportation, but also stimulates automobile manufacturers and technology companies to invest in the research and development of self-driving technology. The project demonstrated the potential of autonomous driving technology in improving traffic safety and travel efficiency, and triggered extensive discussions on autonomous vehicle safety, legal and ethical issues. As a continuation of this project, Waymo has made important progress in the commercialization of self-driving technology. For example, Waymo One self-driving taxi service has been launched, providing the public with the opportunity to experience self-driving technology. The project has also promoted the development of sensor technology, machine learning algorithms and other related fields to meet the needs of self-driving vehicles for environmental perception, decision planning, etc.

Transfer Learning

this study gives a comprehensive overview of the concept, classification, methods and applications of Transfer Learning, laying a foundation for the development of this field.

Time: October 2010

characters: Sinno Jialin Pan,Qiang Yang

paper: Pan, S. J., & Yang, Q. (2010). A Survey on Transfer Learning, published in IEEE Transactions on Knowledge and Data Engineering, Volume 22, Issue 10, 1345-1359.

Impact: This paper systematically expounds the theoretical framework of migration learning, summarizes migration strategies and applications in different scenarios, and provides clear cognition for researchers. This paper has greatly promoted the development of migration learning in the field of artificial intelligence, provided the basis and guidance for subsequent research, and stimulated more exploration on new algorithms and applications of migration learning. In the aspect of technological innovation, this research urges researchers to further improve and optimize migration learning technology based on its classification and methods; In the aspect of application promotion, this enables migration learning to be more widely used in many fields such as Natural Language Processing and Computer Vision, improving model training efficiency, reducing data dependence, and reducing labeling costs.

Potential Dirichlet allocation (LDA)

LDA is a probability model for generating discrete data sets (such as text corpus). Through a three-layer Bayesian hierarchy, it reveals the unobserved data hidden behind the observed data and explains the similarity of the data.

Time: April 2003

characters: David M. Blei, Andrew Y. Ng, Michael I. Jordan

paper: Latent Dirichlet Allocation, published in Journal of Machine Learning Research, Volume 3, pages 993-1022, 2003. [jmlr.org]

impact: LDA has become one of the most popular topic modeling methods and is widely used in text analysis, information retrieval and natural language processing. It has had a significant impact on the development of probabilistic modeling in machine learning and has been extended to fields other than text analysis, including computer vision and bioinformatics. LDA can discover potential topics in large-scale document collections, making it a valuable tool for tasks such as content analysis, recommend systems, and document classification. This model models each document as a mixture of topics, where topics are defined as word distributions, thus automatically discovering hidden topic structures in large text datasets. LDA has become the basic technology of natural language processing and machine learning, promoting the development of document classification, collaborative filtering and bioinformatics, it also stimulates further research on dynamic theme models and related theme models.

Conditional random field (CRF)

Conditional Random Fields (CRFs) is a probability graph model for marking and segmentation of sequence data. It overcomes the limitations of traditional models (such as Hidden Markov Model) and is widely used in Natural Language Processing, Bioinformatics and other fields.

Time: 2003

characters: John Lafferty,Andrew McCallum,Fernando Pereira

paper: Conditional Random Fields: statistic Models for Segmenting and Labeling Sequence Data, published in Proceedings of the 20th International Conference on Machine Learning (ICML-2003).

Impact: conditional random fields play an important role in the labeling and analysis of sequence data in natural language processing, bioinformatics and other fields. It solves the problem that traditional models have too strong assumptions on data independence and can handle complex dependencies in sequence data more flexibly. CRFs has become a powerful tool for sequence analysis in these fields, promoting the development of related technologies and providing an important reference for the research and application of subsequent probability graph models.

Semantic Web

semantic Web aims to achieve more efficient retrieval and automatic processing of Internet information through machine-readable semantic descriptions.

Time: May, 2001

characters: Tim Berners-Lee, James Hendler, Ora Lassila

paper: The Semantic Web, published in Scientific American, May 2001

impact: Semantic Web is an important milestone in the development of the World Wide Web. It brings deeper semantic understanding and information processing capabilities to the Internet. This concept has given birth to key technologies such as Resource Description Framework (IDF) and Network Ontology Language (OWL), and has promoted the rapid development of knowledge mapping, intelligent search and other fields. Semantic Web not only improves the accuracy and efficiency of information retrieval, but also lays a foundation for the application of artificial intelligence in knowledge representation and reasoning. Although the full realization of semantic web still faces challenges, its core ideas have deeply affected the development direction of Internet technology.

Dark blue defeated Kasparov

in May 1997, IBM's supercomputer "dark Blue" defeated world champion Garry Kasparov in a chess competition, it became the first computer system to win the champion of human chess under standard competition conditions.

Time: May 1997

people: IBM dark blue team (Leader: Feng-hsiung Hsu), Kasparov (Garry Kasparov)

paper: Beating the World Chess Champion, published in Artificial Intelligence, January 2002

impact: The victory of dark blue is an important milestone in the development history of artificial intelligence, demonstrating the great potential of computers in complex decision-making tasks. It has aroused public attention to artificial intelligence, promoted research and development in related fields, and provided examples for the application of artificial intelligence in other fields.

Support vector machine (SVM)

Support Vector Machine (SVM) is a supervised learning model for classification and regression analysis. It distinguishes different types of data points by constructing an optimal hyperplane in high-dimensional space.

Time: 1995

characters: Corinna Cortes,Vladimir N. Vapnik

paper: Support-vector networks, published in Machine Learning, Volume 20 (1995):273-297.

Impact: SVM has had a profound impact on the field of machine learning. It provides a new method to solve the classification problem, improving the generalization ability of the model by maximizing the interval between categories. SVM performs well in processing high-dimensional data and small sample data, so it has been widely used in image recognition, text classification, bioinformatics and other fields. In addition, SVM also promotes the development of Kernel Methods, making nonlinear classification problems effectively solved. The successful application and theoretical contribution of SVM have laid a foundation for the research and development of subsequent machine learning algorithms.

Boston Dynamics

Boston Dynamics has made remarkable progress in robot motion, balance and adaptability to complex environments. It has developed a series of landmark robots the field has a profound impact.

Time:

1992: The company was established

2005: BigDog released

2013: Atlas released

2015: Released on the Spot

characters: Marc Raibert (founder), Rooney Brooks (early team member)

paper: The research results of Boston Dynamics are mainly published in the form of technical reports, patents and demonstration videos, and less academic papers are published. Relevant information can be found on its official website and technical publications.

Impact: Boston Dynamics has promoted the development of robotics, especially in dynamic balance, motion control and adaptability to complex environments. The robots developed by it have shown application prospects for military, industrial, search and rescue fields, and stimulated further exploration of robot technology in academia and industry. These technological advances have also promoted the integration of artificial intelligence and robot technology, laying a foundation for the development of intelligent robots.

Bayesian network

Bayesian Network is a graphical model based on Probabilistic Inference, which is used to represent and deal with uncertain knowledge. It describes the conditional dependence between variables through directed acyclic graphs (DAG), and uses Bayes' Theorem for probability calculation and reasoning.

Time: 1988

character: Judea Pearl

paper: Bayesian Networks: A Model of Self-Activated Memory for Evidential Reasoning, published in UCLA Technical Report CSD-880017, it was later published in Proceedings of the 7th Conference of the Cognitive Science Society (May 1, 1989).

Impact: Bayesian network provides a powerful framework for dealing with uncertainty problems in the field of artificial intelligence, and promotes the development of Expert Systems, this enables machines to make reasoning and decisions under incomplete information. It has been widely used in expert system, fault diagnosis, medical diagnosis, information retrieval and other fields, promoting the progress of artificial intelligence in dealing with complex and uncertain problems in the real world. Bayesian networks have laid an important foundation for the application of probabilistic reasoning and decision-making theory in the field of artificial intelligence. Many subsequent uncertain reasoning and Machine Learning the algorithms and models of both refer to the idea of Bayesian network, such as information fusion and decision-making in Data Mining and Multi-agent Systems.

Behavior-based robot control

rooney A. Brooks proposed A behavior-based robot control architecture, emphasizing the realization of complex tasks through the combination of simple behaviors, so that robots can better adapt to the dynamic environment.

Time: 1986

character: Rooney A. Brooks

paper: A robust layered control system for a mobile robot, published in IEEE Journal of Robotics and Automation.

Impact: This research has changed people's understanding of robot control and promoted the transformation of robot technology from rule-based programming to more flexible and adaptable behavior control, it lays a foundation for the autonomous operation of modern robots in complex environments.

Expert System

Expert System is an important early research in the field of artificial intelligence, aiming at simulating the decision-making process of human experts in specific fields. By coding the knowledge of field experts into rules and data that can be recognized by computers, it enables computers to make reasoning and decisions like experts.

Time: 1983

characters: Edward Feigenbaum, Pamela McCorduck

paper: The Fifth Generation: Artificial Intelligence and Japan's Computer Challenge to the World (The Fifth Generation: Artificial Intelligence and Japan's Computer Challenge to the World),1983

impact: the proposal and development of expert systems have had a profound impact on the field of artificial intelligence. It has pushed artificial intelligence from theoretical research to practical application, and has achieved initial success in medical treatment, finance, geological exploration and other fields. The success of expert system shows that artificial intelligence technology has great potential in solving problems in specific fields, which has stimulated people's interest and investment in artificial intelligence, it lays a foundation for the subsequent development of artificial intelligence technology.

Computer beat world champion

computer programs beat the world champion in the backgammon game, demonstrating the potential of artificial intelligence in complex strategy games.

Time: July 15, 1979

person: Hans J. Berliner (developer of BKG 9.8 program)

paper: "Backgammon Computer Program Beats World Champion", published in the September 1980 issue of Artificial Intelligence.

Impact: This event is the first time that a computer program has defeated the human world champion in an intellectual game, which is of milestone significance. It proves the ability of artificial intelligence in complex strategy decision-making, stimulates more researchers to explore computer game algorithms, and promotes the further development of artificial intelligence in the game field. This victory has also improved public awareness of artificial intelligence and promoted the application of artificial intelligence technology in a wider range of fields. Although Berliner admitted that the luck factor played a role in it, this victory was still of great significance, for the later dark Blue (Deep Blue), artificial intelligence programs such as AlphaGo have laid the foundation for breakthroughs in other chess games.

Application of meta-level knowledge in large knowledge base system

this study discusses the application of meta-level knowledge in large knowledge base system, and puts forward the method of using meta-level knowledge to control reasoning process and effectively using large knowledge base, which is used for knowledge representation, the development of reasoning mechanism and expert system has laid the foundation.

Time: 1977

person: The research was completed by researchers from the artificial intelligence laboratory of Stanford University's Department of Computer Science. The main contributors may include Edward Feigenbaum and other scholars who participated in the early research on artificial intelligence knowledge representation.

Paper: Title: Applications of meta level knowledge to the control of inference and use of large knowledge bases published in: 1977 Published by: Artificial Intelligence Laboratory, Department of Computer Science, Stanford University

impact: This study has laid a foundation for the construction and application of large-scale knowledge base systems and promoted the development of artificial intelligence in dealing with complex knowledge and reasoning tasks. The meta-level knowledge concept proposed by it has played an important enlightening role in the subsequent development of artificial intelligence technologies that rely on large-scale knowledge bases, such as knowledge maps and expert systems, it has promoted the expansion of artificial intelligence in knowledge-intensive applications. The research results have had a profound impact on the application and development of knowledge representation, reasoning mechanism, expert system and knowledge base technology in medical treatment, finance and other fields.

Computing theory of early processing of visual information

David Marr proposed a computational theoretical framework for early processing of visual information, emphasizing how the visual system extracts useful information from original image data, laying a foundation for the development of computer vision.

Time: 1976

character: David Marr

paper: "Early processing of visual information", published in Philosophical Transactions of the Royal Society of London. B, Biological Sciences, 1976, Vol. 275, No. 942, page 483-519.

Impact: Marr's visual computing theory has had a profound impact on the field of computer vision. It provides a systematic thinking framework for computer vision research and guides subsequent researchers to understand and solve visual information processing problems from different levels and angles. This theory promoted the development of early computer vision algorithms and models, and played an important role in the subsequent research based on feature extraction and Image Understanding, this makes computer vision research gradually shift from simple image processing to understanding and explaining visual information. Marr's work inspired the design of many subsequent computer vision algorithms and models, and promoted the wide application of computer vision technology in image recognition, target detection and other fields.

Knowledge Representation Framework (Frame Theory)

the "framework" theory put forward by Marvin Minsky in 1974 provides a theoretical basis for the storage, organization and reasoning of knowledge in the field of artificial intelligence.

Time: June, 1974

character: Marvin Minsky

paper: "A Framework for Representing Knowledge", published in MIT Artificial Intelligence Laboratory memorandum 306 in 1974, and later included in computer visual psychology (P. Edited by Winston), published in McGraw-Hill, 1975.

Impact: Minsky's framework theory is an important milestone in knowledge representation in the field of artificial intelligence. It provides a structured method for organizing and inferring complex information, it lays a theoretical foundation for the construction of expert system and semantic network. Framework theory also affects the formation of object-oriented programming and provides theoretical support for the construction of modern knowledge graph and ontology. The concept of framework helps bridge the gap between symbolic artificial intelligence and more flexible, context-dependent knowledge representation, and promotes the further development of cognitive architecture, it has become a key theory for understanding and reasoning the design of artificial intelligence systems in real world scenarios.

Automatic mathematician (AM)

the automatic mathematician (AM) program, developed by Douglas B. Lenat, shows a method of artificial intelligence in mathematical discovery, and shows its ability by rediscovering Jordan curve theorem.

Time: 1972

character: Douglas B. Lenat (Stanford University)

paper: "AM: An artificial intelligence approach to discovery in mathematics as illustrated by the discovery of the Jordan Curve Theorem" (Doctoral thesis)

impact: AM is a pioneer system in the field of automatic mathematical discovery and heuristic search. It uses a set of heuristic methods to guide its exploration of mathematical concepts and generate new concepts and conjectures. This work has made an important contribution to the development of artificial intelligence systems capable of creative problem solving and mathematical discovery. It also lays a foundation for future research in automatic reasoning and knowledge discovery systems. The ability of AM program to rediscover important mathematical theorems demonstrates the potential of artificial intelligence in scientific discovery and opens up a new research approach for computational creativity. Lenat's work on AM affects the subsequent development of machine learning and knowledge representation, especially in the field of automatic theory formation.

DENDRAL, the first expert system

DENDRAL is the first expert system aimed at analyzing mass spectrometry data and determining the molecular structure of organic compounds. The system adopts heuristic programming and imitates the problem-solving methods of Expert chemists.

Time: 1965-1968

figures: Edward Feigenbaum, Joshua Lederberg, Bruce Buchanan and other researchers at Stanford University.

Paper: "DENDRAL: A case study of the first expert system for scientific hypothesis formation", by Edward A. Feigenbaum, Bruce G. Buchanan and Joshua Lederberg, published in the journal Artificial Intelligence, Vol. 11, No. 1-2, August 1978, pp. 5-24.

Impact: DENDRAL is a pioneer project in the field of artificial intelligence, demonstrating the potential of expert systems. It lays a foundation for the future application of artificial intelligence in the field of science and shows how to encode knowledge in specific fields into computer programs to solve complex problems. The project has also promoted the development of other expert systems and the development of artificial intelligence as a research field. DENDRAL's success has proved the potential of artificial intelligence in simulating the decision-making ability of human experts, and has had a profound impact on the subsequent research and application of artificial intelligence, especially in chemical informatics, drug Discovery and other fields. It also stimulates people's research on key artificial intelligence technologies such as knowledge representation, reasoning and machine learning.

ELIZA chat robot

ELIZA is one of the earliest natural language processing programs designed to simulate human conversations and is one of the pioneers of chat robots.

Time: developed from 1964 to 1966; Published in January 1966

person: Joseph Weizenbaum, Massachusetts Institute of Technology

paper: "ELIZA-A Computer Program for the Study of Natural Language Communication Between Man and Machine" (Communications of the ACM, January 1966).

Impact: ELIZA is one of the earliest chat robots, demonstrating the potential of computers in natural language processing. By simulating the dialogue mode of psychotherapists, it enables users to have the illusion of communicating with machines, thus triggering extensive discussions on artificial intelligence, human-computer interaction and whether machines can possess human intelligence. ELIZA's simple and effective pattern matching method had a profound impact on the evolution of chat robots and dialogue artificial intelligence systems, at the same time, important issues such as artificial intelligence ethics and the possibility of establishing relationships between machines and human beings are also raised.

Pattern recognition program for self-adjusting operators

Leonard Uhr and Charles Vossler have developed an early pattern recognition program that can generate, evaluate and adjust their own operators. It is Machine Learning and Adaptive AI Systems. The important pioneer.

Time: May 1961

characters: Leonard Uhr,Charles Vossler

paper: A pattern recognition program that generates, evaluates and adjusts its own operators, published in Western Joint IRE-AIEE-ACM Computer Conference (May 1961).

Impact: This research is of great significance in the fields of Computer Vision and Pattern Recognition. The concept of self-improvement system is put forward for the first time, it lays a foundation for the subsequent development of adaptive artificial intelligence algorithms. The program was able to modify its own operations according to performance feedback, which was revolutionary at that time and had a profound impact on the follow-up research of artificial intelligence and Cognitive Science.

Perceptron

Frank Rosenblatt put forward the concept of Perceptron in 1957 as an early Artificial Neural Network model. The ability of sensors to learn and execute binary classification tasks is an important milestone in the field of Machine Learning.

Time: 1957

character: Frank Rosenblatt

paper: The Perceptron: A Perceiving and Recognizing Automaton(Project Para),1957; The Perceptron: A Perceptron Model for Information Storage and Organization in the Brain,1958

Impact: as the first artificial neural network model that can pass trial and error learning, perceptron lays a foundation for modern machine learning algorithms and stimulates people's understanding of neural networks and Pattern Recognition. And promoted the development of artificial intelligence. Although perceptron has certain limitations in solving more complex problems, its core concept was re-valued in the 1980 s, and it is more advanced in neural network structure and Deep Learning. The development of the model has made important contributions.

Dartmouth Conference

In the Summer of 1956, a landmark conference-Dartmouth Summer Research Project on Artificial Intelligence was held at Dartmouth College in the United States. This meeting marked the formal birth of artificial intelligence (AI) as an independent discipline and laid the foundation for the future development of this field.

Time: Summer 1956 (specific time: June 18th to August 17th, 1956)

characters: main organizers: John McCarthy, Marvin Minsky, Claude Shannon, Nathaniel Rochester; other participants: about 10 scientists from mathematics, psychology, neurology, computer science and electrical engineering

paper: there was no officially published paper at the meeting, but the organizer submitted A Proposal entitled A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence in 1955, it laid the foundation for the meeting.

Impact: Dartmouth Conference had a profound impact on the field of artificial intelligence. It first proposed the term "artificial intelligence" and established the name of this field, it has aroused people's extensive attention to machine intelligence. The meeting marked the formal birth of artificial intelligence as an independent discipline, attracted a large number of scholars to devote themselves to the research of artificial intelligence, and laid a systematic framework for the development of this discipline. The meeting set the initial research direction for the subsequent development of artificial intelligence, promoting Heuristic Algorithms, Natural Language Processing and Machine Translation. Such as the development of early research projects. The scientists who participated in the meeting later became the leaders in the field of artificial intelligence, and their research had a profound impact on the later development. Although artificial intelligence has experienced fluctuations in the subsequent development process, Dartmouth Conference is undoubtedly a milestone in the development history of artificial intelligence, laying a solid foundation for the development of the whole discipline.

Turing test

Turing Test is an important experiment in the field of artificial intelligence to evaluate whether machines have the same level of intelligence as human beings.

Time: October, 1950

character: Alan Turing

paper: Computing Machinery and Intelligence, published in the October 1950 issue of Mind.

Impact: Turing test is a milestone concept in the field of artificial intelligence and provides a benchmark for measuring machine intelligence. It has stimulated extensive discussions and researches on the nature of intelligence, machine capabilities and the future development of artificial intelligence. This test has promoted the development of Natural Language Processing, Machine Learning and other fields, and triggered discussions on the definition of intelligence at the philosophical level. Turing test also inspired later generations to design various tests to evaluate the degree of artificial intelligence.

Research on the foundation of computer game

Claude E. The paper "Programming a Computer for Playing Chess" published by Shannon) in 1950 systematically put forward the theoretical framework of Computer Playing Chess for the first time, laying a foundation for the field of Computer game and artificial intelligence.

Time: March 1950

characters: Claude E. Shannon, Bell Labs.

Paper: Title: Programming a Computer for Playing Chess; Published: March 1950; Journal: Philosophical Magazine, Series 7, Volume 41, Issue 314, page 256-275.

Impact: Shannon's paper lays a theoretical foundation for the field of computer game, and puts forward key concepts such as chess evaluation function and search Algorithm (such as Minimax Algorithm and Minimax Algorithm), these concepts are still the core components of modern chess engines and other game AI systems. The research not only directly affected the development of early computer chess programs and milestone projects such as "dark Blue", but more importantly, it stimulates the research on complex decision-making problems in the field of artificial intelligence, promotes the development of search algorithms, heuristic methods and Game Theory, applications in decision support and other fields have had a profound impact and provided important theoretical references for the development of modern AI systems such as AlphaGo and AlphaZero.

The above content is transferred from Andy730