Article reprint source: AI DreamWorks
Original source: Thinking about things
Image source: Generated by Unbounded AI
What are the main problems with AI so far?
It is obvious that the business loop is not closed.
The Internet made companies like Google successful, and since then the products of these companies have driven the Internet to develop in depth, such as cloud computing and Android, and ultimately the influence of the Internet has become increasingly greater.
AI is not good, it always flows in waves.
After a wave of attention from researchers, VCs, entrepreneurs, and the media, it became silent. This has been the case in the past few times.
This time we finally see the dawn of the second wave. From this perspective, Sam Altman is right. We should think about products based on the premise that AGI will come.
How can we be more specific? How can we think about future products based on the fact that AGI will come?
In this thinking process, I personally think two points are the most critical:
The first is the Turing Test 2.0 mentioned earlier.
The second is native intelligence.
Specific judgments on these two aspects are likely to become more and more critical as intelligence improves.
What is Smart Native?
Intelligence native can be seen as a new computing paradigm. This computing paradigm is based on real-time perception and full digitalization, centered on AGI, integrates various algorithms to make intelligent decisions, and adheres to the principle of intelligence priority throughout the process.
This may be a bit conceptual, so let's take a few examples:
Apple's VisionPro has just been shipped. What will this product look like in the future?
It will be like the headgear of Star-Lord in The Avengers.
This advanced helmet automatically scans the surrounding environment, performs a large amount of analysis and judgment without human intervention, and then leaves the parts that require human intervention to Star-Lord.
If we go back from this kind of science fiction product, we will get to VisionPro, smart speakers, and autonomous driving.
Even products like VisionPro and smart speakers that seem to be very primitive, their computing methods are very different from the GUI trend set off by Steve Jobs and Bill Gates. Icons, menus, and buttons are actually functional decompositions of the operation process, while smart interaction is centered on the purpose and minimizes the cost of achieving the purpose.
Once you put them together with mobile phones and computers, the difference becomes more obvious.
This is the basic model of intelligent native: real-time feedback, everything is counted, central decision-making, and intelligence first (intelligence first is a little difficult to understand, and will be explained later).
Smart native is not limited to terminal products. With the above ideas, we can further trace back the clues of smart native in ancient products.
What is the most relevant product in our time?
It is a search engine. Although it is not very intelligent (so it cannot be called intelligent native), its entire calculation process is highly consistent with the above characteristics.
For search engines, the production of content is completely handled by crawlers and is almost free. Due to the huge amount of content, the entire crawling process must be carried out under the control of the program. It is unlikely that all employees will be mobilized to manually crawl more content to supplement the database considering the current urgent task. This is meaningless in the face of massive data, and once the basic crawling rules are violated, the data will be harmful.
After acquiring a large amount of content, the matching content is returned based on countless user inputs (certain keywords). Because the volume is too large and the real-time requirements are extremely high, it can only be decided and driven by algorithms. In the entire business process, human intervention can only be to adjust preferences and strategies through parameters. The entire end-user use process does not require Google's manual intervention.
Here the search engine itself can be seen as a self-operating system driven by a primary intelligent agent.
Although it is not so real-time when crawling online data, it is basically close to real-time feedback and central decision-making. Algorithms such as PageRank are not very intelligent, but they do not require human intervention to give feedback based on user input and perform the function of central decision-making.
The decision-making body has changed in this process, raising Silicon-based decision-making power to at least the same level as the users themselves.
What is Smart First?
This can be interpreted in the most vulgar way: when you report to your boss, your boss takes priority. When intelligence has the power to make decisions, the entire operating system must adapt to it, otherwise it will not work, so it takes priority.
But this cheesy metaphor is crucial and affects the supporting behavior of using it well. In extreme cases, when it cannot distinguish the authenticity of input, it may even need to configure a special input quality review system to let it see the real input.
For example, when changing data in the future, the first thing to consider is not whether the boss likes it or not, but whether it affects the performance of intelligence.
Smart native application space
If general intelligence really arrives, then it is clear that the native model of intelligence is not limited to search, recommendation, smart speakers, and VisionPro. Instead, it will cover areas that were covered by applications in the past, or even areas that were not covered before.
This time we take the enterprise field as an example.
SaaS is actually the cloudification of software. Now it seems that this is just like the first wave of AI, and it is basically hopeless.
But what happens when SaaS is combined with intelligent native? Will it work?
At this time, SaaS is no longer an independent function, but an operating model of the enterprise, and it may create a new intelligent native enterprise. All SaaS is integrated here.
According to the same framework, referring to autonomous driving, smart native enterprises can also be classified as follows:
This may give people a feeling that ERP has become sophisticated, with a considerable part of the management's decision-making functions being separated out.
When intelligence is sufficient and it generally doubles productivity, all companies will be intelligence-native companies, otherwise you will not be able to survive.
Is this an ancient prototype?
In fact, there are some, a typical example being the management model of Meituan’s delivery guys and Didi’s drivers.
It’s still not smart enough, so it’s not truly natively intelligent, but we can see the progress of intelligence from food delivery to TikTok.
Similar situations can obviously be extended to all fields that have spent a lot of effort on digitalization in the past decade or so:
If enterprises can do it, then why can't chemical plants and medical institutions do it?
This progressive process will gradually unfold based on the Turing Test 2.0.
Only one premise is required: general intelligence is really general enough.
Smart native form
There are many forms of applications that are related to intelligence, but not all of them are native smart applications.
If we put it on a timeline, it is more likely to be a progressive development process like this:
Phase 1: The emergence and mixing of intelligent elements. For example, toilets can suddenly support voice, and payment requires face recognition. At this time, intelligence is in a supporting position and is a new functional element. The role and compression algorithm library are actually not much different. It expands the scope of application capabilities, but it is not native intelligence.
The second stage: the central decision stage. Applications are matched with their own centers, building a perception-decision architecture, but many of these decisions are fixed rules with very narrow adaptation boundaries. For example, a convenience store cannot open a pharmacy, and the weather must be remade after the music is done on the speaker.
The third stage: the native intelligence stage. At this time, intelligence fully covers central decision-making and perception, forming an independently operating octopus-like structure with general intelligence as the core.
According to this stage division, we are actually in the transition period between the second and third stages, as the first stage has already occurred.
Potential areas and order of implementation of smart native
If we look at it from the perspective of B-end and C-end, C-end will be earlier than B-end, and B-end will be earlier than the industry. If we look at it from the perspective of bits or atoms, bits will be earlier than atoms. (See: AI is not short of concepts, or even technology, but it is really short of products)
What kind of results and impacts will be brought about after the application is significantly refactored?
The most superficial impact is the reconstruction of the application landscape.
If every application exists in a native smart form, then many apps that exist by function, such as search, browser, music, etc., will be folded. However, there are too many commercial interests involved behind this, so there will be a lasting friction between technological trends and the existing interest structure.
The gradual impact is the rebalancing of the roles of silicon-based people and carbon-based people. The more developed the native intelligence is, the more the boundary between silicon-based people and carbon-based people will move toward silicon-based people. This will lead to multiple models, which I mentioned in "The Most Certain Thing in the Next Decade", and I will repeat it here:
The first is the food delivery boy model, which is characterized by full transparency of activities and is defined by silicon-based intelligence.
The second model is the OpenAI model, which is characterized by strong individuals being linked by fields and having a high degree of freedom of movement.
The third model is very similar to the ultimate wisdom of the Cree people. It cuts away some of the rights and responsibilities vertically, but maintains individual space and independence.
Regardless of the situation, the more silicon-based products are involved, the more output value they create will increase under the existing measurement system. This implies a new order of wealth distribution.
The ultimate impact is the reconstruction of production relations. It seems difficult for us to imagine a world where very few people are responsible for production and the vast majority of people are responsible for consumption. This seems to have disastrous consequences under the existing measurement model, so what kind of new form is positive?
It is worth mentioning that in the early days of the bit world, smart native applications are likely to follow the exponential growth curve of Internet applications. And the cost of implementation has been greatly reduced due to general intelligence itself, which is a rare opportunity for the general public.
summary
The big model itself doesn’t have any real business opportunities, but the intelligent native application is probably a blue ocean for a while, and it’s worth spending enough time thinking and judging. In addition, Agent is basically an intelligent native application, but the intelligent native application may not be the Agent that we always talk about now.
