Artificial intelligence is a branch of computer science. It attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a similar way to human intelligence. Regarding the emerging industry of artificial intelligence AI, many people have many misunderstandings. Today, the editor has collected some knowledge about artificial intelligence AI for everyone to popularize science.
• What is artificial intelligence
• Seven misunderstandings about artificial intelligence AI
• Artificial intelligence in the future
What is artificial intelligence ?
If you are a business executive (rather than a data scientist or machine learning expert), you may have been exposed to artificial intelligence from mainstream media reports. You may have read related articles in “The Economist” and “Forbes”, or read stories about Tesla’s autonomous driving, or Stephen Hawking’s article about the threat of AI to humans, and even read about artificial intelligence and human intelligence. Caricature. Therefore, if you are an executive who cares about the development of your company, these media reports about AI may raise two annoying questions: First, is the commercial potential of AI true or false? Second, how can AI be applied to my products? The answer to the first question is yes, AI has commercial potential. Today, companies can use AI to change automated operating processes that require human intelligence. AI can increase the workload of human-intensive companies by 100 times while reducing unit economic efficiency by 90%. It takes a little more time to answer the second question. First, we must eliminate the AI myth promoted by mainstream media. Only by eliminating these misunderstandings can you have a framework for how to apply AI to your business.
Seven misunderstandings about artificial intelligence AI
Myth 1: AI is magic
Many mainstream media describe AI as magical as if we only need to applaud senior magicians from big companies such as Google, Facebook, Apple, Amazon, and Microsoft. This description is unhelpful. If we want companies to adopt AI, then we need to make entrepreneurs understand AI. AI is not magic. AI is data, mathematics, model, and iteration. In order for AI to be accepted by enterprises, we need to be more transparent. The following are explanations of 3 key concepts related to AI:
Training data (TD): Training data is the initial data set for machine learning. Training data includes input and pre-answer output, so machine learning models can find patterns for any given output. For example, the input can be a customer support ticket with an email thread between the customer and a corporate support representative (CSR), and the output can be a category label from 1 to 5 based on a company-specific category definition.
Machine Learning (ML): Machine learning is software that can learn patterns from training data and apply these patterns to new input data. For example, when you receive a new customer support ticket with an email thread between the customer and the CSR, the machine learning model can predict its classification and tell you its confidence in the prediction. The main feature of machine learning is that it learns new rather than applying inherent rules. Therefore, it can adjust its rules by digesting new data.
Human-in-the-Loop (HITL): Human-in-the-Loop is the third core element of AI. We cannot expect machine learning models to be absolutely reliable. A good machine learning model may only have 70% accuracy. Therefore, when the confidence of the model is low, people need to use the Human-in-the-Loop workflow.
So, don’t be fooled by the myth that AI is magic. The basic formula for understanding AI is: AI=TD+ML+HITL.
Myth 2: AI is only for the technical elite
Media reports can easily give people the illusion that AI only belongs to the technical elite-big companies such as Amazon, Apple, Facebook, Google, IBM, Microsoft, Salesforce, Tesla, Uber-only they can form a large team of machine learning experts, and receive a billion-dollar investment. This concept is wrong.
Today, you can start applying AI to your business without $100,000. So, if you are one of the 26,000 companies in the U.S. whose revenue is greater than $50 million, you can invest 0.2% of the revenue in AI applications.
Therefore, AI is not exclusive to technical elites. It belongs to every business.
Myth 3: AI is only to solve billion-dollar problems
Mainstream media tend to report on futuristic things, such as self-driving cars or unmanned aircraft used to deliver express delivery. Companies like Google, Tesla, and Uber have invested tens of billions of dollars in order to seize the leading position in the future driverless car market due to the “winner takes all” mentality. These give the impression that AI is only used to solve new problems at the billion-dollar level. But this is another mistake.
AI is also used to solve existing smaller problems, such as million-dollar problems. Let me explain: The core requirement of any company is to understand the customer. This is the case from the agora market in ancient Greece and the personal trading square in ancient Rome. This is also true today, even if business transactions have moved explosively to the Internet. Many companies sit on a treasure trove of unstructured data from customers, which comes from email threads or Twitter comments. AI can be applied to these classifications to support ticket challenges, or to understand the sentiment of tweets.
Therefore, AI can not only be applied to new and exciting problems at the billion-dollar level, such as self-driving cars. AI is also used for existing “uninteresting” small problems, such as better understanding of customers by supporting ticket classification or social media sentiment analysis.
Myth 4: Algorithms are more important than data
Reports on AI in mainstream media tend to believe that machine learning algorithms are the most important element. They seem to equate algorithms with the human brain. They imply that it is the algorithm that makes the magic work, and that more sophisticated algorithms can surpass the human brain. Reports about machines defeating humans in Go and Chess are examples. The media is concerned about “deep neural networks”, “deep learning” and how machines make decisions.
Such reports may give companies the impression that if they want to apply AI, they must first hire machine learning experts to build a perfect algorithm. But if companies do not consider how to obtain higher quality and larger amounts of customized training data for machine learning models to learn, even with perfect algorithms, they may not get the desired results (“We have great algorithms” and “We The model only has an accuracy of 60%.”
Buying commercial machine learning services from companies such as Microsoft, Amazon, and Google without a training data plan or budget is like buying a car and failing to reach the gas station. You just bought a large piece of very expensive metal. The analogy between cars and gasoline is not appropriate, because if you give a machine learning model more training data, the better the model will become. It’s like every time a car runs out of gasoline, the greater the mileage accumulated. So training data is even more important than gasoline. Therefore, the quality and quantity of training data are at least as important as algorithms.
Myth 5: Machines>People
For the past 30 years, the media has always liked to describe AI as a machine that is stronger than humans, such as Schwarzenegger in “Terminator” and Alicia in “Ex Machina” Vikander. It is understandable for the media to do so, because the media wants to establish a simple narrative structure of who will win between the machine and the human. However, this does not match the actual situation.
For example, the recent news that Google’s DeepMind/AlphaGo defeated Li Shishi was simply described by the media as the victory of machines over humans. This is inaccurate, and the real situation is not so simple. A more accurate description should be “the machine unites many people to defeat one person”.
The core reason for eliminating this misunderstanding is that machines and humans have complementary capabilities. Please see the picture above. The machine’s specialty is processing structured calculations, and they will perform well on the task of “finding feature vectors”. Humans’ specialty is to understand meaning and context. They perform well on the task of “finding leopard print dresses”, and it is not so easy for humans to do the task of “finding feature vectors”.
Therefore, the correct framework for enterprises is to realize the complementarity of machines and humans, and AI is the joint work of machines and humans.
Myth 6: AI is the replacement of humans by machines
The mainstream media like to portray the future of dystopia because they think it can attract attention. This may indeed attract readers’ attention, but it does not help to truly understand how machines and humans work together.
For example, let’s go back to the business of enterprise classification to support ticket. In most companies today, this is still a 100% manual process. Therefore, this process is slow and costly, and the number of things that can be done is limited. Suppose you have a model with 70% accuracy after categorizing 10,000 support tickets. The result is wrong 30% of the time, but then Human-in-the-loop can intervene. You can set the acceptable confidence level to 95% and only accept output results with a confidence level of 95% or higher. Then the machine learning model can only do a small part of the work initially, such as 5%-10%. But when the model gets new artificially labeled data, it can learn and improve. Therefore, as time goes by, the model can handle more customer support ticket classification work, and enterprises can also greatly increase the number of classified tickets.
Therefore, the combination of machines and people can increase the workload while maintaining quality and reducing the unit economic benefits of important businesses. This eliminates the AI myth that machines replace humans. The truth is that AI is a machine that strengthens humans.
Myth 7: AI=ML
The last myth about AI in mainstream media is that artificial intelligence and machine learning are treated as the same thing. This may make corporate management think that as long as they buy a commercial machine learning service from Microsoft, Amazon, or Google, they can turn AI into products.
To implement an AI solution, in addition to machine learning, you also need training data, which requires human-in-the-loop. Machine learning without training data is like a car without gasoline. Although it is expensive, it can’t go anywhere. The lack of human-in-the-loop machine learning can also lead to undesirable consequences. You need people to overturn the low-confidence predictions of machine learning models.
Artificial intelligence in the future
There will definitely be a big change in transportation, from the current manual driving to the future unmanned driving. Now in Silicon Valley, the United States can often see those unmanned vehicles put into use, not only unmanned cars, aircraft can also use unmanned technology to soar in the sky, are you hungry, commercial use of small drones Food delivery has already begun. So there will be a big change in traffic. There will certainly be a huge gap in medical treatment. Artificial intelligence will automate diagnosis by automatically browsing the user’s condition. At the same time, wearable medical devices and mobile applications can enable us to go further in the future of artificial intelligence medical treatment. It can also be further improved in terms of wheelchairs and intelligent bones. In terms of security, artificial intelligence must also be an indispensable part of the future. In the future, artificial intelligence will become a very important part of public security, whether it is from the face recognition technology on surveillance or the robot police judge in the future, it will have an important position. At present, face recognition technology has been used in most of cameras, which is of great help to the police in finding suspects. It is believed that artificial intelligence will be of greater help to the police in the future.