"Narrow AI" is a special artificial intelligence that efficiently completes tasks in certain specific fields, such as recognizing the content in the picture or searching a large number of medical clinical cases to provide treatment recommendations for doctors. Currently available The essence of artificial intelligence is that people come to propose goals, and machines analyze large amounts of data to find answers efficiently.
Classification of artificial intelligence applications
In many cases,
artificial intelligence cannot give a 100% correct answer (in fact, humans are the same), how to find the problem that artificial intelligence is good at solving has become the primary task.
Artificial intelligence applications can be divided into three categories:
Core business, failure is unacceptable. Medical, banking, legal.
For core business, the failure rate is acceptable. Autonomous driving, natural language understanding.
Non-core business, not sensitive to failure. Used to improve user experience.
From the perspective of the development and application of artificial intelligence, through the simulation of perception, help humans make decisions until they completely replace humans in dealing with large amounts of repeated data.
On the other hand,
artificial intelligence driven by huge commercial benefits will soon become a reality, and autonomous driving commercial applications will bring objective commercial value such as:
Man-made traffic accidents are reduced, insurance costs are reduced, unattended driving, the cost of car use is reduced to one-fifth
If you use a car on demand, the number of cars will be reduced to one third, leading to a change in the car business model
Changes in vehicle flow have saved a lot of road and parking area, leading to changes in urban planning.
Manual + intelligent is the best combination
Did Kasparov and Li Shishi really lose to the machine?
Opposite human players, it is the intelligent flow algorithm that artificial intelligence brings together all human wisdom and experience. If this is the case, humanity will definitely lose.
But conversely, what if humans also have an artificial intelligence assistant to play? The outcome is unknown.
After defeated by Deep Blue, Kasparov launched a freestyle chess game, which can use artificial + intelligent (Centaur players) to participate in the game. Artificial intelligence gives advice, and humans decide whether to adopt the advice. In 2014, the freestyle chess match was won by 42 centaur players and 53 games. The current best chess team is composed of humans and artificial intelligence. Since artificial intelligence can help humans become the best chess players, it can be speculated that artificial intelligence can also help humans become the best doctors, pilots, judges and teachers, even O&M and developers.
How AI works and the kinds of problems it solves
When
DevOps meets AI, the golden age of intelligent operation and maintenance
Typical machine learning extracts features through unsupervised learning and supervised learning, and then through machine learning algorithms,
Realize grouping based on common features, draw a prediction model, and label new data with the prediction model.
Machine learning can solve four types of problems based on data: logical reasoning prediction, planner, communicator, experience and emotion
A blog from Ajit summarizes 12 problems that AI is good at solving
Domain experts: simulation field experts give advice
Field expansion: Give new insights and new methods.
Complex planner: easier to optimize than non-AI algorithms
Better communicators: intelligent agents, automatic language translation
New perception capabilities: machine vision produces autonomous vehicles
Enterprise AI: Improve business processes
ERP AI: Enhance ERP through cognitive systems
Prediction of cross-border impacts: For example, autonomous vehicles reduce the demand for drivers; man-made traffic accidents reduce insurance premiums; on-demand car consumption leads to changes in the business model of car companies, changes in vehicle flow, and changes in city planning.
At present, the algorithm and hardware problems cannot be solved well: speech recognition reaches the ability of people.
Better expert system: gain knowledge through unsupervised learning of data
Ultra-long sequence pattern recognition: time series prediction model
Sentiment analysis: predicting changes in human emotions through behavior
Operation and maintenance development process and the role of artificial intelligence
The operation and maintenance industry has gone through the initial, specialized, instrumentalized, platformized, cloudized and intelligent processes. From the manual operation and maintenance stage, there is basically no data, to the trend of large-scale structured data and intelligent unstructured data.
In the early stage of the development of artificial intelligence, it acted as an assistant to assist humans, with the goal of increasing sales, improving user experience, optimizing production processes and saving costs.
Manual operation stage
Operation and maintenance workload The main job of the operation and maintenance personnel is to look at the monitoring screen. As the requirements for operation and maintenance increase, the division of work occurs at this stage, resulting in stable, convenient, reliable, and fast working principles.
What artificial intelligence can do is: Based on human experience, perform business intelligence analysis (BI) on structured sales data to find out the knowledge in the data, thereby increasing sales. The main problem is that data experts discover knowledge in business data based on experience, and the degree of understanding of the business becomes the biggest bottleneck of BI effectiveness. The lack of talents who understand business rules and data mining hinders the
development of business intelligence.
Scale stage
With the introduction of the DevOps concept, a large number of tools have emerged to assist the operation and maintenance work. The operation and maintenance capabilities have been greatly improved. The problem is that few companies can produce tools that cover all DevOps life cycles, and learn a variety of tools from different vendors to complete the task. Bring a high technical threshold. With the rise of some entrepreneurial companies and the explosive growth of O&M workloads, SRE also emerged during this period to ensure business continuity. The main goal is to use software engineering technology to achieve substantial business growth while O&M work has remained stable.
What artificial intelligence can do is: there are industrial-grade solutions based on structured data, commercial-based problems are solved using algorithms, and the typical problem is to increase the utilization rate of personnel and accelerate the creation of value.
At the same time, there is also the problem of how much efficiency improvement of applying industrial-grade intelligent solutions is difficult to estimate and it is difficult to track and optimize when data knowledge changes.
Ecological stage
With the development of the scale of the Internet, a few large companies have undertaken the work of infrastructure, and through highly concentrated operations and maintenance efficiency has increased several times (purchasing a $1 infrastructure on Amazon can bring the same calculations as a $7 investment in a traditional data center Force), this change allows cloud computing customers to focus on business development and hand over infrastructure operation and maintenance to cloud computing platforms. The market continues to grow. A company cannot use a set of solutions to cover the needs of all market segments, resulting in ecologicalization. Therefore, a large amount of data lays the foundation for the practical application of artificial intelligence.
What artificial intelligence can do is: There is a general technical framework based on non-institutional data. Different companies are responsible for a part of the problem to form an ecosystem, assist business personnel to complete the work, and complete the previous manual work through new perception capabilities semi-automatically or automatically.
How to combine new perception capabilities to help humans make decisions in the huge amount of data and changing laws has become a new problem.
Analyze DevOps from a contradictory perspective
The essence of DevOps is to solve the problem of contradiction and unity
There are two contradictory aspects of DevOps. What we do is nothing more than two, and ultimately two into one
This is the first time that Western DevOps methodologies have been combined with Chinese contradictions. In fact, the so-called methodologies are either considered nonsense (general principles) or not understood (too profound). Let’s take a look at what it means to divide into two and into one.
Let’s put aside the definition of DevOps first. Assuming what DevOps is going to do, he is like a coin thrown by the referee at the beginning of a football game with the front or back facing up to decide which side will serve first. Advantages, but both sides recognize the coin as a way acceptable to both parties to start a game. This is the low-cost communication and coordination role played by DevOps in R&D and operation and maintenance.
It is very interesting that as
DevOps theory proposes a large number of tools (coins) emerge, these tools only provide rules that are more complicated than coin tossing. And artificial intelligence will bring enhanced effects to these tools.
Can't blindly pursue one aspect of things and ignore the other
When DevOps meets AI, the golden age of intelligent operation and maintenance
We look back to see what it means to divide into two.
The R&D pursuit of functional throughput mainly focuses on the demand realization time, release frequency and deployment lead time. While O&M pursues stability, it mainly focuses on deployment success rate, application error rate, accident severity and serious bugs. This was originally an irreconcilable contradiction.
However, from a higher perspective, only good throughput or stability can not bring performance improvement, experience improvement and business success. When we determined the common goal of O&M and R&D--that is, the success of the business, the problem becomes: For common business success, R&D and O&M will not blindly seek throughput or stability in the DevOps collaboration process .
Why is artificial intelligence promising in DevOps?
DevOps can get almost all types of data
The index system framework comes from "Lean Software Measurement"
We understand that the problems solved by artificial intelligence are all based on data, so with the value, efficiency, quality and ability of indicators and data, we can solve the problems through artificial intelligence in the DevOps process.
Find out that artificial intelligence can improve problems in the entire life cycle of DevOps
There are still many tools in the DevOps life cycle that cannot be automated. These processes often involve a lot of manpower and communication costs, and there are many scenarios where information is insufficient to make a good decision. In these scenarios, artificial intelligence can be based on a large number of previous The model of data training gives suggestions, so as to give a working method that can be recognized by R&D and operation and maintenance, improve work efficiency and improve work quality.
