Shulou(Shulou.com)06/02 Report--

Since artificial intelligence began to grow, its storm has been continuous, especially in recent years, artificial intelligence has entered a golden period of development, and the storm of public opinion is wave after wave. "how far is artificial intelligence from threatening human beings?"eight phenomena demonstrate that artificial intelligence threat theory really exists." AI warning! why do tech bigwigs stand in line with AI threat theory?

But no matter how much public opinion pushes us, we have to face the fact that "AI really makes us live a good life!" And it will continue to be "good" Artificial intelligence will continue to exert its power in the coming year.

The following is the development trend of artificial intelligence in 2019. Are you ready for it?

1. New technology to realize partial task automation

In 2019, automation will be carried out in phases. Although it is still a long way from full automation, many workflows and tasks are partially automated. Based on existing technology, 5 per cent of occupations are expected to be fully automated and 60 per cent can achieve 30 per cent automation, according to McKinsey.

We have seen many products and services that rely on computer vision and voice technology, and we will see more in 2019. The continuous development of language models and robotics will achieve more comprehensive textual and physical solutions. Competition will drive enterprises to implement some automation solutions, while the success of automation projects will promote the development of full automation.

two。 Artificial intelligence in the enterprise will be based on existing analytical applications

Over the past few years, the company has focused on building processes and infrastructure to unlock different data sources in order to improve the analysis of critical tasks, including business analysis, recommendation, personalization, prediction, anomaly detection, and monitoring.

In addition to using visual and voice technologies, we look forward to deepening deep learning into the company-owned data and machine learning areas. For example, injecting systems into time and geospatial data through deep learning, resulting in a cocoa scalable and more accurate hybrid system (that is, a system that combines deep learning with other machine learning methods)

3. UX / UI design will become crucial

The current AI solution is the collaboration of consumers, human workers, and experts in the field. These systems increase the productivity of users and enable them to perform tasks with incredible scale and accuracy. Rich user experience or user interface design can not only simplify the operation of tasks, but also directly enhance the loyalty of users to a product.

4. Hardware will become more specialized for sensing, model training and model reasoning

Deep learning became popular in 2011, shaping the Genesis model in terms of voice and computer vision. Today, there is enough evidence to justify professional hardware-Facebook alone makes trillions of predictions a day. Google also has enough scale to justify the dedicated hardware it produces. Google has been using the Tensor processing Unit (TPU) since last year. As a result, more proprietary hardware will begin to emerge in 2019. More companies in China and the United States will develop hardware for model building and reasoning based on data centers and edge devices.

5. Hybrid models are still important.

Although deep learning has been going on, most end-to-end solutions are implemented in hybrid systems. In 2019, other components and methods will continue to emerge, including model-based methods, such as Bayesian reasoning, tree search, evolution, knowledge graph, simulation platform, and so on. We may see more machine learning that is not based on neural networks.

6. The investment will be spent on new tools and processes

We are in an era of highly empirical machine learning. ML development tools need to consider the importance of data, experimental, and model search, as well as model deployment and monitoring. With only one step: model creation, the company can begin to develop tools such as data inheritance, metadata management and analysis, efficient utilization of computing resources, efficient model search and hyperparameter tuning. In 2019, we look forward to new tools to simplify the development and actual deployment of AI and ML products and services.

7. The challenge of machine deception will increase.

At present, we are still in the early stage of machine-generated content (image, video, audio, text, etc.), and there is an endless stream of news about "machine deception". At present, detection and forensics technology has been able to retrieve false videos and images. But the tools used to generate fake content are still improving rapidly, so we must ensure that detection technology keeps pace.

Machine deception is not only deceiving human machines, but also machine deceiving machines (robots) and human deceiving machines (troll army and click farm). The dissemination of information and click farms will continue to play a role in the ranking system of deceptive content and retail platforms, and we must solve this problem as soon as possible before new forms of machine spoofing are developed.

8. Automation needs the guarantee of security.

What is encouraging is that both researchers and practitioners are interested in privacy, fairness and ethical issues. As artificial intelligence begins to go deep into critical applications, the improvement of automation efficiency needs the guarantee of security and reliability. The increase in machine deception on online platforms, as well as recent accidents involving self-driving cars, have completely broken the problem. We look forward to a more in-depth discussion of security issues in 2019.

9. Accessing more data helps to take advantage of ungenerated data

Because many of our dependency models (including deep learning and reinforcement learning) lack data, it is the large companies and countries that have a lot of data that are more likely to win in the field of artificial intelligence. However, as the services that generate tagged datasets begin to use machine learning, in some areas, new tools such as generating countermeasure networks (GAN) and simulation platforms can provide real comprehensive data for training machine learning models. With new security and privacy technologies, companies can take advantage of the data they create. So smaller companies can use machine learning and artificial intelligence to win competitiveness.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.