Text / brain body
One of the major events in the AI community these days is that Facebook, known as "North American AI Big Five," suddenly announced a complete reorganization of its AI team and management architecture.
One of the most significant changes to Facebook's AI business card, artificial intelligence godfather LeCun announced that they no longer serve as head of the FAIR team, focused as chief scientist, will be more energy into academic work.
This may seem like a mere corporate change, but Facebook's big move to adjust AI is obviously not a purposeless chaos. Hidden behind its strategic thinking, perhaps hinting at today's node, the giant companies how to treat the development of the AI industry.
Today we come to open a hole in the brain, trying to strip behind this thing, hiding what details and hints about the direction of the industry.
From a larger perspective, the two different AI paths are affecting the AI world today.
The pace of commercialization: AI's two-way battle
The first thing to understand is exactly what Facebook is doing: why Zuckerberg and Schlumpf are willing to downplay LeCun as an AI card, and insist on AI reorganization?
The most direct reason is apparently due to LeCun's personality and scientist status, its team style is too free and academic, although the research strength is strong, but the research and development has always been with the layout of Facebook's industry at random.
Obviously, dissatisfaction with the AI industrialization progress is the fundamental reason for this system adjustment. Following the reorganization, the management structure of the Facebook AI department has been streamlined and AML will work more closely with the FAIR. This means that Facebook hopes to rapidly accelerate the commercialization of ICT research results.
What really deserves our attention is the Facebook internal AI "revolution" that seems to convey the two options that technology companies face in the face of AI today. If these two roads compared to the fork in the AI, then they are like this:
Road A: The company is heavily invested to create a research system, allowing the research system to develop on its own. Waiting for the results of natural maturity, remitted to the company's business. This is the common European and American technology giant to open AI when the basic game, basically in accordance with Google's model.
Road B: Quickly make the academic system and industrial system frequent, orderly and scale connections, in accordance with the needs of the industry reverse customization academic breakthroughs, high-frequency open to create industry-university integration.
Obviously, Facebook hopes to adjust the pace, began to move from A to B's own industrial deployment.
It is interesting to note that this choice is not the only one in the fight, it can even be said that this is a very mainstream thing in AI today.
Its most direct frame of reference, probably Baidu nearly a year of AI business and industrial restructuring. In particular, not long ago we saw Baidu Research Institute has just carried out a new round of upgrades. Introduced a new top-level scientist and focused on two highly commercially viable topics, "Business Intelligence" and "Robotics and Autopilot".
So tacit synchronization, no wonder there are comments in the Western media that this is Facebook many times to maintain the strategy with Baidu AI synchronization.
However, this interesting phenomenon may hide behind the general question of the AI community: How do enterprises deal with the relationship between academic research and commercialization? How to deal with the relationship between ecology and the entire AI industry? Is the traditional road A dead in this fast-moving AI lane?
Inter-industry balance: "Baidu model" growth of all things
In today's AI world, a common question is whether academia should dominate the AI industry and allow academics to develop without rules in the corporate world. Is not it a good thing?
Of course it is a good thing to respect academic freedom and give great support to academic issues. But today there are two criticisms of AI's tradition of industry-university segregation: First, whether to allow academic research that is hard to get into commercial applications back to schools and purely research institutes, otherwise wasting money from investors to do "research and development" It seems inappropriate; Second, after all, limited enterprise resources, whether to find a balance between production and learning, focused on breaking the most awaited AI enterprises to solve the problem. Such as Baidu research system in a high degree of investment in driverless and industry-academia, has been verified in this year's Apollo growth is correct.
Coincidentally, in recent years, Microsoft, IBM, Amazon, are one after another to adjust their own academic research system and industrial connectivity, planning out a clear overall, there is a clear strategic intent of the production-and-learning frequency system. Including recent adjustments made by Baidu, are focused on focusing on the direction of the industry, upgrading the research system, strengthening the integration of industry-system docking and so on.
Another interesting anecdote is that Facebook first started doing AI, setting up two AI labs, which is said to have been recommended to Zach Burke by Robin Li.
So is it possible, Facebook one of the underlying drivers of structural adjustment is that hope to learn and imitate Baidu's success?
In stark contrast, perhaps the Google model has been subject to much controversy in recent times. For example, in the product of high-level personnel and engineering personnel AI today, Google is still flying and continue to lead Li Fei Fei entrusted with the task. On the other hand, Google's industrial layout still maintains a large and comprehensive "full blossom pattern"; there is almost no AI products Google does not do, but at the same time it is hard to say which product has done particularly well. Lack of focus and focus, the lack of product thinking, have become the shadow of Google in the past year.
The DeepMind continued to maintain a state of loss, Google's brain difficult to provide clear support for the industry and other issues, but also shows the academic side of the rate seems to be different from the industry side.
This may indicate that the traditional AI model of the separation of production and education is increasingly trying to integrate industries and schools in the direction of business representatives such as Baidu. Balance began to favor AI forces with clear plans and a high degree of industrial organization.
It is a problem that the increasingly closed or continuous opening up
Hidden in the back of the balance of production and learning, is how companies think AI, AI industry awareness.
Last May Google 2017 conference in Google, Google fully defined "Al first" strategy. And then the first performance is full array of "hardware + software package", created by far the world's largest AI industry group.
Obviously, Google's understanding of AI first is Google first in AI. In the strategic starting point to choose the big and the whole model, Google's AI ecology began to show two significant manifestations: accounted for pit and exclusive.
The so-called "accounting pit" is that whenever someone came up with AI products, Google should have a similar. So we see that Google has a Google Voice Assistant similar to Apple's Siri, smart speakers like Amazon Echo, Facebook's information and photo assistants, and more. More phone, tablet, camera, wearable devices, etc. AI hardware.
And "exclusive" is Google in promoting AI first, gradually began to abandon the cooperation and division of labor in the Internet industry guidelines. Start the whole process of AI Google. For example, we have seen in the calculation of the power of TPU and Google Cloud provides AI integration, the algorithm to TensorFlow as the center to create a closed-loop ecology, Google data blockade in the data on the possibility of foreign cooperation in the talent started all over the world R & D talent battle.
In fact, Google's AI strategy today, you can use Chinese Internet users are very familiar with a word to explain: Happy Valley. This is why Google does not care about the synchronization and connection between production and learning. Because in Google's progress, collecting and occupying the best academic resources, and a steady stream of results into the Google system is the first. Therefore, the academic atmosphere of the arms race enterprise is still popular in Google.
Can be described as large and comprehensive, means increasingly satisfied with the self-cycle and closed with the outside world. So in today's European and American AI world, an anti-Google mood may be brewing. The reason behind this, there are media attributed to the strict implementation of Google, and increasing ecological closure and all-Google strategy. In other words, hegemony is daunting.
As AI is a highly diverse and diverse category of technology, the implementation of these strategies means that Google is increasingly squeezing the power of its American counterparts. From the exclusion of TensorFlow, the absolute Google cloud service and TPU, to the closed social business of voice services, Google is cracking down on Facebook as "latecomers." Today's Google is very much like Tencent before the war 3Q, there is a "where AI will Google" temperament.
But AI must be closed and dominance it? This may also be a question that we should think about today.
In contrast, another model of AI industry can be said that Baidu's representative of the open cooperation model.
To be sure, Baidu is practicing Luci's "multiple routes to multiple friends". In the industrial ecology, technological openness and strategic cooperation has an open mind and high-quality cooperation case. This is a model that is more feasible today in the AI sector and more accepted by the industry.
Another focus on open or closed industries comes from the impact on developers. Google's development ecology is indeed continuing to increase the technological absorption, but the real progress has not been as much as Google expected the ideal, a large part of the reason is the over-closed industrial ecology, the selectivity of developers too small.
For example, Google's AI course and training program is completely targeted at TensorFlow, providing developers with hardware APIs and access to TensorFlow systems. The TensorFlow community resources and new AI development tools provided are fully deployed on Google Cloud.
In simple terms, once a developer enters the Google system, it is basically prohibited to introduce the advantages of any other company or platform on any one port. After frameworks, cloud computing and hardware triple-barrier architecture, Google's developer system has become more and more closed, and even to some extent forced developers to choose the situation.
AI is a rapidly changing, there are surprises in the world of technology, the choice of Google is equivalent to giving up the world, apparently many developers are reluctant. This point Baidu's compatibility and create the foundation is relatively better, for the developer's enabling plan is also more diverse. Many US technology companies, it seems, are also now more inclined to Baidu represented by "platform and developers to explore" mode, gradually away from the strict and rigid platform logic.
In general, Google is becoming increasingly closed today, and most of the new AI forces represented by Baidu are continuing to tap the power of eco-discourse through opening up. This forms a very representative AI road, more and more companies are adjusting themselves to find ways for themselves to get along with the times. Of course, how the future is hard to judge. But constant attempts and self-doubt are in fact a symbol of a dynamic industry.
Two models of the dispute, may become 2018's AI annual drama.