Up to now, Facebook, Google, Microsoft and IBM have successively opened source artificial intelligence. The so-called open source means that the source code of the program is open and anyone can use and modify it at will.
Nothing for nothing. What is the purpose of this move by several major Internet companies? What impact will it have on the development of global artificial intelligence?
Open source what
On February 7th, 65438, IBM announced that it would provide the source code of the artificial intelligence tool system ML (machine learning) for external programmers free of charge through Apache Software Foundation. Rob Rob Thomas, vice president of IBM development, expressed the hope that this will be the first step for the widespread popularization of system ML.
It is understood that System ML was originally developed by IBM 10 years ago, which can simplify the difficulty for developers to customize machine learning software. For example, the software can help banks to write risk modeling software to warn in advance when fraudulent activities are discovered. The current version hopes to be used together with another software to help enterprises deal with a large amount of data on smartphones or fitness bracelets.
IBM is the fourth technology giant to open its proprietary machine learning technology to the outside world this year. Facebook opened some artificial intelligence software of Torch Software in February. Google11/kloc-0 opened TensorFlow system in October. The system can be used in many machine deep learning fields such as speech recognition or photo recognition, and can run on a server cluster composed of thousands of computers or a single smart phone. Currently used in search, photos, inbox applications and other products.
In the same month, Microsoft also launched the machine learning open source toolkit DMTK. This open source plan allows small and medium-sized enterprises to realize the work of thousands of supercomputers with several computers-because artificial intelligence needs to deal with huge amounts of big data. The full name of DMTK is dstributed machine learning toolkit, that is, distributed machine learning toolkit. One of the tools is called LightLDA, which is an extensible, fast and lightweight system for large-scale theme models.
Microsoft said, "In the distributed implementation, a lot of system optimization has been done to enable LightLDA to process very large-scale data and models on an ordinary computer cluster. For example, on a cluster of 8 computers, an LDA model with 1 10,000 words and 1 10,000 topics (about 1 trillion parameters) can be trained on a data set with 200 billion training samples. "
George thomas Jr of Microsoft Research Institute said that with these tools, developers can deploy large-scale machine learning with fewer servers.
What's the intention?
Why should the software that big companies have worked so hard to develop be open to the public for free?
The understanding in the industry is that various artificial intelligence companies are actively recruiting machine learning talents, and open source code can attract external talents to participate in project collaboration and improve related technologies. It is also possible for them to recruit some talents from the third-party community.
Thomas said that IBM hopes to attract more programmers to use and study this software by opening the source code of System ML, thus accelerating the development of this software. "Our current research and development is limited by the budget, so we need to open source and accelerate innovation."
"China's domestic artificial intelligence research has not really started, so there is no great demand for talents. We still focus on traditional business, and we have never heard that any company should focus on developing artificial intelligence. " A BAT executive told reporters.
Google said in official website that it hopes to attract more researchers, find new uses and improve the system by releasing it. "TensorFlow has not been completed, and needs to be adjusted, modified and expanded."
It is worth noting that although the system turned to open source, Google left behind some things that can make its machine learning technology unique: massive data, computer networks that can run software, and a huge team of artificial intelligence experts that can adjust algorithms.
Christiani, a professor of artificial intelligence at the University of Bristol in the UK, said: "Google's move is not rational. Deep learning is not plug and play, it requires a lot of testing, adjustment and adaptation. "
One of the important purposes of Google's disclosure of its own system is to attract more artificial intelligence experts to make suggestions for the improvement and application of software. "There are millions of parameters in this system that need to be adjusted. If there is no engineer to do this work, then the deep learning algorithm released by Google this time is extremely limited. "
How to make a profit
Companies are competing for open source, so that they can also use each other's open source programs.
Google will use the artificial intelligence systems of Microsoft, IBM and Facebook, and it will also be used by others. In addition, developers can integrate all open source programs and create their own new systems.
The giant Internet companies in China will also use the open source programs of companies such as Google, Facebook, Microsoft and IBM, and modify them to become their own artificial intelligence programs. But the research on artificial intelligence in China is not as hot as that in the United States. "Except Baidu, no company in China regards artificial intelligence as an independent business unit." BAT's technical executives told the reporter of International Finance News.
The research and development of artificial intelligence software needs a lot of manpower and material resources, so can these open source software still achieve commercial profit?
A core technical executive of BAT analyzed with the reporter of International Finance News, "Just like Google's Android system is open source, the more people you use, the more related things around you. Google can make money from other places, such as providing supporting services, such as advertising, such as special equipment. As long as there are many people, you can make money. "
"Among them, advertising is a very important source of income. Google's open source system has a strong correlation with Android, and more people use it. As long as there are many people, you can make money. " The technical executive told reporters. "It is also possible to produce hardware equipment and use more people. Specialized equipment can be provided, such as intelligent hardware, and software and hardware form a more complete ecosystem. "
However, the technical executive said, "There is no clear scenario for the profit model of artificial intelligence services. Although it can improve and upgrade existing systems, machine learning itself has no clear mode of making money. "