"Popular Science: What Level Does AI Need to Reach to Fully Replace Human Work?
The author of this article, Mr. Chen Yunwen, is the CEO of Daguan Data. He has won championships in international data mining competitions such as ACM, and has held positions like the former director of the Tencent Literature Data Center, senior director, chief data officer of Shanda Literature, and core technical engineer at Baidu. He holds a Ph.D. in Computer Science from Fudan University.
For different application scenarios, the practical level of artificial intelligence technology in the industry varies. According to internationally renowned AI expert Professor Sandeep Rajani's classification method described in his book "Artificial Intelligence: Man or Machine," AI capabilities are horizontally compared with human abilities and divided into four distinct levels:
Peak Level - Achieved an unsurpassable optimal capability.
Beyond Human Level - Stronger than all humans.
Strong Human Level - Stronger than most humans.
Weak Human Level - Weaker than most humans.
In his textbook "Intelligent Web Algorithms" (second edition), which introduces computer algorithms and data mining techniques, Mr. Chen Yunwen translated and presented Professor Rajani’s categorization. Given the current era’s context, here are the levels of AI technology in various existing application areas:
Peak Level:
In certain application scenarios where the rules are comprehensive and the strategy space is small, such as playing Go on a 19x19 board, today’s computer technology can exhaust all possible game conditions to ensure optimal solutions. Similarly, in games like Tic-Tac-Toe and Checkers, as well as statistical data applications, computers have already reached peak performance.
Beyond Human Level:
AlphaGo’s victory over Ke Jie in Go and IBM’s Deep Blue defeating chess champion Garry Kasparov show that AI has surpassed human capabilities in these complex board games. In specific application areas like fingerprint recognition and iris recognition, the technology has become highly mature, achieving a level beyond human capabilities.
Strong Human Level:
Some intellectual activities require deep domain expertise. Compared to a small number of professionals, computers still lag behind, but they excel over the general population. For instance, in areas like Texas Hold'em and Bridge, computers have outperformed most ordinary humans in specialized fields such as facial recognition under ideal conditions (no obstructions, lighting, or angle issues) and speech recognition without regional accents or noisy environments. In identifying plant species, computer capabilities have also reached a strong human level.
Weak Human Level:
Many skills are easily grasped by most humans, like driving a car. However, for a computer system, the complexity of the signals to be collected and the data analyzed makes reaching the level of an average human challenging, placing it at a weak human level. Common examples include writing articles, reading comprehension, and human language translation.
The reasons for the varying levels of AI development across different fields are primarily influenced by three factors:
Factor 1: Rules and Clarity of Evaluation Methods
The simpler and clearer the rules, and the more quantifiable the problem evaluation, the lower the cost to implement AI, such as in chess and card games, where computers can perform exceptionally well. Uncertainty in problems presents significant challenges for AI learning. For example, driving a vehicle involves numerous variables and lacks strict win/loss or good/bad criteria. Similarly, games like Mahjong or Poker involve randomness and luck, with inconsistent conditions, posing additional challenges for AI.
Thus, the clearer the rules and the more objective the criteria for assessing quality, the better the practical effectiveness of AI technology. In areas with vague rules, human thinking and solutions commonly used in everyday life tend to yield better results.
Factor 2: Handling Special Cases
Many application problems are addressed under typical scenarios but include unique exceptions, leading to varying degrees of difficulty.
Take face recognition as an example. Existing machine learning technologies achieve high accuracy under ideal conditions like proper lighting, correct angles, and no obstructions. However, in practical applications, it encounters numerous special cases, such as interference from light, angle, makeup, jewelry, partial obstructions, aging, and deliberate attempts to deceive the system. These factors significantly impact the practical effectiveness.
Autonomous driving technology also faces numerous challenges, from normal weather conditions to adverse weather like rain and snow. Different road conditions present substantial challenges to its practical implementation.
Many existing AI applications perform excellently under laboratory conditions. However, in industrial applications, the conditions are more complex and harsher than in labs, requiring the handling of various anomalies and interference factors. Consequently, the practical level of many applications remains between the strong human level and weak human level.
Currently, transitioning AI technology from lab settings to practical applications presents numerous obstacles, and significant progress is yet to be made. Pragmatically, the first step is to restrict specific scenarios, minimizing uncertainty and simplifying problems. For example, in developing autonomous driving technology, limiting applications to fixed routes or closed circuits greatly simplifies technical difficulties, often elevating the practical level from weak to human-level performance. In text comprehension, limiting industries, types, and text understanding can significantly improve system accuracy to a practical level.
Of course, from the perspective of 'uncertainty,' machines have advantages. Humans’ energy is limited, and prolonged focus leads to decreased judgment accuracy and inconsistent work quality. Machines, however, can operate stably under long-term, high-pressure environments without fatigue or emotional influence. In such scenarios, human work appears 'uncertain' while machine output remains 'deterministic.'
Factor 3: Data Accumulation
We all know that 'Big Data + Algorithm Model = Artificial Intelligence.' It’s evident that the key to AI lies in big data. As the saying goes, even a clever woman needs firewood to cook. Only by accumulating vast amounts of training data can the level of AI be enhanced. AlphaGo defeated the best human players by accumulating tens of millions of Go game datasets and conducting comprehensive model training.
In the real world, the accumulation of training data has only just begun. Especially the accumulation of labeled data required for supervised learning often demands substantial manual effort, which incurs high costs and severely limits AI advancements in related fields. Additionally, some data areas are restricted by policy factors, such as medical data, or monopolized by certain industries. These factors hinder data circulation and slow AI progress.
With the rapid decline in computer hardware storage costs in recent years, cloud computing becoming increasingly popular, and the hardware environment for data accumulation improving, data acquisition awareness is gradually awakening. It is hoped that with application demand promotion, more data will be digitized and recorded, and excellent algorithm models will be trained to enhance effects.
Starting from the weak human level, progressing to the strong human level, beyond the human level, and the peak level, there is still a long way to go. The pace of scientific development is usually easier first and then more difficult and complex. As technology accumulates and progresses, it is believed that in the future, more and more applications will enable AI technology to replace humans in completing increasingly valuable work.
Chen Yunwen, founder and CEO of Daguan Data, holds a Ph.D. in Computer Science from Fudan University, is the vice president of the Multimedia Society of the Shanghai Computer Society, and a senior member of the International Computer Society (ACM) and the Institute of Electrical and Electronics Engineers (IEEE). He once served as the chief data officer of Shanda Literature, senior director of Tencent Literature, head of the data center, and Baidu’s core technology R&D engineer. He represented China multiple times in international data mining competitions such as ACM and won championships."
The author of this article, Mr. Chen Yunwen, is the CEO of Daguan Data. He has won championships in international data mining competitions such as ACM, and has held positions like the former director of the Tencent Literature Data Center, senior director, chief data officer of Shanda Literature, and core technical engineer at Baidu. He holds a Ph.D. in Computer Science from Fudan University.
For different application scenarios, the practical level of artificial intelligence technology in the industry varies. According to internationally renowned AI expert Professor Sandeep Rajani's classification method described in his book "Artificial Intelligence: Man or Machine," AI capabilities are horizontally compared with human abilities and divided into four distinct levels:
Peak Level - Achieved an unsurpassable optimal capability.
Beyond Human Level - Stronger than all humans.
Strong Human Level - Stronger than most humans.
Weak Human Level - Weaker than most humans.
In his textbook "Intelligent Web Algorithms" (second edition), which introduces computer algorithms and data mining techniques, Mr. Chen Yunwen translated and presented Professor Rajani’s categorization. Given the current era’s context, here are the levels of AI technology in various existing application areas:
Peak Level:
In certain application scenarios where the rules are comprehensive and the strategy space is small, such as playing Go on a 19x19 board, today’s computer technology can exhaust all possible game conditions to ensure optimal solutions. Similarly, in games like Tic-Tac-Toe and Checkers, as well as statistical data applications, computers have already reached peak performance.
Beyond Human Level:
AlphaGo’s victory over Ke Jie in Go and IBM’s Deep Blue defeating chess champion Garry Kasparov show that AI has surpassed human capabilities in these complex board games. In specific application areas like fingerprint recognition and iris recognition, the technology has become highly mature, achieving a level beyond human capabilities.
Strong Human Level:
Some intellectual activities require deep domain expertise. Compared to a small number of professionals, computers still lag behind, but they excel over the general population. For instance, in areas like Texas Hold'em and Bridge, computers have outperformed most ordinary humans in specialized fields such as facial recognition under ideal conditions (no obstructions, lighting, or angle issues) and speech recognition without regional accents or noisy environments. In identifying plant species, computer capabilities have also reached a strong human level.
Weak Human Level:
Many skills are easily grasped by most humans, like driving a car. However, for a computer system, the complexity of the signals to be collected and the data analyzed makes reaching the level of an average human challenging, placing it at a weak human level. Common examples include writing articles, reading comprehension, and human language translation.
The reasons for the varying levels of AI development across different fields are primarily influenced by three factors:
Factor 1: Rules and Clarity of Evaluation Methods
The simpler and clearer the rules, and the more quantifiable the problem evaluation, the lower the cost to implement AI, such as in chess and card games, where computers can perform exceptionally well. Uncertainty in problems presents significant challenges for AI learning. For example, driving a vehicle involves numerous variables and lacks strict win/loss or good/bad criteria. Similarly, games like Mahjong or Poker involve randomness and luck, with inconsistent conditions, posing additional challenges for AI.
Thus, the clearer the rules and the more objective the criteria for assessing quality, the better the practical effectiveness of AI technology. In areas with vague rules, human thinking and solutions commonly used in everyday life tend to yield better results.
Factor 2: Handling Special Cases
Many application problems are addressed under typical scenarios but include unique exceptions, leading to varying degrees of difficulty.
Take face recognition as an example. Existing machine learning technologies achieve high accuracy under ideal conditions like proper lighting, correct angles, and no obstructions. However, in practical applications, it encounters numerous special cases, such as interference from light, angle, makeup, jewelry, partial obstructions, aging, and deliberate attempts to deceive the system. These factors significantly impact the practical effectiveness.
Autonomous driving technology also faces numerous challenges, from normal weather conditions to adverse weather like rain and snow. Different road conditions present substantial challenges to its practical implementation.
Many existing AI applications perform excellently under laboratory conditions. However, in industrial applications, the conditions are more complex and harsher than in labs, requiring the handling of various anomalies and interference factors. Consequently, the practical level of many applications remains between the strong human level and weak human level.
Currently, transitioning AI technology from lab settings to practical applications presents numerous obstacles, and significant progress is yet to be made. Pragmatically, the first step is to restrict specific scenarios, minimizing uncertainty and simplifying problems. For example, in developing autonomous driving technology, limiting applications to fixed routes or closed circuits greatly simplifies technical difficulties, often elevating the practical level from weak to human-level performance. In text comprehension, limiting industries, types, and text understanding can significantly improve system accuracy to a practical level.
Of course, from the perspective of 'uncertainty,' machines have advantages. Humans’ energy is limited, and prolonged focus leads to decreased judgment accuracy and inconsistent work quality. Machines, however, can operate stably under long-term, high-pressure environments without fatigue or emotional influence. In such scenarios, human work appears 'uncertain' while machine output remains 'deterministic.'
Factor 3: Data Accumulation
We all know that 'Big Data + Algorithm Model = Artificial Intelligence.' It’s evident that the key to AI lies in big data. As the saying goes, even a clever woman needs firewood to cook. Only by accumulating vast amounts of training data can the level of AI be enhanced. AlphaGo defeated the best human players by accumulating tens of millions of Go game datasets and conducting comprehensive model training.
In the real world, the accumulation of training data has only just begun. Especially the accumulation of labeled data required for supervised learning often demands substantial manual effort, which incurs high costs and severely limits AI advancements in related fields. Additionally, some data areas are restricted by policy factors, such as medical data, or monopolized by certain industries. These factors hinder data circulation and slow AI progress.
With the rapid decline in computer hardware storage costs in recent years, cloud computing becoming increasingly popular, and the hardware environment for data accumulation improving, data acquisition awareness is gradually awakening. It is hoped that with application demand promotion, more data will be digitized and recorded, and excellent algorithm models will be trained to enhance effects.
Starting from the weak human level, progressing to the strong human level, beyond the human level, and the peak level, there is still a long way to go. The pace of scientific development is usually easier first and then more difficult and complex. As technology accumulates and progresses, it is believed that in the future, more and more applications will enable AI technology to replace humans in completing increasingly valuable work.
Chen Yunwen, founder and CEO of Daguan Data, holds a Ph.D. in Computer Science from Fudan University, is the vice president of the Multimedia Society of the Shanghai Computer Society, and a senior member of the International Computer Society (ACM) and the Institute of Electrical and Electronics Engineers (IEEE). He once served as the chief data officer of Shanda Literature, senior director of Tencent Literature, head of the data center, and Baidu’s core technology R&D engineer. He represented China multiple times in international data mining competitions such as ACM and won championships."
Wall Mounted Battery,Wall-Mounted Lithium-Ion Battery,Residential Wall-Mount Battery,Wall Mounted Lithium Battery
JIANGMEN RONDA LITHIUM BATTERY CO., LTD. , https://www.ronda-battery.com