AI report: How to apply the core

Most of the domestic discussion of artificial intelligence is fragmentation of fragmentation, it is difficult to understand the development of artificial intelligence and technical system, it is very difficult to have practical reference significance. Deloitte DUP recently released a report detailing the history, core technologies and applications of artificial intelligence, especially the important cognitive techniques. This report will help us to learn more about artificial intelligence and cognitive technology, and help companies in various industries consider the real value of AI applications.

This report is translated by the heart of the machine, welcome to the micro-signal: The Heart of the Machine (id:almosthuman2014).

I. Overview

Interest in AI has soared in recent years, and since 2011 companies that have developed and commercialized artificial intelligence-related products and technologies have gained more than $2 billion trillion in venture capital, while technology giants are investing billions of of billions of dollars to acquire artificial intelligence start-ups. Reports are rampant, and problems such as huge investments and computer-induced unemployment are also emerging, and the assertion that computers are smarter than people and threatens human survival is widely cited and raised by the media.

IBM has pledged to set aside 1 billion dollars to commercialize their cognitive computing platform, Watson.

Google's investments in recent years have focused on artificial intelligence, such as the acquisition of 8 robotics companies and a machine learning company.

Facebook hired Yann LeCun, a leading AI scholar, to create its own artificial intelligence lab, hoping for a major breakthrough in the field.

Researchers at Oxford University published a report showing that about 47% of the work in the United States has become precarious because of automated machine-awareness technology.

The New York Times bestselling book "The Second Machine age" argues that the era of great positive changes in digital technology and artificial intelligence has come, but there are also negative effects of massive unemployment.

Elon Musk, a Silicon Valley entrepreneur, keeps an eye on AI through constant investment. He even believes that artificial intelligence is more dangerous than nuclear weapons.

The famous theoretical physicist Stephen Hawking that if the successful creation of artificial intelligence means the end of human history, "unless we know how to avoid risk." ”

Even with so much hype, there are notable business practices in the field of artificial intelligence that are or are about to have an impact on industries and organizations. Business leaders need to have a thorough understanding of the implications and trends of artificial intelligence.

Ii. Artificial Intelligence and cognitive technology

The first step in revealing artificial intelligence is to define the terminology, outline the history, and describe the core technologies of the foundation.

1, the definition of artificial intelligence

The field of artificial intelligence suffers from a variety of concepts and definitions, and some too many are not enough. Nils Nilsson, one of the founders of the field, wrote: "Artificial intelligence lacks a common definition." "An authoritative artificial intelligence textbook, now revised in version three, gives eight definitions, but the book does not reveal what kind of definition the author prefers." For us, a practical definition is that artificial intelligence is a theoretical study of how computer systems can perform tasks that can only be accomplished by relying on human ingenuity. For example, visual perception, speech recognition, decision making in uncertain conditions, learning, and language translation. Rather than studying how humans conduct their thinking activities and define artificial intelligence from the perspective of the tasks that humans can accomplish, instead of how humans think, in today's era, we can make a precise definition of intelligence around the neural-mechanism level to directly explore its practical application. It is worth mentioning that as computers are upgraded to address the challenges of new tasks, the threshold for defining tasks that need to be solved by human ingenuity is becoming more and more high. Therefore, the definition of artificial intelligence evolves over time, which is called the "artificial Intelligence effect", which is summed up as "artificial intelligence is to achieve all the current can not without the wisdom of mankind to achieve the collection of tasks." ”

2. The history of artificial intelligence

AI is not a term. In fact, this field was launched in the 1950s, and the history of this exploration is known as the "era of tumult and longing, frustration and disappointment"-a more appropriate assessment recently given.

The bold goal of artificial intelligence to simulate human intelligence was clarified in the 1950s, and the researchers carried out a series of research projects throughout the the 1960s and into the 70, which showed that computers could accomplish a series of tasks that were within the scope of human capabilities, such as proof theorems, To solve calculus, to respond to commands through planning, to perform physical actions, and even to simulate psychologists and compose such activities.

However, the simplistic algorithms, the lack of the ability to deal with uncertain environments (which are ubiquitous in life), and the limitations of computational power seriously hamper our use of artificial intelligence to solve more difficult and diverse problems. Along with disappointment at the lack of continued efforts, artificial intelligence gradually faded out of public view in the mid 1970s.

In the early the 1980s, Japan launched a project aimed at developing a computer architecture that is leading in the field of artificial intelligence. The West began to worry about losing to Japan in this area, prompting them to decide to start investing in artificial intelligence again. Commercial suppliers of artificial intelligence technology products have emerged in the 1980s, some of which have been listed, such as Intellicorp, Symbolics, and Teknowledge.

In the the late 1980s, almost half of the Fortune 500 were developing or using an "expert system", an artificial intelligence technique that simulates human experts to solve problems in the field by modeling human experts ' ability to solve problems.

The high hopes of expert system potential completely obscure its own limitations, including the obvious lack of common sense, difficulty in capturing the tacit knowledge of experts, the complexity and cost of building and maintaining large systems, and when this is recognized by more and more people, artificial intelligence research is once again out of orbit.

The 1990s in the field of artificial intelligence technology has been at a low ebb, the results are few. Instead, technologies such as neural networks and genetic algorithms have gained new attention, partly because they avoid some of the limitations of expert systems, and because new algorithms make them more efficient to run.

The design of neural networks is inspired by the structure of the brain. The mechanism of the genetic algorithm is to first iteratively generate alternative solutions, then eliminate the worst scenarios, and finally, by introducing random variables to generate new solutions, "evolve" the best solution to the problem.

3. Catalysts for the advancement of Artificial intelligence

By the end of the 10 years of the 21st century, a series of elements for the revival of artificial intelligence research, especially some of the core technologies, emerged. These important factors and techniques are described in detail below.

1 Moore's Law

Under the condition of Price and volume unchanged, the computing power of computer can be increased. This is known as Moore's Law, named after the Intel co-founder Gordon Moore. Gordon Moore profited from various forms of computing, including the type of calculation used by artificial intelligence researchers. A few years ago, advanced system design could only be theoretically set up but could not be achieved because it required too much computer resources or computer incompetence. Today, we have the computing resources we need to implement these designs. To give a fantastic example, the performance of the latest generation of microprocessors is now 4 million times times that of the first generation of Single-chip microcomputers in the 1971.

2 large data

Thanks to the Internet, social media, mobile devices and cheap sensors, the volume of data generated by the world has increased dramatically. As the value of these data continues to be recognized, new technologies for managing and analyzing data have been developed. Large data is a booster for the development of artificial intelligence, this is because some artificial intelligence techniques use statistical models to calculate the probability of data, such as images, texts, or sounds, by exposing them to the ocean of data, to optimize them, or to call them "training"--conditions that are now readily available.

3 Internet and cloud computing

Closely related to the large data phenomenon, the Internet and cloud computing can be considered to be the cornerstone of artificial intelligence for two reasons, first, they can make all networked computer devices have access to massive data. These data are needed to promote the development of artificial intelligence, so it can promote artificial intelligence. Second, they provide a workable way of cooperating-sometimes explicit-sometimes implicit-to help the AI system train. Some researchers, for example, use cloud-based crowdsourcing services like Mechanical Turk to hire thousands of people to depict digital images. This allows the image recognition algorithm to learn from these descriptions. Google translation improves the quality of its automated translations by analyzing user feedback and the free contribution of users.

4) New algorithm

An algorithm is a path method that solves a design program or completes a task. In recent years, the development of new algorithms has greatly improved the ability of machine learning, which is important in itself and is also a catalyst for other technologies, such as computer vision (this technology will be described later). Machine learning algorithms are currently being used in open source, and this situation will lead to greater progress because developers can complement and enhance each other's work in an open source environment.

4. Cognitive technology

We will distinguish between the field of artificial intelligence and the technologies that extend it. The mass media portrays AI as a computer that is as intelligent or smarter than humans. And the technology in the past only people can do a specific task on the better performance. We call these technologies cognitive technologies (pictured below), and cognitive technology is the product of the artificial intelligence field, which can accomplish tasks that only people can accomplish in the past. And they are what business and public sector leaders should focus on. Below we will introduce several of the most important cognitive technologies that are being widely adopted and progressing rapidly, and are also heavily invested.

1 Computer Vision

Refers to the ability of a computer to recognize objects, scenes, and activities from an image. Computer vision technology uses sequences composed of image processing operations and other techniques to decompose image analysis tasks into manageable small tasks. For example, some techniques can detect the edges and textures of an object from an image. Classification techniques can be used to determine whether a recognized feature can represent a class of objects known to the system.

Computer vision is widely used. This includes medical imaging analysis that is used to improve disease prediction, diagnosis and treatment; Face recognition is used by Facebook to automatically identify the characters in the picture; in the area of security and surveillance, the consumer can now photograph products with a smartphone to get more options.

As a related subject, machine vision refers to the visual application in the field of industrial automation. In these applications, computers identify objects such as manufacturing parts in highly restricted factory environments, so it is simpler to target computer vision than to seek to operate in an unrestricted environment. Computer vision is a research in progress, while machine vision is a problem that has been solved, and it is a project of system engineering rather than research level. As the scope of applications continues to grow, start-ups in the field of computer vision have attracted hundreds of millions of of dollars of wind investments since 2011.

2 machine Learning

Refers to the ability of a computer system not to follow explicit program instructions but to increase its performance by exposing it to data. The core is that machine learning is the automatic discovery of patterns from the data, which can be used to make predictions once they are discovered. For example, to give the machine learning system a database of credit card transactions, such as trading time, business, location, price, and transaction legitimacy, the system learns about patterns that can be used to predict credit card fraud. The more transaction data is processed, the better the forecast will be.

Machine learning has a wide range of applications, and it has the potential to improve all performance for activities that generate huge amounts of data. In addition to fraud screening, these activities include sales forecasts, inventory management, oil and gas exploration, and public health. Machine learning technology also plays an important role in other cognitive technology fields, such as computer vision, which can improve the ability of recognizing objects by constantly training and improving the visual model in mass images. Today, machine learning has become one of the hottest research areas in cognitive technology and has attracted nearly 1 billion of billions of dollars in venture capital over the 2011-2014-year period. Google also spent $400 million in 2014 to buy DeepMind, a company that studies machine-learning technology.

3 Natural Language Processing

Refers to the ability of the computer to have human-like text processing, such as extracting meaning from the text, and even reading it from the readable, natural, grammatically correct text. A natural language processing system does not understand how humans handle text, but it can skillfully handle text in very sophisticated and sophisticated ways, such as automatically identifying all the people and places mentioned in a document, identifying the core issues of the document, or in a pile of human-readable contracts, The various terms and conditions are extracted and made into a table. These tasks are impossible to accomplish with traditional text processing software, which can only operate on simple text matching and patterns. Consider a commonplace example that reflects a challenge to natural language processing. The meaning of every word in the sentence "Time flies" (Time flies likes arrow) seems to be clear until the system encounters the sentence "fruit flies like bananas (Fruit flies kind banana)" and substituting "fruit (Fruit)". Time, and substituting "arrow" with "banana" (banana), changes the meaning of the two words "fleeting/flying (like)" and "likes/dislikes".

Natural language processing, like computer vision technology, incorporates a variety of technologies that contribute to achieving goals. A language model is established to predict the probability distribution of language expression, for example, the maximum probability of a given character or word expressing a particular semantics. Selected features can be combined with some elements in the text to identify a piece of text, by identifying these elements can be a certain type of text from other text, such as spam and normal mail. The machine-learning-driven classification method will be the standard for filtering to determine whether a message belongs to spam.

Because context is so important to understand the difference between "time flies" and "fruit flies (Drosophila)", natural language processing technology has a relatively narrow range of practical applications, including the analysis of customer feedback on a particular product or service, Automatic discovery of certain meanings in civil litigation or government investigations, and the automatic writing of formulations such as corporate revenue and sports.

4 Robot Technology

Integrating machine vision, automatic planning, and other cognitive technologies into very small but high-performance sensors, actuators, and cleverly designed hardware, this gives birth to a new generation of robots capable of working with humans to flexibly handle different tasks in a variety of unknown environments. For example, drones, and "cobots" that can work in workshops for humans, include consumer products from toys to domestic helpers.

5 Speech recognition technology

It focuses on automatic and accurate transcription of human speech. This technique must face some problems similar to the natural language processing, with some difficulties in dealing with different accents, background noise, distinguishing between homonym ("buy" and "by"), as well as having to keep pace with normal speed. Speech recognition systems use the same techniques as natural language processing systems, supplemented by other techniques, such as acoustic models that describe sounds and their probability in specific sequences and languages. The main applications of speech recognition include medical dictation, voice writing, computer system voice control, telephone customer service and so on. Domino's Pizza, for example, recently launched a mobile app that allows users to move through voice orders.

The cognitive advances mentioned above are speeding up and attracting a lot of investment, and other relatively mature cognitive technologies are still an important part of the enterprise software system. These maturing cognitive technologies include decision optimization-automatic completion of the best trade-off for complex decisions or limited resources, planning and scheduling-to design a series of action processes to meet goals and observe constraints; rule-oriented systems--the technology that provides the basis for an expert system, A database of knowledge and rules is used to automate the process of inferring from information.

Iii. the widespread use of cognitive technology

Various economic sectors have applied cognitive technology to a variety of business functions.

1) Banking

Automated fraud detection systems use machine learning to identify behavioral patterns that portend fraudulent payment actions; voice recognition can automatically fulfill telephone customer service; Sound recognition verifies the identity of the caller.

2 Medical and health field

Half of America's hospitals use automated speech recognition to help doctors automate their medical orders, and the rate of use is growing rapidly; the machine vision system completes the analysis of mammograms and other medical effects automatically; IBM's Watson has used natural language processing techniques to read and understand a great deal of medical literature, By assuming automatic generation to complete automatic diagnosis, the use of machine learning can improve the accuracy rate.

3 Life Sciences

Machine learning systems are used to predict the causal relationship between biological data and the activity of compounds, thus helping pharmaceutical companies identify the most promising drugs.

4 Media and entertainment industry

Many companies are using data analysis and natural language generation technology to automatically draft data based documents, such as company revenue status, sporting events overview, etc.

5 Oil and gas

Machine learning is widely used in mineral resource location, drilling equipment fault diagnosis and many other aspects.

6 Public sector

The public sector has also begun to use cognitive technology for specific purposes such as monitoring, compliance and fraud detection. Georgia, for example, is digitizing financial disclosures and campaign-donation forms through crowdsourcing, and in the process they have adopted an automated handwriting recognition system.

7) Retailers

The retailer uses machine learning to automatically discover attractive cross-selling pricing and effective promotions.

8) technology companies

They are using machine vision, machine learning and other cognitive techniques to improve the product or develop new products, such as the Roomba robot vacuum cleaner, nest intelligent thermostat.

The above examples show that the potential business benefits of understanding technology are much greater than the cost savings of automation, which is mainly reflected in:

Faster action and decision making (e.g., automated fraud detection, planning, and scheduling)

Better results (e.g., medical diagnostics, oil exploration, demand forecasting)

Higher efficiency (i.e., better use of highly skilled personnel and expensive equipment)

Lower costs (e.g., automatic telephone service reduces labor costs)

Larger scale (i.e., massive tasks that people cannot carry out)

Product and service innovation (from adding new features to creating new products)

Iv. reasons for the increasing influence of cognitive technology

In the next five years, the impact of cognitive technology on the business sector will grow significantly. The reason is two, first of all, in recent years, the technical ability has the substantial progress, and is in the continuous research and development condition. Second, hundreds of millions of of billions of dollars have been invested in technology commercialization, and many companies are working to provide customized development and packaging solutions for the broad needs of the business sector to make these technologies easier to purchase and configure. While not all technology providers can survive, their efforts will work together to push the market forward. Improvements in technological performance and commercialization are expanding the range of applications of cognitive technologies that will continue over the next few years.

1, technology upgrade expanded the scope of application

There are many examples of cognitive technology striding forward. Google's voice-recognition system, for example, shows that Google, in less than two years, has raised the accuracy of speech recognition from 84% in 2012 to 98% today. Computer vision Technology has also made rapid development. If the technical standards set by computer vision Technology researchers, the accuracy of image classification and recognition has increased 4 times times from 2010 to 2014. Facebook's DeepFace technology is in the peer review report (translator Note: Peer review is an academic review process in which an author's scholarly work or program is reviewed by other experts and scholars in the same field.) is highly certain, the accuracy of the face recognition rate reaches 97%. By 2011, IBM had optimized Watson to give Watson a twice-fold answer in order to make it a success in the intelligence program, "the brink of danger." Now, IBM claims that Watson is now 2,400% more intelligent than it was at the time.

With the improvement and enhancement of technology, the scope of technology application is also expanding. For example, in speech recognition, the machine once needed a lot of training to be barely recognized in the limited thesaurus, and the medical application extended by speech recognition technology is very difficult to get true popularity. Now every month there are millions of voice searches on the internet. In addition, computer vision technology was narrowly understood to be deployed in industrial automation, but now we have seen it widely used in surveillance, security, and a wide variety of consumer applications. IBM is now expanding the use of Watson beyond competition, from medical diagnostics to medical research to financial advice and automated call centers.

Not all cognitive technologies have such remarkable development. Machine translation has a certain development, but the amplitude is very small. A survey found that from 2009 to 2012, the accuracy of translating Arabic into English only increased by 13%. Although these technologies are imperfect, they can already affect how professional organizations work. Many professional translators rely on machine translation to improve the accuracy of translations, and give some regular translations to the machine and focus on more challenging tasks.

Many companies are working to further develop cognitive technology and gradually integrate it into more products, especially enterprise-class products, to facilitate the purchase and deployment of enterprise users.

2. Large-scale investment in commercialization

From 2011 to 2014 May, more than 2 billion dollars of venture capital flowed into products and services based on cognitive technology research. Meanwhile, more than 100 companies have been merged or acquired, some of them by internet giants such as Amazon, Apple, Google, IBM or Facebook. All of these investments are nurturing a diverse corporate atlas that is accelerating the commercialization of cognitive technologies.

Here, we do not provide details about a company's commercialization of cognitive technology, and we want to show that cognitive technology products are rich in diversity. Here is a list of companies dedicated to the commercialization of cognitive technology, a list that is neither intact nor fixed, but a dynamic indicator of how markets are driven and nurtured.

Data management and analysis tools are mainly used in natural language processing, machine learning and other cognitive techniques. These tools use natural language processing to extract meaning from unstructured text or machine learning to help analysts discover deep meanings from large datasets. Companies in this area include context relevant, a large data mining and analytics company in the United States, and Palantir Technologies, a company that says it wants to connect data, technology, human and environment. and Skytree: A big Data company that uses machine learning for market analysis and decision making.

The various parts of cognitive technology can be integrated into various applications and business decisions, respectively, to enhance the function and improve the efficiency of the role. For example, Wise.io offers a set of modules to promote business decisions, such as customer support, marketing, and sales, which use machine learning models to predict which customers are more likely to be lost and which are more likely to be converted. Nuance offers a voice-recognition technology to help developers develop mobile apps that require voice control.

Single point of solution. The hallmark of many cognitive technologies is that they are being constantly embedded in solutions to specific business problems. These solutions are designed to be more effective than the company's original solutions and require little expertise in cognitive technology. Application areas with high popularity include advertising, marketing and sales automation, forecasting, and planning.

Technology platform. The purpose of the platform is to provide the foundation for a highly customized business solution. They provide a range of capabilities, including data management, machine learning tools, natural language processing, knowledge representation and inference, and a unified framework for integrating these customized software.

3. Emerging applications

If the performance and commercialization trends of these technologies continue to evolve, we can boldly predict that the application of cognitive technology will be wider and accepted much more. The influx of hundreds of millions of of dollars into these companies, based on machine learning, natural language processing, machine vision or robotics, bodes well for many new applications coming into the market. We also see tremendous scope in the way business organizations rely on cognitive technology to build automated business processes and enhance products and services.

V. The application path of cognitive technology in enterprises

Cognitive technology will become popular in the next few years. In the next 2-5 years, technological advances and commercialization will expand the impact of cognitive technology on enterprises. More and more companies will find innovative applications to significantly improve their performance or create new features to enhance their competitive position. The IT departments of the enterprise can now act to increase their understanding of these technologies, assess the opportunities to apply them, and report the potential value of these technologies to the leadership. Senior business and public sector leaders should think about how cognitive technology will affect their departments and the company as a whole, how these technologies will inspire innovation and improve business performance.

(Responsible editor: Mengyishan)