Pain Points and Prospects of AI's Comprehensive Commercial Applications
-
The AI boom has been around for a long time, but its current commercial applications remain relatively rudimentary. The deeper exploration and commercial utilization of AI technology have yet to reach their core potential. This article delves into the phenomenon, analyzing further commercial possibilities for AI.
As a professional working on AI-enabled transformation in traditional industries, I’ve seen how AI algorithms can liberate the productivity of traditional production line operators. However, I still believe the journey from AI exploration to widespread adoption and application is a long one.
Just like the invention of electricity 200 years ago, the emergence and exploration of technology itself hold far less significance for an era than its actual commercial application. Humanity has only truly understood electricity for about 400 years, with the first 200 years being largely experimental. By the time AI truly empowers and delivers efficiency, it may no longer resemble the 'AI' we recognize today.
Artificial Intelligence (AI) is a new technological science that studies and develops theories, methods, technologies, and application systems to simulate, extend, and expand human intelligence.
In fact, as early as 1956, 17 years after Americans invented computers, a few visionary scientists were already researching how to simulate human intelligence with machines.
When IBM's Deep Blue defeated the world chess champion in 1997, it didn’t cause much of a stir. The real turning point that shifted human perception and even transformed the global AI industry structure was in 2016, when AlphaGo defeated Go grandmaster Lee Sedol. From then on, AI became a household name—AlphaGo not only calculated m to the power of n possibilities in its next move but also mastered the overall control of the game (including deception techniques), which was what truly amazed humanity.
So, what exactly is AI? AI is when machines learn from a subset of sample data, perform calculations using algorithms that simulate human thinking, and finally validate the results with another subset of real-world sample data to produce the final output model.
The applications of AI mentioned above span various dimensions such as numbers, images, speech, and text. At its core, AI is about the process of learning from, imitating, and outputting data.
AI Processing Data Across Different Dimensions
The mobile internet era has captured every aspect of daily life—food, clothing, housing, and transportation—in data. Informatization and digitization have left data traces of all business operations. These data manifest in diverse forms: numbers, images, speech, and text. Datafication is the foundation of AI development because the essence of AI computation is feature-based learning and feature-based engineering.
Feature learning, also known as representation learning, refers to the process by which machines automatically extract features or representations from data. For example, from 6,000 temperature data points, the machine might identify 23–39°C as the high-temperature range and 0–8°C as the low-temperature range.
Feature engineering, on the other hand, involves human intervention in processing and extracting data, sometimes referred to as 'data cleaning.' Unlike feature learning, this is a manual process. It involves human efforts to shape data into a format deemed suitable for subsequent models. For instance, we might consider temperatures from June to September as valid summer data while discarding anomalies like a 20°C reading in December as meaningless. Such manually processed data carries the significance of feature engineering.
One reason AI hasn’t advanced rapidly over the decades is the lack of sufficient computing power. Chips and processing speeds couldn’t keep up. However, with decades of hardware development following Moore’s Law, computing power has now become a prerequisite for AI’s widespread adoption. Meanwhile, humans have begun reverse-engineering how our own brains function.
Algorithms Are the Soul of AI
The human brain doesn’t store vast rules or statistical data. Instead, it operates through neuron activation. Each neuron receives inputs from others and produces outputs to stimulate other neurons. Countless neurons interact, ultimately generating various outcomes.
For example, when people see a beautiful woman and their pupils dilate, it’s not because the brain applies rule-based judgments on body proportions or tallies up all the beautiful women one has ever seen. Instead, neurons trigger from the retina to the brain and back to the pupils. In this process, it’s nearly impossible to pinpoint the exact role each neuron plays in the final outcome—they just do.
Humans have tried inputting all factors influencing judgments of 'beauty'—height, weight, facial features, skin, smile, body proportions, attire—as input conditions. For any individual, if exposed to an infinite sample of 'women' and asked to rate their beauty (across different dimensions), these rated samples and scores can be fed into a machine. The machine can then automatically assign beauty scores to images and even determine the weight of each input factor (height, weight, facial features, skin, smile, body proportions) in the scoring process.
This is how machines simulate the predictive function of the human brain.
With an understanding of AI’s working principles, let’s explore its commercial applications.
The following diagram, sourced from Baidu AI Research’s official website, illustrates their definition of AI applications.
Currently, AI applications are still in the weak AI stage, characterized by narrow and shallow business scenarios. They can only address repetitive 'point' problems in specific contexts. For broader business scenarios or workflows, integration with other information systems is required.
Thus, while AI is a hot topic, its practical implementation remains a challenge. Just as AlphaGo couldn’t perform other tasks after defeating a Go grandmaster—or even 'forgot' the game strategies it learned—AI in commercial applications faces several pain points:
Businesses adopt AI to create value, but deploying AI requires significant human resources. Moreover, because algorithms are black boxes, they face skepticism from operators and cannot fully replace human labor for a long time.
AI’s technological advancement alone isn’t enough to justify widespread adoption. Factors like product stability, accuracy, and error rates are more critical for large-scale deployment. Currently, AI tends to perform either exceptionally well or poorly. Even a 1% chance of extreme failure can have devastating effects—for example, refunding 30 million yuan to a customer who only spent 300 yuan.
For businesses, rectifying errors caused by AI is far more challenging than replacing a product. If adapting to AI’s precision requires cross-departmental system or data changes, companies would rather opt for a low-risk, stable product that meets mass-user needs.
Every company’s workflow is unique. Applying AI within an organization inevitably takes the form of 'AI + information systems,' with AI’s role shaped by the business context. For instance, in my work in the lingerie industry, attempting to migrate key features like age and tag prices to the fast-moving consumer goods (FMCG) sector for alcohol reveals that business scenarios are rarely reusable, often necessitating model rebuilding.
Despite these pain points, AI remains a transformative technology. Andrew Ng, one of the most influential figures in AI, once said: 'AI is the new electricity.' Once AI empowers the commercial world, it will redefine industries just as the internet did—only faster and with shorter iteration cycles.
In 2016, Robin Li pointed out that the internet had undergone massive changes over the past 16 years, broadly evolving through three acts:
Each era has its unique characteristics, and thus, they evolve at different paces. The PC internet era relied heavily on rapid software responsiveness, while the mobile internet era built its own ecosystem around user attention. In the third act—the AI era—machines will replace certain human "organs," such as the brain, eyes, hands, and ears, enabling intelligent observation, reading, writing, calculation, and speech. Humans and machines will collaborate to create a new society that blends the virtual and physical worlds.
This integration of intelligence and virtual-physical fusion will manifest across various fields, industries, enterprises, and business scenarios.
Humans and machines are destined to work together to write a new chapter for the future of society.