Machine Hearts released
Machine Heart Editorial Department
In a flash, 2021 is gone. This year, the global economic and social life of the COVID-19 pandemic has brought many challenges, and at the same time, the power of science and technology has continued to emerge. The speed of technological progress and industrial development has further accelerated, and digital technology and intelligent technology have brought more imagination space to people's lives.
Standing ahead of the starting point of 2022, what are the technology trends that deserve our attention? Recently, Baidu Research Institute officially released the 2022 Technology Trend Forecast, covering ten major directions including AI core technology, interdisciplinary and cross-field research, and industrial and social value. In the high-speed changing and unknown world of science and technology, we will explore more certain values and directions.
All along, Baidu Research Institute has continued to focus on the forward-looking basic research of artificial intelligence and explore the frontier direction of technology, and the research field almost covers the entire field of AI from the underlying foundation to the perception and cognitive technology. Thanks to the continuous deepening of the FIELD of AI over the years, baidu Research Institute's Science and Technology Trend Report has a strong "AI attribute", and also pays more attention to the landing value of the AI field, providing guidance for every participant in this era of rapid change.
In the past 2021, Baidu has been at the forefront of technology exploration and application landing. Joint Pengcheng Lab Released the World's First Knowledge-Enhanced 100 Billion Model Pengcheng - Baidu · Launched the industry's first mRNA vaccine sequence design algorithm LinearDesign, which can find a stable vaccine sequence within ten minutes; at the same time, the open source platform Flying Propeller with the deep learning framework as the core continues to reduce the threshold for the development of AI technology, and universal AI is no longer only about the majority of AI practitioners, but also about a wider range of AI technology beneficiaries.
What is the evolution trend of hyperscale pre-trained models in the next 2022? How will it become more practical? What cross-cutting areas will AI technology bring more surprises to? In which areas does it stand out for its landing value? All of these will find some answers in this report.
preface
In the era of steady progress, we use technology to explore the "unknown" of the world; in the era of uncertainty, we use technology to anchor the "certainty" of the world.
At the moment, our mission is more inclined to the latter. In the face of the 2022 year that has arrived, the impact of the new crown epidemic continues to bring, and the global economy has ushered in many challenges, technology has become an important force leading innovation and development. The artificial intelligence that integrates the development trend of technology and industrial value creation, the rapid evolution of core technologies, the enhancement of cross-field connectivity, and the increasingly solid industrial base are precisely such scientific and technological forces.
First of all, the AI core technology continues to make breakthroughs and evolutions, and the integration and innovation are becoming more and more significant. Taking pre-trained large models as an example, technical paths such as knowledge enhancement, cross-modality, and cross-language have activated the imagination space of large models, and even ignited the hope of general artificial intelligence.
Second, in interdisciplinary and interdisciplinary research, AI constitutes a common variable in scientific research and technological development. In the fields of quantum, biology, chemistry, etc., "+AI" has become one of the most exciting research directions.
Third, in terms of value creation, AI is promoting the development of autonomous driving, robotics, aerospace, life and health, and other fields, and plays an important role in achieving goals such as "double carbon" and scientific and technological inclusion.
Today, Baidu Research Institute shares with you the 2022 technology trend prediction, hoping that in the uncertain era, AI will be used as a lamp to illuminate the road to innovation; with AI as a paddle, rowing the boat of development.
——Baidu CTO, Baidu Research Institute President Wang Haifeng
1. The hyperscale pre-training model shows the trend of knowledge enhancement, cross-modal unified modeling, and multi-learning methods to evolve together, and gradually becomes practical
The large model performs self-supervised learning based on massive data, uses a unified model and paradigm to solve various AI tasks, breaks the dependence of traditional technology on large-scale annotation data, and significantly improves the effect, versatility and generalization of AI models.
It is expected that in 2022, the direction of large model research and development will change from continuously increasing the scale of parameters to practicality, based on knowledge enhancement, cross-modal unified modeling, prompt learning, continuous learning, combined with model distillation, sparseness and other technologies, the effect, versatility, generalization, interpretability and operational efficiency of large models will continue to improve, and the application threshold will continue to be reduced, so as to achieve a wide range of landing in the Internet, intelligent office, smart finance and other scenarios. For example, AIGC (AI generated content, artificial intelligence to create content), with the help of large models of cross-modal comprehensive technical capabilities, can stimulate creativity, enhance content diversity, reduce production costs, will achieve large-scale application.
2.AI emerging research areas for Science are expected to bring about a paradigm shift in research
Machine learning has helped mathematicians discover two big conjectures, and solved real-time simulation problems of superscale quantum stochastic circuits using a combination of machine learning, multiscale modeling, and high-performance computing, showing the great potential of artificial intelligence to be applied to scientific research, processing data, designing new types of experiments, and creating more efficient computational models.
The emerging AI for Science is expected to promote the deep integration of two major scientific paradigms of data-driven and theoretical deduction. It is expected that in the next few years, AI will be further deeply integrated with different fields such as mathematics, physics, chemistry, materials, and engineering, and play a greater role in the progress of basic science.
3. AI-based biocomputing will continue to develop at a high speed, and new breakthroughs will be achieved in collaborative innovation in basic research and application scenarios
In the context of human society is still fighting against the new crown virus, the demand for technological innovation in the life and health industry is more urgent. AI enables gene editing to find targets more accurately and quickly, and AI has helped make significant breakthroughs in protein structure prediction. The success of the new crown mRNA vaccine technology has brought about an outbreak of drug design and vaccine research and development based on macromolecular drugs such as RNA and proteins, and international mainstream pharmaceutical companies have accelerated the landing of mRNA technology.
In the future, AI-based biocomputing will also make breakthroughs in more basic research and application scenarios, such as protein-based drug design, synthesis, and screening, anti-cancer drugs based on mRNA technology, monoclonal antibodies, immunotherapy, etc. The deep integration of the two will significantly shorten the drug research and development cycle, reduce the cost of research and development, and promote precision medicine and personalized diagnosis and treatment.
4. Privacy computing technology is in the spotlight and will become a breakthrough in the release of data value and an infrastructure for building trust
With the increasingly sound global personal information and data security regulations, it has become the industry consensus that security compliance is the premise of promoting the effective release of data value.
Privacy computing technologies, represented by trusted confidential computing and federated computing, have attracted much attention because they take into account data security protection and data sharing and circulation from a technical point of view. With the improvement of the performance of privacy computing technology, the mutual promotion of technology and compliance standards, and the synergy of multiple parties to enhance the credibility of technology, relevant typical applications will appear in scenarios such as biocomputing, financial analysis and data transactions.
In the long run, privacy computing technology may promote data circulation and computing based on dense forms as the default option, and gradually become an infrastructure for building trust.
5. Quantum software and hardware integration solutions have become the mainstream trend, and the actual demand accelerates the integration and innovation of quantum computing and various industries
It is expected that in 2022, the design, preparation and measurement and control technology of quantum chips will continue to develop, and the number of qubits will increase in scale, and a breakthrough will be sought along the two ideas of reducing noise or adapting to noise. Quantum software and services are developing across platforms, users will get richer quantum back-end choices on cloud-native quantum computing platforms, and quantum platforms carrying quantum software-hardware integration solutions will gradually show their application value.
With the deep integration and innovation of quantum computing and intelligent manufacturing, artificial intelligence, chemical medicine, financial technology and other fields, a number of practical application solutions with significant quantum advantages will be produced. Government agencies, research institutes and industry will also work more closely together to build high-quality quantum devices, cultivate quantum science and technology talents, and initially open up the quantum computing industry chain.
6. Autonomous driving technology has entered a new stage of unmanned landing, and multiple "automotive robots" are constantly emerging, connecting technologies and scenarios
In 2022, under the dual impetus of policies, regulations and technological progress, automatic driving will soar in unmanned, and the automobile form represented by diversified "automotive robots" will develop rapidly.
Through passenger cars, buses, trunk logistics, warehousing and distribution, special operations in mines and ports, retail, sanitation and other rich scene applications, diversified "automotive robots" will provide services to users more widely, create value for customers, and then gradually achieve stable commercial benefits, promote the development of science and technology and social progress.
7.AI the integration and innovation of technology and aerospace science and technology, and promote deep space exploration to a new stage of intelligence
Deep space exploration carries human curiosity and reverie about the universe and itself. Realizing the residence of the moon and planets, carrying out scientific exploration and resource development and utilization as the main program, and carrying out deep space exploration in distant and unknown environments, the demand for the autonomy of probes is increasingly strong.
The field of construction machinery automation has realized the actual engineering scenario of 24-hour continuous unmanned excavation operation, and the relevant AI algorithms such as autonomous environment perception and motion planning will also enable the detector to have functions such as autonomous obstacle avoidance and decision-making, and flexible and autonomous operation of the robotic arm. In addition, AI technology is also expected to play an important supporting role in spacecraft fault detection and repair, construction digital twin simulation laboratory, deep space big data detection and analysis, etc.
8. "Social distancing" accelerates the symbiosis of man and machine, and supports the rapid integration of virtual and real technology and intelligent interaction technology into production and life
The COVID-19 pandemic has set up "social distancing" for people's communication, and the development of digital technology has allowed us to shorten this distance and accelerate the symbiosis between humans and digital people and robots. The future world of virtual and real combined with intelligent interaction is no longer far away from us.
Supporting this change is the continuous progress of AI technologies such as vision, speech, natural language processing, and XR in cross-modal understanding and generation, continuous learning, and the cross-technical support system formed by integrating hardware, networking, computing, ecosystem platform, and content.
With the acceleration of integration and innovation of related technologies and the maturity of the cross-technical support system, more virtual and real combinations and intelligent interactive products for industry and consumption scenarios will emerge, thereby promoting the deep integration of the digital economy and the real economy and enriching people's production and life experience.
9. Green and low carbon are more integrated into the AI blueprint to help achieve the goal of carbon peak carbon neutrality
As AI technology accelerates the integration and innovation of various industries, data centers and large-scale AI computing achieve important economic and social value, but their energy consumption and environmental impact cannot be ignored, and it is urgent to develop more environmentally friendly "green AI" technologies to reduce the energy consumption of model training and use.
In the next few years, "green AI" related technologies will continue to flourish, around the energy-efficient architecture design, training and reasoning strategies, data utilization and other construction system, the formation of performance and energy consumption evaluation standards; higher computing power, lower energy consumption AI chips will continue to emerge; leading AI companies to build intensive large computing power and large models, improve downstream performance, reduce the overall energy consumption costs; the policy will also encourage the construction of green and low-carbon data centers, the introduction of AI technology to improve infrastructure energy efficiency ratio and other initiatives.
10.AI more inclusive and inclusive, and the value creation orientation has made the needs of small and medium-sized enterprises and vulnerable groups pay more attention
Inclusive AI is not only about ai practitioners at large, but also about the broader beneficiaries of AI technologies.
The open source platform with the deep learning framework as the core has greatly reduced the threshold for the development of AI technology, and public data sets, large model bases, and regional intelligent computing centers will be further developed to help small and medium-sized enterprises achieve cost reduction and efficiency increase, and stimulate innovation vitality. The national AI training system will also be gradually built to promote the reemployment of traditional industry personnel and AI science education.
The well-being of AI should also benefit all groups in society, with the promotion of policy guidance and sustainable development of the ESG concept, the focus of enterprises will shift to value creation, AI service providers will strengthen the attention to the needs of the elderly, children and other vulnerable groups, develop corresponding inclusive AI services and products, so that everyone can enjoy the convenience of digital technology.