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

Through the research and analysis of more than 7000 "AI startups", I find that most people underestimate the difficulties and challenges facing the development of AI. Are you involved in this? "

You may have heard Andrew Ng say something similar: "AI is like electricity in our new era! the development of electricity has changed countless industries; the same will be true of AI today."

I basically agree with this view. The problem, however, is that this statement ignores the huge obstacles that AI faces in its rapid adoption. After all, AI won't be everywhere overnight. As far as electric power is concerned, it took more than 40 years for it to become a ubiquitous technology. By 1882, the world had discovered the key elements of modern electricity. However, many challenges such as expensive infrastructure, lack of talent and opaque regulation have hindered the immediate and large-scale adoption of electricity. Taken together, the average American household did not have access to electricity until 1925!

AI is like a new kind of electricity, it will change all kinds of industries. But like the development of electricity, it will take decades. For the AI world, we are in 1882, not 1925.

What friction is hindering the application of AI? where will AI succeed in the first place? What will it lag behind? Unless we start this dialogue, many technically feasible and reasonable AI enterprises will face failure. This dialogue is crucial because the world is betting heavily on AI, though it may not be wise. I searched the Internet and found 7192 "AI startups" claiming to be an AI company or using machine learning. These startups have raised more than $19 billion and employed more than 150000 people.

When will your AI business succeed?-- Framework

Fortunately, you can predict that your AI business is more likely to succeed in the near, medium or long term. The capabilities and challenges of AI are easy to understand-all you have to do is take a comprehensive look at them and then think critically about your AI use cases. To do this, consider using a simple framework: the speed with which AI solutions are adopted is a function of value potential and unique friction. Many frictions slow down the application of AI technology, but these frictions can more slow down the development of some enterprises. The reason is that some AI solutions can create more value than others. When an AI solution has great value potential, it is easier for companies, investors, regulators and consumers to unite to overcome friction. This simple relationship between value and friction creates a useful framework:

AI adoption rate = f (AI friction, AI value)

So what will be the path to large-scale adoption of AI for your AI bet? For any problem, enterprise or industry, the framework can be operated directly. This is a more detailed classification.

Top friction hindering the rapid application of AI

The first step is a thoughtful analysis of AI frictions, which may slow down the adoption of your AI business. Human, data, and market frictions all slow the adoption of proven AI solutions. They complicate development, limit scalability, and introduce use case kill risks. Not all frictions can be generalized. Some are much more dangerous than others:

AI friction predictor

Human restrictions on AI man-machine loop requirements: many algorithms need human supervision. For example, Facebook employs more than 15000 people to assist with their content review algorithms. Manual data annotation requirements: many use cases of AI require humans to teach algorithmic prediction (or, in technical terms, "annotate" data). For example, Baidu has to hire thousands of translators to train its Chinese translation algorithm. Lack of talent: there is a global shortage of data scientists, machine learning engineers and other AI talents. This makes it challenging for enterprises to build capable AI teams. In 2018, Indeed.com posted three times more information about AI-related jobs than searches. AI's data limits organic data creation: some business models cannot naturally generate the data needed for AI. For example, traditional retailers do not have access to rich data on customer shopping patterns. To integrate artificial intelligence, retailers need to adopt new business models, such as online sales and "direct to consumers" models. Lack of data infrastructure: AI requires significant investment at each level of the technology stack. On-prem hardware and legacy software solutions are the curse of AI. In order to achieve artificial intelligence, enterprises must invest in cloud platforms, data concentration platforms, data security and artificial intelligence development tools. Existing data is cluttered: data is rarely organized in a table of clean, centralized rows and columns. Instead, most of the data exists in messy documents or legacy software systems. Companies tend to store data across teams and organizations, often unable to maintain documents with different data, and do not enforce standards for how data is captured and stored. Dependence on third-party data: artificial intelligence craves data. When your company doesn't have enough proprietary data, you have to buy it, and it's very expensive to license and maintain API to access third-party data. Data speed is low: most artificial intelligence requires thousands of examples of complete feedback loops to learn. This is challenging in areas where the feedback cycle is slow. For example, obtaining data on long-term health care outcomes for chronic diseases is a costly process. AI's market limitations capture the business model changes required for AI value: in order to capture AI value, many industries will have to change the way they deliver products and services. Self-driving cars, for example, will force carmakers to adopt a "transport as a service" strategy. Near-perfect algorithm performance requirements: some artificial intelligence use cases have high failure costs. In the case of medical and self-driving car diagnostic decisions, for example, AI solutions pose significant risks. AI requires process change: products that support AI often introduce very different workflows. For example, artificial intelligence recruitment solutions tend to favor non-traditional interviews and job applications. This scares more traditional HR teams. Unexplained algorithms: in many cases, consumers (and even regulators) need AI tools that can explain themselves. Unfortunately, it is difficult to explain how many artificial intelligence algorithms are making decisions. For example, if a bank refuses to provide credit to customers, they must explain why, which makes it difficult for artificial intelligence to lend. Biased algorithm: AI algorithm often makes biased decisions. This is illegal and offensive in many areas, such as law enforcement, human resources and education. Onerous privacy standards: AI is a threat to privacy. Artificial intelligence creates motivation for enterprises to collect a large amount of private information. In addition, artificial intelligence can infer personal information (such as a person's emotional state) from harmless data, such as typing patterns. These AI solutions that threaten privacy may face regulatory and consumer resistance. Evaluate the value of AI

Once you understand the AI friction faced by the enterprise, it is necessary to carry out value analysis. Can your AI solution reduce costs, save time, mitigate risks, and create new consumer value? If so, how much will it cost? There is no one-size-fits-all way to do this. Once you have evaluated your AI solution, think critically about how this value will motivate stakeholders to overcome friction. In the process, you should consider trends at the macro level. This is dangerous in areas where AI cannot create important value more generally. If you really come to this point, then you will become a lonely artificial intelligence advocate. The McKinsey Global Institute (MGI) recently estimated the potential of artificial intelligence and analysis at more than $9 trillion, and importantly, this value is not distributed proportionally across use cases and industries.

AI use case

After evaluating a list of more than 400 known AI use cases, MGI found that common business problems-supply chain, sales, and marketing-were the most valuable use cases for AI.

Dividing AI value by use case

Cross-industry AI value

By mapping use cases to various industries, MGI assesses the importance of AI to each industry. They found that industries with complex problems in high-end features, such as sales, would benefit most from artificial intelligence.

AI value as a percentage of industry revenue

The Future of AI-- Application Framework

So which industries are most vulnerable to the slower-than-expected speed of artificial intelligence applications? Who is most likely to be the inopportune cannon fodder of AI bets? We can use the framework to find the answer at the macro level. I interviewed several AI experts to estimate the AI friction intensity in each industry, and then compiled this information and plotted a chart corresponding to MGI's AI value estimate:

According to my analysis, AI will sweep across all industries in three waves:

The first wave of AI-fast adopters: this wave of integration of consumer technology and media has begun well. Advanced technologies from companies such as Google, Facebook and Netflix are leading the trend. The second wave of AI-slow adopters: this wave has also begun, but may advance more slowly. Some adopters, such as manufacturers and supply chain operators, are reluctant to adopt artificial intelligence. Some other companies, such as banks, are well aware that if they succeed, they will get huge returns, but they must face major challenges in the process of adopting artificial intelligence. The third wave of AI-hard adopters: healthcare, automotive and (possibly) retail may have lower-than-expected AI adoption rates. They all face huge obstacles to the adoption of artificial intelligence. On the basis of every penny, they are reluctant to use artificial intelligence. It is worth noting that retail is somewhat different from other industries here: traditional retailers face significant frictions in some areas (sales and marketing), but in others (supply chain operations) are rapid adopters of artificial intelligence.

So, when will your artificial intelligence business succeed? You need to analyze the artificial intelligence frictions you face, assess how much value you want to create, and then look at the gap between your business and known AI successes. If the conclusion is more friction and less value, then maybe now is not the time to make that bet. But if you have a high-value, low-friction artificial intelligence solution, don't read this article again. Hit at full speed!

Author: AJ Christensen

Https://www.toutiao.com/i6729308642094350855/

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.