Home > News content

Aiming at "AI patent hooligans", the U.S. Patent Office launched the operation.

via:博客园     time:2019/9/18 12:23:28     readed:106

Guo Yipu Annie comes from Wafei Temple

Quantum bit public number QbitAI

Google's hands, holding countless AI patents, give every AI practitioner a headache.

For example, DropOut by Geoffrey Hinton and RNN by DeepMind, Google Brothers, are legally owned by Google.

These are the basic tools for almost all AI research institutes and AI companies on the planet. Without them, wives can hardly cook without rice. Small companies that have just taken angel investments have to play out. Eighty percent of young people who read machine learning doctorates will be forced to give up halfway.

Although Jeff Dean said in Tokyo that the AI patents are just defensive devices to prevent being snatched by hooligans, the ghost knows if Google's family runs out of food one day, it will become a patent hooligan on its own.

After all, the beast has taken up his weapon and put on his armor, although not hitting people now does not mean that he will not go crazy in the future.

Moreover, the patent protection period is as long as more than ten years. If you do so, the AI development of Earthlings can be locked up for more than a decade.


Now, the Patent Office of the United States Patent Office, which has gone through many Google patents, has finally come to a sobering sense. It has begun to seriously consider the patent system of AI, and has made a widely available suggestion. It has made a public opinion solicitation. Before October 11th, it can find out the problems of AI patents for AI.


As soon as the news came out, Reza Zadeh, a visiting professor at Stanford at Matroid CEO, a computer vision startup, had a meal on the spot.


It's not too late to get lost.

Curious Baby U.S. Patent Office

The focus of the U.S. Patent Office's help is on 12 questions about AI patents, and they say the answers to these 12 questions

1. The invention using AI technology and the invention developed by AI are called

Examples include the problems to be solved, the database structure for training and operation, the training process for data-based algorithms, the algorithm itself, the principles and weights of the data that affect the results, and so on.

It feels like the US Patent Office wants everyone to fill out a form when they apply for a patent, but they don't know what to fill in.

2. What contribution must natural persons make to AI inventions in order to be worthy of being inventors?

For example, designing algorithms and adjusting weights, building running algorithm data, and, for example, designing algorithms based on data and coming up with results, sounds a little like

3. The contributor of an invention is not necessarily a human being (maybe a company, NGO or something). Should the current patent law and the regulations concerning inventors be amended?

4. Should an entity other than a natural person, or a company whose patent is transferred by a natural person, have a patent for an AI invention? For example, you rubbed Google's TPU to train AI. Does this AI Google own?

5. Are there any special requirements for AI inventions? (Patent Office Inner OS: People in your AI circle are afraid, sir, what do you really want?)

6. When announcing an AI invention, do you need to make any special rules?

For example, it is now common to require full disclosure of algorithms. What are the requirements on this detailed scale? Especially some deep learning systems with a large number of hidden layers, in which the weights are not artificially interfered during learning and training.

7. How can AI patent applications for inventions meet the actual landing requirements, especially considering the unpredictability of some AI systems?

8, will AI affect the average level of workers in its field of action? Be able to give AI at the level of these ordinary workers

9. What are the different technical considerations for AI inventions?

10. Do AI inventions require new intellectual property protection strategies, such as data protection?

After all, you are what you eat, what data sets you eat, and what models you train, data sets are also an important factor affecting machine learning models.

11. Are there any other issues related to AI invention patents that need to be considered?

12. Are there any strategies or practices of other patent agencies that can help the U.S. Trademark Patent Office understand the policies and practices related to AI invention patents?

These issues cover the whole process of proposals from patent review policy to whether new forms of intellectual property protection are needed.

The U.S. Trademark and Patent Office said that if you have any views on the above issues, you can send them to the relevant mailbox by October 11 this year.


Patent or open source?

One of the reasons why patent offices need to carry out AI patents separately is the open source tradition in the field of pan-computer.

According to convention, after the invention and creation of open source in the field of computer, people will be allowed to use these open source projects free of charge under certain conditions, even for commercial use, according to some agreements prevailing in the industry.

In this way, software developers can learn from each other, or simply use other people's ready-made tools, do not repeat the wheel, save time, but also improve the efficiency of the progress of the information society.

So, open source is absolutely politically correct in the field of computers. Open source contributions are like achievements in games and medals on the battlefield. They can be shown off.

CS convention, natural and natural, want to appear compelling, No Code No BB.

But the trouble is that GitHub is not opened by the US government, and so are the open source rules.

Once an invention is applied for a patent, and the owner pursues the patent right, the customary rules of the river and lake will not work. Others will be prosecuted by the patent owner if they use it casually.

In the field of AI, if you turn over the papers and look at the citations, you will understand:

Almost all innovations and applications are based on previous studies.

Moreover, these open source predecessor studies may be cited in four digits. Numerous other people in the industry around the world are using them to create updated results, which are further cited by more people.

Quoting multiple quotations, everyone is a grasshopper on a rope.

So, once some important open source projects are patented and the patent owners are held accountable, a group of people will be involved, and a group of very basic AI tools will be involved. If these are not available, then we simply do not study AI and do not commercialize AI.

Last year, it was discovered that DeepMind, Google Brothers, patented a large number of basic AI research, not only using neural networks to generate audio and video, process text, train behavior, but even RNN, the most basic tool in AI, became DeepMind's home.

Patent numberPatent namePriority date
WO 2018/048934 Audio Generation Using Neural Network 6 Sep 2016
WO 2018/048945 Sequence Processing Using Convolutional Neural Networks 6 Sep 2016
WO 2018064591 Generating Video Frames Using Neural Networks 6 Sep 2016
WO 2018071392 Neural Network for Selecting Executing Behavior for Robot Agents 10 Oct 2016
WO 2018/081089 Text Sequence Processing Using Neural Network 26 Oct 2016
WO 2018/083532 Training Behavior Selection Using Neural Networks 3 Nov 2016
WO 2018/083667 Reinforcement Learning System 4 Nov 2016
WO 2018/083668 Realization of Scene Understanding and Generation by Neural Network 4 Nov 2016
WO 2018/083669 Cyclic neural network (RNN) 4 Nov 2016
WO 2018083670 Sequence Conversion Neural Network 4 Nov 2016
WO 2018083671 Reinforcement Learning with Assisted Tasks 4 Nov 2016
WO 2018/083672 Enhanced Learning for Environmental Navigation 4 Nov 2016

On June 25, Google's Dropout patent came into effect. Dropout is a method to solve the over-fitting problem of neural networks. It can be said that Dropout is one of the basic tools in AI.

Earlier this month, Google also worked fo

News from the UK is that grave of some guy named Turing has been heard making noises since this came out.

Turing's coffin boards can't be pressed.

In addition, there are many AI applications that have been patented by Google. From the name point of view, they include the use of neural networks to complete image processing, video classification and other tasks, as well as speech recognition, image classification and other basic AI tasks.


This is equivalent to the fact that some people have applied for patents for pans, ovens and rice cookers, and chefs who use these tools face the risk of being charged a patent fee if they want to sell several dishes to their guests.

As soon as the cost is passed on, where can diners afford to go to restaurants and where can chefs open stores? Everyone can go home and work on their own.

As a result, Google has been scolded for many years. Many people in the industry feel that this large-scale patent application stifles AI innovation, and many researchers and AI entrepreneurs feel that once they are held accountable, they can not continue their business.

However, Jeff Dean later said that Google applied for patents on so many AI projects as a defensive means to prevent the touch of porcelain, and did not intend to profit from them.

After all, in the final analysis, the patent system in the United States does not adapt to the development of the AI industry, so that the industry and Google fall into

Is AI Patent Reasonable?

Despite the wide opening, the US Patent Office also received a wave of Tucao.


In response to this notification document, someone found bug:

Artificial intelligence developed and invented itself? This Dr. AI is in a complicated mood. I wonder if the Patent Office really thinks

The deeper problem is that today's patent system may not really be suitable for AI. Many people feel that the current patent system in the United States is not reliable.


It would be great if the public had done so before they started talking. To be honest, they should jump out of the idea of generic software patents and do ridiculous things over and over again.


Strictly speaking, you can only apply for software patents for one thing:

Source code.

If someone else uses different code to achieve the same result, you have nothing to say. It makes no sense for someone to apply for a software patent that a subset can't make. Can I apply for a patent for a software I don't have the skills, needs and funds to develop, and then someone else does it, and I have the patent? That makes no sense.


I think the patent office is for small developers.

1. Initial costs start at $5000-10000 and actual patents start at $10000-50000. This is an absolute obstacle for small developers who do not rely on big companies. It is not good for society, but it protects big companies. Keep in mind that an empty patent can't be eaten as a meal, but requires you to spend more money to protect your patent. Patent litigation takes a long time and costs more than $100,000 at random. Think about how many products you have to sell to support the initial patent fees and litigation fees. The patent application process is not only a joke, but our legal system also needs to be reformed.

2. 95% of patents were dropped in the first application. Such a high number means that either the application procedure is wrong or the patent office is too incompetent. Whatever the reason, the patent application process is too cumbersome and needs to be simplified.


In my opinion, the whole concept of software patent is problematic. Software is logic, logic may be complex, but it can still be simplified into a flow chart, as if you can patent something built out of Lego building blocks, but you can't patent a building block itself.


Of course, there are also many people who are actively involved, such as this doctoral student at the University of Michigan, who is actively calling on his alma mater to participate in rule-making.

After all, if the Patent Office really recognizes it, it will have to let the real AI industry have a say.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments