Is Dilira a real person? Zhihu (AI doesn't have a "no man's land") Dilira: turni

 

1. The advantages and disadvantages of artificial intelligence

AI artificial intelligence has been applied in many fields; Recently, the art field has also been invaded by AI. At an art exhibition, a contestant used AI drawing tools to generate works and won an award. In a flash, discussions about the development of AI artificial intelligence have become lively, and people have also pondered the pros and cons of AI learning.

2. Advantages and disadvantages of artificial intelligence

The gap between AI and AI artists is more the "human" gap in AI. In an art exhibition held in the United States, Allen used AI mapping tool Midjournal to generate space opera, which won the championship in the digital art category. Many artists accused Allen of cheating by using AI to generate works.

3. Debate on the pros and cons of artificial intelligence

In fact, this work was not generated by AI software and directly taken to the contest. In the middle of the contest, the author also used Photoshop to color the work in contrast to the champion work space opera, which was posted on the social platform by netizens through the AI keyword "salmon backtracking".

4. Advantages and Disadvantages of Artificial Intelligence PPT

The photos of ideal salmon swimming in pairs did not appear, but instead were pieces of salmon floating on the water trying to swim. They were also AI charted, one won an art prize, and the other was ridiculed for the differences in model design and data selection among artificial intelligence production teams, including the different descriptions used by users when using the software, resulting in different presentation effects.

5. Discussion Paper on the Advantages and Disadvantages of Artificial Intelligence

From this perspective, the people behind AI seem to be more important than artificial intelligence. Firstly, when AI is applied to a certain industry, the level of informatization in that industry should be at a high level. In fact, the level of informatization construction varies among different industries, which also means that AI will first face varying degrees of informatization challenges when entering the market.

6. Advantages and disadvantages of artificial intelligence

For the medical industry, although medical data has been digitized, the degree of digitization is not high, internal systems are dispersed, and data is difficult to circulate and share. AI pharmaceutical research requires data from the entire medical process. In order to explore the value of the data, professionals need to use intermediate systems to unify medical information in different systems or use federated learning methods for processing.

7. Advantages and Disadvantages of Artificial Intelligence: 800 Word High School Model Essay

After completing the unification of the information system, relevant personnel often need to perform secondary processing on the information itself. Some medical industry insiders told Ginkgo Technology that some doctors use slang that only industry insiders can understand when entering information, or only record the omitted parts of keywords that need to be manually "translated", and the manually standardized data can be used for AI model training.

8. Advantages and Disadvantages of Artificial Intelligence Paper

The language used for AI training is usually more standardized, while people's daily language is more popular. In order to make AI more humane in daily applications, specialized staff will "polish" their language. For example, when users communicate with intelligent voice, they often omit some key modifiers and assume that intelligent voice understands.

9. Materials on the pros and cons of artificial intelligence

And the key information that users default to intelligent voice understanding requires designers to "supplement" Tencent's car phone system. When entering the car market, they encountered a small incident where users requested voice assistants to play "On the Road". However, this is not only a song name, but also a radio station name. Faced with such polysemy, The correct approach seems to be for the voice assistant to ask the user whether to play the song "On the Road" or the radio station "On the Road".

10. Advantages and Disadvantages of Artificial Intelligence Writing 800 Words

In the actual use process, due to customer requirements, designers make the song version the default option. Some manufacturers, when facing such problems, consider that the radio version has not received many followers, and the song version has been popular recently. They prioritize playing the song "On the Road" as the default option. This ranking is a program setting for AI to increase users' familiarity with artificial intelligence during use, However, this setting will reduce the favorability of voice assistants for users who want to listen to the radio version of 'On the Road'.

When AI is sufficiently intelligent, it can automatically generate user preferences or habits based on user preferences, from data collection, product testing, to formal use. People participate throughout the process, adjust models and data in a timely manner based on AI feedback, ensuring that the desired effect is achieved technically. From another perspective, whether to play songs or radio depends on platform intervention, Thus, AI platforms have some kind of "power" to choose based on their likes and dislikes.

2、 One of the skills to teach AI "good" AI is to find out the information that people are trying to hide according to clues, which can help Holmes solve the case, and may also lead to the Amazon smart speaker echo persuading suicide event that made a lot of noise in the past few years before the information of the technology company was leaked. The smart speaker only retrieves information from its own database as a response to users' questions, and it does not understand the meaning of it, It is also difficult to empathize with the emotional harm that such a response would bring to users.

From a technical perspective, voice assistant Alexa responded to the user's question and the answer was not intellectually impaired (leaving aside human stance), but people were unable to emotionally accept this answer. Therefore, they protested to Amazon's engineer Echo and explained that Alexa likely made such an anti human response by downloading text containing malicious files from Wikipedia.

This is why it will persuade users to commit suicide. Its data source appears to be polluted. "Robots have learned toxic stereotypes through these flawed neural network models." Andrew Hunter, a postdoctoral researcher at Georgia Tech University, said that in essence, Alexa's wrong statements are extreme statements of human society, which are transmitted back to human society through artificial intelligence.

It is also because of this consideration that even if intelligent products have been in mature operation in the market for a period of time, technology companies usually arrange dedicated personnel to monitor the database and model of the product. If the intelligent speaker data is contaminated, persuading users to commit suicide may bring potential harm, while if the data of intelligent driving cars is contaminated, the harm is almost inevitable.

Saboteurs mix disruptive data into the system, thereby affecting the vehicle's judgment of the objective environment, interfering with the normal driving of the vehicle, and potentially causing the vehicle to lose control. In severe cases, attackers can even restore the system's algorithm logic, conducting indiscriminate attacks on cars loaded with the same system, bringing incalculable losses to the manufacturer.

AI has been involved in the production and circulation of Internet content. Most of the traffic distribution and content review of Tiktok and Kwai are handed over to AI completion machines to be responsible for the first information screening. According to past data, the suspected illegal content is intercepted or marked red; Afterwards, the reviewers will verify the highlighted information.

Why can't we completely hand it over to the machine? A reviewer of Tiktok said to Ginkgo Technology that the machine makes decisions based on quantitative indicators. For example, once the image of "fruit body" or related keywords are identified, it will automatically intercept or limit the flow, but many things cannot be quantified, such as artistic conception and connotation. The machine is still unable to make accurate judgments on information that can only be understood but not communicated.

Nowadays, many borderline videos do not contain sensitive words in their visuals and text, but when combined together, they are appropriate sexual cues. In the documentary "Surveillance Capitalism: Intelligent Traps", where two groups of people view the information they see as truth in the cocoon created by algorithm recommendations, and constantly strengthen this belief with algorithm recommendations, the differences in beliefs between the two sides gradually become violent incidents, And their initial argument may only be whether the Earth is round or square.

A tendentious statement, a recommendation mechanism that appeals to its own interests, can be seen separately without any problem. Together with the catalysis of time, it turns into a violent event. Neither party is responsible, and neither party seems to be responsible. Compared with the first two, the externality of AI technology may need more attention from technology companies. 3. AI is not a panacea

After crossing technical barriers and avoiding various risks, can the AI we use obediently bring better economic benefits to the company? Not necessarily for music platforms with a library of up to 70 million songs, it is not realistic to require music editors to recommend works after listening to all the content. Indeed, AI can quickly "listen" to all the music and recommend niche works with similar styles based on the audience's aesthetic preferences, thereby improving the utilization rate of the entire library. Back then, with its unique algorithm logic, Xiami Music could still compete with NetEase Tencent despite copyright weaknesses.

However, algorithm recommendations also have drawbacks. Many users have reported that the recommended music style is too single, and music platforms have also tried to change the logic of the algorithm by adding works with significant style differences in daily recommendations, in order to alleviate the audience's aesthetic fatigue. Even algorithm recommendations that appear to be completely handed over to machines are still executed according to human will.

Machine recommendations always appear a bit cold, and this lack of human touch is supplemented by various song lists spontaneously uploaded by users. The style of the songs in the user created playlists is not high in coherence, but enthusiastic and full. People spontaneously collect comments, which brings the distance between users and the platform closer. For overall commercial interests, most content platforms use AI and manual recommendation methods to learn from each other and make up for their shortcomings.

Due to its cultural uniqueness, the content industry cannot be completely handed over to machines; And it is entirely possible to use AI instead of industries that have better results. Many companies still use manual labor, partly because the benefits brought by using AI cannot cover the initial investment costs. A practitioner engaged in traditional industries told Ginkgo Technology.

If the adoption of AI technology brings a profit of 1 million yuan but increases costs by 1.1 million yuan, and this cost may not necessarily decrease within three to five years, then he tends to maintain the status quo. If the cost and benefits are basically the same, but the cost is expected to significantly decrease within three to five years, then he will be willing to try to solve the cost problem. The above practitioner also expressed that he hopes to ensure the absolute privacy of the company's data Conducting AI model training under absolute safety conditions, if this cannot be guaranteed, even if the benefits of using AI technology are higher, it is likely that they will not adopt AI technology.

In fact, the practitioner is concerned about data security and privacy leakage issues, even if they are completely using ordinary machines or manual labor. He told Ginkgo Technology that in his opinion, AI information leakage must have a greater impact than traditional technology information leakage. Moreover, the law has clear relevant regulations for issues such as leakage in traditional technology, and if problems arise, they can directly go through legal procedures.

And the legal provisions for AI technology in this regard are still being formulated. He stated that he is more willing to consider the integration of AI technology from the technology itself, to the integration of technology with human society, and to the true value that AI brings to users after these supporting facilities are basically completed. Every step of this requires human participation, and the role of human beings changes from being frontline workers to being regulators who command AI to carry out frontline labor.

People use data and models to cultivate an assistant. How to guide this assistant to achieve their desired market effect is still in the exploratory stage. Author: Ginkgo Technology Source official account: Ginkgo Technology (ID: yinxingcj), bring you the best business figures and stories. This article is published by everyone who is a product manager, the cooperative media @ Ginkgo Finance and Economics. It is prohibited to reprint without permission.

The title image is from Unsplash, and based on the CC0 protocol, this article only represents the author's viewpoint. Everyone is a product manager, and the platform only provides information storage space services.