Machine Intelligence: The "Steam Engine" in the Post-Internet Age

Machine intelligence is not part of the Internet, but technology that is at least as good as the Internet, or even more promising. There is always something called IT and the Internet to be called the third industrial revolution, but this may be a bit inappropriate. The Internet is like a lubricant to accelerate the society, and the optimization is more important in the existing field. If machine intelligence develops, the impact on existing areas will certainly be greater than the Internet. After many years, the changes brought by machine intelligence will at least be as important as the steam engine.

Big data is not machine intelligence

We face too many new technologies every day, such as VR/AR, 3D printing, big data, etc., but the level of energy and the range of impact of each technology are actually different. What is very confusing is big data and machine intelligence. There is an intersection between the two technologies, because the data and computing platforms used by machine intelligence are at least temporarily related to the depth of big data. But machine intelligence is not big data, and this is made with flour, but the same is true for steamed bread.

From a macro perspective, it is particularly easy to see this problem, and the scope of machine intelligence is broader than big data.

IBM's Dark Blue defeated chess world champion Kasparov in 1997. At that time, IBM focused on its supercomputer; Watson actually brought 4T data when it entered the dangerous edge game and won the championship in 2011. Not connected to the Internet. This means that artificial intelligence can generate tremendous value in some relatively closed environments, more like a big upgrade of PCs, but this kind of upgrade usually requires a large amount of data to support.

Of course, the application of machine intelligence is more broad in a networked environment. It has been said that the Palantir created by Peter Til is related to the capture of Osama bin Laden. Although it is difficult to confirm, it is indeed possible. Analysis of a large number of transaction data, find the pattern of abnormalities, narrow the scope, which will undoubtedly improve the efficiency of arrest. This is not just a matter of terrorists. Other crimes can be used as a breakthrough through this model, as long as the corresponding person is behaving abnormally when committing crimes, such as a drug dealer who may use a mobile phone for one day.

The application of machine intelligence is actually without boundaries. It will fully penetrate into various ends and various fields, such as voice assistant, photo album application, Pepper robot, medical, education and so on. It's not a subset of big data, it's like the result of mixed evolution of big data and other fields.

Many people are still not aware of this, even some of the more famous technical conferences, but also the topic of machine learning is listed under big data. But after a few years, it is very likely that the situation will fall, and big data will become a subset of machine intelligence.

Who will be crushed by the wave of machine intelligence?

The power of current machine intelligence is more reflected in 2B, and it is not obvious in 2C. The two types of 2C products that best represent machine intelligence achievements are voice assistants such as GoogleNow, and one is self-driving cars, and even the voice assistant needs to increase the application ratio. If machine intelligence is a technological change in the Internet, then this is definitely the beginning, and the story is still long.

Who is more likely to be crushed in this wave?

The first category is people who are unaware of machine intelligence. There are news reports that many car manufacturers feel the threat of Google's driverless cars and are ready to start driving. Under normal circumstances, this is very difficult to succeed, because the core of Google's driverless car is not the car, but the machine intelligence. How much difference between the two companies in terms of machine intelligence is the gap in this product. And can we think that car companies have a high level of machine intelligence? If not, how can it be successful?

In the same way, I also doubt the prospect of Pepper, the more robotic intelligence (requires big data support) to determine the experience and value of such products. It is undoubtedly dangerous to treat it as a terminal product. Not to mention it is still a peculiar partial entertainment 2C positioning.

The second category is people who fall into the trap of thinking. This is especially meaningful for domestic companies. Because some of the most popular thinking patterns in China will die in this wave.

The most dangerous kind of thing here is that "the Internet age does not require core competitiveness." This view is in line with the fact that some of the top Internet companies are basically winning the model. There is really no core technology, and the persuasive power is increasing immediately.

But the actual view is very one-sided. At this point, Peter Till is more realistic. In "From 0 to 1," he not only listed "patent technology" as the first point of the core characteristics of a monopoly enterprise, but then It is a network effect and a scale effect. In addition, a chapter is written specifically to write "secrets." At this point, Huawei's choice is indeed different from other companies.

Our current thinking that does not require core competencies poses a great danger. Where there is a network effect and scale effect support, it may be okay. There is really no need for core competitiveness. As long as you can run fast, you can have the same results, but the characteristics of machine intelligence are different from those of the Internet, so it brings The above rules are likely to be inaccessible to the world.

The Internet itself is much like infrastructure, so it has often been called the information superhighway in the past few years, and the infrastructure is geared to everyone, so the model is more critical.

Machine intelligence is more like a single-point but high-value technology, such as advanced encryption algorithms, high-performance CPUs, etc., so the technical level will be critical. Your design and production model are no better, the level of technical technology is not enough, and the aero engine can't be built. And once you need to buy at a high price, you will become the downstream of the industrial chain.

Because of this difference, the model suitable for the former is not necessarily suitable for the latter. This has the greatest impact on the R&D model.

Should the engine be built with the plane?

There are two ways to get an aero engine. One is to think that it is part of the aircraft, so that the engine development will be placed in the development process of the aircraft. One that considers the engine to be independent is that it needs to be developed independently, while the specific aircraft products are chosen to fit different types of engines.

This problem is very typical. It is common in many industries. It is not clear that the choice is better because of the specific circumstances, but under the specific constraints, the right and wrong will be obvious.

Aircraft engines are clearly not suitable for development with aircraft. Because the investment and cycle it requires is not supported by the aircraft. The image point is that the product development cycle is inconsistent with the core technology cycle. This is not a problem in the era of pure product and operation, because there was not much barrier in the core technology at that time, so the core technology development cycle is relatively short, mainly because of the time to absorb the existing technology, so the problem is not big. .

The relationship between machine intelligence and the specific products that use it is a bit like an aero engine and a specific model. I feel that many big companies have realized this, so they have to work hard to set up various research labs and dig very cattle to grab high points.

The things mentioned above are only true in one situation. That is, machine intelligence becomes a standardized and open technology. This is indeed possible, such as machine intelligence as a service, such as open source.

Now, these trends do exist at the same time. Large Internet companies prefer to have complete machine intelligence technology to improve their products, while companies like IBM prefer artificial intelligence as a service, while Numenta The company is thoroughly supporting open source.

If a few large companies ran to the front, it would lead to the death of similar products without machine intelligence. In the future, Google’s unmanned driving against traditional cars may be the case.

If it is a machine intelligence-as-a-service run to the front, it is possible to rise here as a giant, such as Palantir, which is up to ten times larger.

If open source runs the fastest, such as Linux on a machine intelligence, it will return to the old story of the pattern to win.