The wave of AI is sweeping across the globe, but do people really understand AI? Alaric, the author of the Vancisco website, points out three common misconceptions people have about AI today.
1. Equating AI with Generative AI or Large Models
The first misconception is equating AI with generative AI or large models. Alaric states that this understanding is narrow and incomplete, as the development of AI has been ongoing for over 70 years. “Before the rise of generative AI, discriminative AI (Discriminative AI) was already widely applied. For example, image recognition and facial recognition technologies, which we are all familiar with, are typical applications based on machine learning for judgment and classification. IBM launched its Watson X product over 12 years ago, and this discriminative AI solution has been serving more than 40,000 customers globally.”
Alaric emphasizes that the explosive development of generative AI is primarily based on breakthroughs in probabilistic models and deep learning. However, we must recognize that generative AI is just one phase in the development of AI, not the entirety. The true AI industry ecosystem is far richer than the large models that are currently being widely discussed in the media.
2. The Misconception of AI’s Boundless Abilities
The second misconception is about the limits of AI’s capabilities. Alaric points out that many people believe AI is all-powerful. In reality, although AI has enormous potential, it is still far from being truly “omnipotent.” “Especially in enterprise applications, the implementation of AI is progressing slowly. A typical phenomenon is that many companies enthusiastically purchase AI devices, such as the DeepSeek integrated machine, but after deployment, they are unsure how to use them.”
Moreover, the problem of AI hallucinations cannot be overlooked. Alaric explains that when early AI was asked, “What fish is in fish-flavored shredded pork?” it would give plausible but incorrect answers based on probability. Such problems reveal the inherent limitations of current AI technology—it excels at pattern recognition and probabilistic reasoning, rather than genuine cognitive understanding.
3. The ROI (Return on Investment) Misconception in AI Implementation
The third misconception is that companies commonly face the issue of ROI not meeting expectations when implementing AI. This is also rooted in a key misunderstanding.
Alaric points out that AI is not a technology that appears out of nowhere. For enterprises, AI applications must be built on a solid digital foundation. “We find that many companies mistakenly allocate investments in digitalization construction to AI costs, which is the main reason for distorted ROI calculations. For instance, when companies start deploying AI, they often face issues like data loss and poor data quality. What they actually need to invest in at this stage is digitalization (data governance, process restructuring, etc.), not AI itself. Although this foundational work may seem costly in the short term, it is an essential long-term investment.”
“The relationship between digitalization and AI can be understood this way: digitalization is the foundation, and AI is the superstructure. Based on our observations, especially among private enterprises, their digital foundation is generally weak. According to the ‘Digital Maturity Model’ issued by the government, on a scale of 5, most companies can only achieve a score of 2-3. Practice shows that if a company’s digital maturity is below 3-4, the effect of AI applications will be significantly compromised,” Alaric adds.
4. AI Will Change the World in Two Directions
In Alaric’s view, AI will undoubtedly change the world, and this change will unfold along two paths:
1.Native Innovation Areas: These include fields like robotics, AI-generated content (AIGC), world models, and other new directions. These areas have no historical baggage, allowing for boundless imagination, and their development speed will be rapid, similar to how the early internet gave rise to search engines, social media, and other native applications.
2.Traditional Industry Transformation: This process is often slower than expected. For example, it took the internet 20 years to transform traditional industries, and even then, the results did not meet the initial optimistic predictions. Technological implementation faces real-world resistance, including existing systems, regulations, and habits, making the transformation more gradual.