2022年1月20日

Five anticipated trends in AI in 2021

Five anticipated trends in AI in 2021

World tide of scientific and technological innovation

◎ Our reporter Liu Xia

Artificial intelligence (AI) has become a hot topic in the field of science and technology in many countries. Governments and companies around the world, not to be outdone, are pouring money into the field and all kinds of innovations are springing up.

In addition, the COVID-19 epidemic has forced us to rely even more on technology, online activities, and artificial intelligence. Among them, artificial intelligence is particularly important for enterprises, which can achieve personalized services on a large scale, while meeting customers’ ever-improving experience needs.

In a report published on March 15, Forbes Biweekly listed five trends we are looking forward to in artificial intelligence in 2021. These include the proliferation of low-code/no-code tools, and the increasing accessibility of artificial intelligence that makes it easy for children to create their own.

Low code/no code tools

Automated Machine Learning (AutomL) is not a new thing. In 2020, Huawei is hiring a doctor of machine learning with an annual salary of $1 million, and one of its research fields is AutomL.

Machine learning allows algorithms to automatically find out a set of rules from the data so as to extract relevant features from the data. With the development of machine learning, more and more parts need to be intervened by human beings, while AutoML automates the whole process of machine learning model from construction to application.

While AutoML can build high-quality AI models without solid data science knowledge, the low-code/no-code platform goes one step further — it can build entire production-level AI-driven applications without deep programming knowledge.

Low-code/no-code tools have been a global phenomenon since last year, with applications ranging from building applications to vertical AI solutions for the enterprise expected to continue this year.

Low-code/no-code tools will be the next frontier for tech giants, a $13.2 billion market that is expected to grow to $45.5 billion by 2025, according to data.

The best proof of this is the Honeycode platform launched by Amazon in June 2020. Honeycode is a code-free development environment similar to a spreadsheet interface, which has been called a “blessing” for product managers.

Advanced pre-training language model

The Bidirectional Encoder Representation from a Converter (Bert) is a new language model developed and released by Google in late 2018. A newcomer to the field of natural language processing (NLP), Bert encapsulates the major NLP advances of the past few years, wounding out his competitors on his first appearance, setting new records in 11 NLP tests and even surpassing human performance.

In recent years, pre-trained language models similar to the Bert model (such as question and answer, named entity recognition, natural language reasoning, text classification, etc.) have played an important role in many NLP tasks.

These pre-trained language models are very powerful and have revolutionized language translation, understanding, summarization, and so on, but they are expensive and the training is time consuming.

The good news is that advanced pre-training models can lead to a new generation of AI services that are efficient and easy to build.

GPT-3 is the best of them all! It is a natural language processing model created by OpenAI with huge investment. With 175 billion super parameters, it is the strongest AI model in the NLP field. Since its first launch last May, the GPT-3 has continued to gain traction across major media platforms thanks to its impressive text generation capabilities. Not only can it answer questions, write articles, write poems, translate articles, but it can also generate code, do mathematical reasoning, data analysis, draw charts, make resumes, and even play games, and the effect is surprisingly good.

Synthetic content generation

Algorithmic innovation in artificial intelligence is not limited to NLP. Generative adversarial networks (GANS) are also spawning innovations that showcase the extraordinary achievements scientists have made in creating art and fake images.

GANS, first proposed by AI scholar Ian Goodfellow at the University of Montreal in Canada, is also complex to train and adjust because they require large data sets to train.

But the scientists’ innovations have greatly reduced the amount of data needed to create GANS. Nvidia, for example, has demonstrated a new way to train Gans more efficiently that requires less data than previous methods. This allows GANS to be used in a wide range of fields, from medical applications, such as the synthesis of cancer histological images, to deeper “Deep fakes”.

“Deep Fake” is a high-powered hack that uses the latest artificial intelligence technology to allow people to edit video clips into anyone’s face using a computer. “The so-called success is nothing, the failure is nothing”, the video “face change” has aroused great attention, but also caused a huge controversy. Just five days after it went live, the dark technology was rejected by the entire network and later banned worldwide.

Artificial intelligence for children

With the popularity of low-code tools, AI creators are also getting younger. A schoolboy can now create artificial intelligence for his own use — from categorizing text to drawing images. American high schools already offer A.I. courses, and middle schools are not far behind.

At the Synopsys 2020 science fair in Silicon Valley, for example, 31 percent of the winning software projects used AI in their innovations. Even more impressive, 27 percent of these AIs were created by students in grades 6 through 8. One of the winners was an eighth-grader who created a convolutional neural network that can detect diabetic retinopathy through eye scans.

Machine learning operation

Machine learning operations (MLOPs) are a relatively new concept in the field of artificial intelligence that involves the best management data scientists and operators to effectively develop, deploy, and monitor models.

In 2020, due to the COVID-19 epidemic, huge changes in operational workflow, inventory management, traffic patterns, etc., caused many AI to show unexpected behavior, which is called drift — input data does not match the expectations of the AI training.

While companies deploying machine learning in production have faced challenges such as drift before, the COVID-19 epidemic has increased the demand for MLOPs. Similarly, with the implementation of privacy regulations such as the California Consumer Privacy Act of 2018, there is a growing need for governance and risk management for companies that operate on customer data. According to data, the market for MLOPs is expected to reach $4 billion by 2025.

These are not all new trends in AI, but they deserve our attention because they highlight three important aspects. First, there is the increasing use of AI in the real world, as evidenced by the problems caused by COVID-19 and the growth of MLOPs. Second, people are coming up with new ideas in the field, just as Bert and Gans followed one another. Finally, the threshold for the creation of artificial intelligence is getting lower and lower, laying a solid foundation for its “flying into ordinary people’s homes”.

The ideal and future of artificial intelligence is always good, but despite all the above innovations, we still need to promote and guide its development in a down-to-earth way, so that it can better benefit mankind.

Leave a Reply

Your email address will not be published. Required fields are marked *

Scroll Up