The excitement of AI-related conferences sometimes brings to mind Zhao Benshan’s sketch “Selling Crutches”. Reflecting on the development process of the metaverse, one can see that it is almost identical to the process of “selling crutches”. It starts with combining a sci-fi concept with new technological advancements, packaging it, increasing the output power to preach various self-justifying stories, and then Zuckerberg plays the role of Fan Wei – even the crutches are sold (Facebook changed to Meta), but where is the metaverse in all of this?
AI conferences also have a similar vibe. Lots of imagination, but limited applications. If AI is truly as powerful as it is claimed to be, shouldn’t it be indispensable to our lives like the internet? Where is its super application?
The fatal mindset lies in viewing applications from a technical standpoint and a strategic standpoint, as these perspectives often fail to identify super applications, or even useful applications.
The common misconception here is that AI, no matter how advanced, is not a mirror image of the entire world. Super applications are destined to emerge at the intersection of the world and AI.
In other words, new super applications can only be realized by spanning two different domains, necessitating an outside-in perspective.
The technical perspective, on the contrary, focuses on looking out from AI. This often leads to the observation of various simplified versions of possibilities, focusing on the framework without fleshing out the products. This is accompanied by a purely top-down strategic thinking approach.
This is essentially the model of research institutions, which may have a great start, but very few great products have emerged directly from research institutions.
At this point, let’s not dwell on first principles too much. First principles serve as a starting point, but they can be misleading as they may make one feel like they have grasped the essence of the world, when in fact, they are just the beginning.
The wave of PCs and the internet has just passed. Let’s take a simple retrospective.
The PC and the internet in retrospect
Xerox’s research institute invented the graphical user interface, but it was Jobs and Gates who transformed it into the foundation of the two trillion-dollar businesses we see today. What exactly did these two individuals do?
Once the basic infrastructure of the internet was in place, figures like Jack Ma and Pony Ma in China successfully implemented the practical applications. What exactly did they do?
On the flip side, Facebook serves as a cautionary tale. The company rose to prominence early on due to its social product within school campuses, which lacked significant technological innovation. However, when they tried to expand their horizons, they faltered. Firstly, their choice of apps was suboptimal, leaning towards H5 web pages, and then they resorted to aggressive acquisitions, including the $19 billion purchase of WhatsApp. Most recently, with the metaverse, the decision to rebrand was a bold move, but it seems to have led to chaos and a stumbling block.
So, if you truly want to find the super application of AI at this juncture, it’s best not to pay too much attention to what others are saying. The core lies in one thing: going back to the field to experience new technologies firsthand, and then contemplating and practicing what it can actually do to form a complete experience.
Returning to the field, establishing your own test sets, you will gradually see many new possibilities. What applications are now achievable that were once deemed impossible?
In the past, when I led product teams, we envisioned numerous products, some of which were abandoned due to limitations in technology at the time. But now, some of these products seem feasible.
For instance, there was a product called VIPKID, which focused on using foreign teachers to help children practice English speaking. Many parents bought it to help their children practice English. However, it gradually disappeared from the scene, reportedly due to high operating costs, including traffic and foreign teacher costs.
Could we replace foreign teachers with AI? What level of proficiency can AI achieve in the role of a foreign teacher?
The challenge in the past was that if AI had to be activated using wake words like “Xiaoi”, the entire interaction process would be cumbersome. If wake words were not used, it would be difficult to differentiate between interactive content and environmental noise.
Now, things have improved. By using models to assess the likelihood of interaction in the backend, it is now possible to create a seamless interaction method without the need for wake words. Since this scenario is not as open as Siri and is relatively vertical, the likelihood of establishing an interaction method without wake words has increased.
Similarly, could we create a real-life scenario of “mother storytelling”? Could AIGC generate interesting stories and narrate them to children in a motherly voice?
Some may argue that this idea has already been explored. In such cases, it’s best not to follow Zuckerberg’s example, but to return to the field and refine the details.
Is the interaction truly smooth? Are the stories genuinely engaging? Smart speakers have covered most of these aspects, but what has changed now?
Upon closer examination, one will notice that the starting point of such products involves the replacement of certain roles. Mother, teacher, girlfriend… it’s essentially an AI version of role-playing.
To play a role effectively, one needs new technology, but the key lies in understanding the role and emotions involved.
Therefore, these products inherently conflict with the institute and strategic top-down thinking. Why didn’t Cisco, which clearly understands networks, pioneer the idea of a campus social network? This is the gist of it.
The discovery of super applications
The more roles are incorporated, the more likely new super applications will emerge.
Consider a hypothetical scenario. In the past, we had a product category called UGC, with platforms like Douyin and Zhihu falling under this category. If AI-powered roles are ubiquitous, is UGC still necessary? Why not generate content directly based on preferences?
Of course, social attributes can still be retained, but the product form would change. After all, the majority of users are consumers, not content producers. If AIGC’s capabilities reach a certain level, all
