Original link: https://yufree.cn/cn/2023/06/24/agi/
The Institute forwarded a notice from NIH on Friday, saying that due to confidentiality requirements, fund reviewers are not allowed to use any tools generated by artificial general intelligence (AGI) for project review, and all reviewers are required to re-sign the relevant consent forms. I looked at this email and wanted to laugh, finally came the ban, but I don’t have any confidence in how effective this ban is, it’s like asking reviewers not to use Internet resources as a reference when the Internet first came out, after all Internet content also belongs to users Generated content also has the problem of being difficult to distinguish between true and false, and search engine companies also know which words are searched more or less.
Now I am worried about the high probability of AGI, and I was also worried about Internet leaks back then. Note that the worry here is about leaks, not the model itself. Technology should be neutral, AGI training does require some materials that may have intellectual property protection, but there is a small paradox here, how to determine intellectual property rights. For example, if I wrote a blog and quoted an ancient poem, whose property right is this sentence? Of course ancient poetry has already entered the public domain knowledge, what if I am quoting a sentence from a current publication? What if I happen to write a wrong word, causing the meaning of the original expression to change, and another person discovers and spreads it? Many times I feel that intellectual property does not have a hard boundary, and it is ridiculous to try to confirm the rights of this vague concept. If there is no economic interest involved, maintaining ambiguity is beneficial to new ideas, because no one wants to check whether something has already been said before uttering a sentence, and then add a citation, so that daily communication becomes academic textual research . However, for so-called professionals, it is necessary to get to the bottom of it. Finding out the source is conducive to quantifying contributions and facilitating subsequent standardized operations. This is the cornerstone of the operation of a modern division of labor society. As long as the concept of general equivalent is still used in modern society, intellectual property rights will always be an issue.
The core issue here is the contradiction between the economic system and the knowledge system. The inheritance of knowledge must be accompanied by the process of discovery, condensing, dissemination, and rediscovery. In order to spread a series of beliefs, people even created religions and assumed that God does not care about intellectual property rights, and God’s will must be spread. But from the perspective of the economic system, knowledge will generate value, so its value must be attributed. Further speaking, the benefits of knowledge must naturally belong to the creator. So I would like to ask, who is the author of the Bible or Quran? Why didn’t I see that their descendants asked for copyright share, but the publishing company was getting the benefits? It can be refuted here that intellectual property rights are time-limited, and after the time-limitation, they belong to all mankind. The income of the publishing company’s reprint is a process of organizing typesetting and processing, which also generates new value. However, I can’t figure out how much new value is there between the “Analects of Confucius” published 20 years ago and the “Analects of Confucius” just printed today? Is there a change of concept here? The book buyer buys the content value of the book, but the book seller advertises the book’s form value. These two are not compatible.
In fact, it doesn’t matter. The economic system is for value circulation. As for whether it is an equivalent exchange or not, this is a false proposition. What I pay 100 yuan to buy a stereo is the experience. The store sells 100 yuan based on the purchase price, labor, rent and the expected profit of the industry. For pricing, the value of these two pieces is not the same thing, it just means that they are equivalent in the default transaction process in the economic system. The value of the store is the yardstick recognized by the entire economic system built by strangers, and it is also the cornerstone of standardization in modern society. There must be a quantification system in order to achieve fairness among strangers and prevent the system from collapsing.
Once the economic system enters the domain of knowledge, however, pricing becomes difficult. Sometimes you don’t know who the owner is, and sometimes you can’t market certain knowledge. The former often happens on communication platforms. Many articles have copyright, but the comments below are sometimes more valuable and referential. At this time, it is impossible to tell who is whoring for nothing. Without articles, there will be no comments, but comments will generate new knowledge. The latter is more common. I have a secret recipe that has been passed down from generation to generation. If you eat it, you can become a world champion in long-distance running. So how much is this recipe worth? How much is a long-distance running world champion worth? At least it’s worthless to me, because the secret recipe says to eat dragon meat. If knowledge is priced without a trading market among strangers, then intellectual property can be very expensive, or it can be worthless at all. This is the most intolerable aspect of the modern economic system.
AGI actually impacts the contradiction between the economic system and the knowledge inheritance system. The training of the AGI model uses the wisdom of previous people, and at the same time, it can also produce some solutions or ideas that people are not aware of. At this time, the economic system is full of pricing problems. Those who train and design this model should be paid, those who provide data should also be paid, and those who use data should pay. This is the simplest economic model. But the current problem is that the person who uses the data may also become the person who provides the data, so there is a risk of leakage. At the same time, the trained model cannot guarantee that the content inside is true. After all, we have all seen the nonsense of ChatGPT, so it seems Prohibition is reasonable.
Common sense, though, doesn’t make much sense disabling a tool that might generate useful data. This model itself is imitating the human learning process. Human learning is actually the extraction of the essence from past experience and then applied. It’s just that the effect of model learning before is not as good as that of human beings. At least it can’t be distinguished now. It is precisely because I can’t tell the difference, so I think the ban is meaningless, because you just want to use the ban to ensure that the information you get is of high quality, but if the fund reviewer uses AGI and he can’t tell it, it’s not a good display Has the level of the current reviewers been met? People who agree with the ban must also think that the opinions of professionals are more informative. They have never thought that the professionals they are looking for are actually not so professional in the frontier field, so they use AGI. If the ban is considered to be of insufficient quality, it is actually untenable It’s lame, as long as you still rely on people to judge, then people will act as a filter for AGI, and you should be able to see the nonsense. Now you issue a ban saying that it is not allowed because of nonsense. The subtext is that the previous reviewers may not even be able to distinguish nonsense. Is it more appropriate for such a reviewer to be replaced by AGI? At least we have expected the nonsense of AGI, and the opinions of unprofessional reviewers are blooming.
As for leaking secrets, it is actually two different things from using them. If the questions we use to ask will leak, the subtext is that these content will be used to train commercial models, leaking content containing intellectual property. This is also possible, but it has nothing to do with AGI. I can use the current open source large language model to train a public welfare model. The settings are all open source and set to not use user data. At least tens of thousands of programmers around the world can watch it. The kind that understands the settings, so there will be no leaks. Even if commercial companies say that they don’t need user data out of their own interests, it’s still hard for you to convince everyone that it’s useless. This problem doesn’t exist after the open-source model architecture. Musk’s open-source twitter code is a good thing. head. What I want to explain here is that open source never means free or unprofitable. After embracing open source, you should charge a fee, and you can also develop some innovative business models. Of course there may be competitors, but I have always believed that the regulation of the market can play its greatest regulating role in a market with no or weak technical barriers. A dominant market is prone to monopoly giants, which usually eventually lead to economic crises. If you don’t want to give up your commercial interests, you will most likely be taken over by the government in the end. I can’t imagine which kind of government would be happy for a certain company to monopolize the market. This will eventually challenge the government’s credibility. After all, Silicon Valley giants have all been heard on Capitol Hill.
Therefore, this new NIH regulation is destined to be crushed by the times. The resistance to new things has happened countless times in history. It would be abnormal if there is no resistance. This kind of administrative ban is basically behind the times. When I typed this sentence, I thought of the same thing that the person who agrees with gun ownership said in Houston at the beginning of the month. Guns are harmless. The problem is people. . Guns are not a new thing. I do agree with gun control but not with the ban on AGI. From a technology-neutral point of view, I am also a fan of double standards. But it doesn’t matter, the things I fight in my head are not worse than this pair, and if I try to be a completely self-consistent person, then I will probably be a thug: chug chug, you heretics! ! !
This article is transferred from: https://yufree.cn/cn/2023/06/24/agi/
This site is only for collection, and the copyright belongs to the original author.