Large language model is a deep learning algorithm。In recent years，The corpus of the model learning continues to increase，Training methods are increasingly perfect，Large language model in text recognition、Understanding、Inference、Translation、Breakthrough progress has been achieved in prediction and generation。OpenAI's ChatGPT released on December 15, 2022 is even more prompting according to the user input，Generate coherent、Natural text，Dialogue between people。But，harmful speech and false information generated by ChatGPT，and the built-in prejudice and security loopholes of the GPT-3.5 model behind it，It has triggered a widespread discussion on the risk of artificial intelligence ethics。Around the potential ethics hazards that may cause potential ethics around large language models、Artificial intelligence technology producers' moral responsibility to stakeholders、Can technology companies maintain topics such as innovation and profitability while maintaining social moral standards，This reporter interviewed relevant scholars。

　　bet365 Play online games

Information Technology Researcher at Monash University, Australia、Terry Yue Zhuo, a researcher at the Australian Federal Science and Industry Research Organization, said，The main application scenarios of large language models include production creative content and making decisions。From a large number of books、Article and script corpus training Language model，The mode of being able to imitate text、Structure and style，to generate a similar story、Poetry、script、Advertising and marketing copywriting。In addition，Large language models can also analyze the context of the text entered by the user，Give a rational judgment and suggestion、Answer user questions, etc.。This kind of function is a chat Bet365 app download robot、Important components of digital products such as virtual assistants and text games based on text。

Advanced language models while creating opportunities，It also challenged the ethical and moral standards of society。Excellent talk to reporters，Chatgpt's moral hazards are mainly reflected in the remarks of production discrimination and exclusion、Propaganda error message、Leaks important information:。First，More than 7,000 languages ​​around the world，Among them, the English corpus (text data) held by ChatGPT is the most，The English works generated or made in English are the highest accuracy，Creation based on other languages ​​is biased。Excellent worry，The cognitive deviation of the large -scale language model of multiculturalism may enhance the prescribed concept of society，Groups that make insignificant representatives from all walks of life suffer from unfair treatment。

Canadian British Columbia University Ph.D. Candidate IFE Adebara said，Eliminate prejudice and discrimination in the language model and have a long way to go。Compared with English，Swahili、African languages ​​such as Zulu and Joruba lack high-quality corpus; developers of natural language processing technology cannot build large language models similar to GPT-3.5 for "low resources" language。In addition，Many languages ​​are different from the Indo -European language system (for example, at least 80%of African languages ​​are vocal language)，This requires developers to process such language data differently，greatly increases the cost and difficulty of development。

Next，Excellence said，If the technician does not update the knowledge reserve of the language model in time，Can't invest more time training model，The accuracy and extensiveness of the content will gradually decrease。But，Chatgpt but dialogue with users with a convincing tone，Created an artificial intelligence fantastic hallucination for users。Error information and false information transmitted by generated artificial intelligence bet365 Play online games by ChatGPT and other generated artificial intelligence will interfere with users' judgment on objective facts，Especially for the elderly and children, such as unreasonable verification of facts，It may cause users to make clinical and legal decisions that damage ethics and morality。

Last，Because ChatGPT cannot distinguish the malicious information with the data field，Some users use this vulnerability，The induction model generates harmful remarks，Reduced model security and robustness (that is, whether the system can automatically repair or ignore the failure to continue running when the system can fail)。In addition，As the language model continues to increase，The unwanted capabilities in some smaller models will suddenly appear in a larger model，Therefore, the developer believes that it is the ability of the model "burst"。but，Models with rough capabilities are also more likely to be maliciously manipulated，Personal information in the training data，Treatment of identity theft、Tracking and harassing risks。

Excellence thinks，Not just ChatGPT，Other language models will also trigger similar ethical challenges and potential moral damage。For example，Constantly improving the safety factor of the language model、Maintenance of moral standards is the long -term goal of artificial intelligence developing responsible。The responsibility of technical developers is to build a language model that complies with moral norms。For example，Reliability of the language model、accuracy、The actual efficacy and efficiency are listed as the focus of development、continuously update text data for model learning、Increase the training time and the number of test times of the model、Make heavy editing and prejudice audit。In addition，Technology companies need to disclose the source of data、Preparation before processing data and the test process of the model，To maintain the transparency of artificial intelligence research and development and deployment。

Comparison of comparison，The responsibility of the user is to eliminate immoral behaviors such as abuse or malicious attack language model。For example，Community can frequently carry out artificial intelligence bet365 best casino games science popularization and application training，Help the public understand the advantages and limitations of artificial intelligence、Clarify the responsibility and ethical meaning of using artificial intelligence technology。In addition，Users should also actively feedback the accuracy of digital technology products、reliability and ethical risk problems，Public discussions participating in artificial intelligence moral standards，Promote the improvement of relevant laws and regulations。

　　Strengthening the sense of responsibility of information technology companies

In recent years，More and more companies、Guarantee of product users and other stakeholders，They are developing "responsible" artificial intelligence technology，Among them, there is no shortage of "moral whitewashing" publicity。Irina Raicu, director of the Internet Ethics Project of Mackolla, the University of California, San Barbara, said，When talking about "technology to good"，People often pay attention to the final result brought by technology，instead of the research and development process and usage method of technology meet the standards of ethics。Relatively speaking，"Moral Technology" from the beginning, the design and development function of technical producers from the perspective of maintaining ethics，Evaluate their tolerance、accessability、Data collection and deletion security, etc.。

Roycu thinks，Before making decisions that affect ethics，Organization must first clarify the stakeholders，That is, all individuals and groups that will be affected by the decision。This step lays the foundation for the moral standards of technological products later。But OpenAI does not seem to use students and educators as its stakeholders。For example，OpenAI when introducing the limitations of ChatGPT，Acknowledge that the product is "too stubborn、Excessive use of certain phrases ... These problems originated from the prejudice caused by the data during the training process (data trainers recognize those that look more comprehensive、Bet365 lotto review longer reply) "。But，Openai didn't realize it，"More comprehensive、Sentences with longer responses "also in line with the characteristics of bad writing。Roycu thinks，If OPENAI is responsible for changing the introduction of ChatGPT to "Our model is to optimize the dialogue experience，instead of the accuracy or excellent writing of expression "，Maybe to stop the abuse of some students and other stakeholders。

Excellence told reporters，Openai adheres to the spirit of developing safe and friendly artificial intelligence，But lack of sufficient goodwill and practical actions to handle the ethical crisis brought by ChatGPT。Current，OpenAI's artificial intelligence text detectors are not high and practical，Unable to track all the content generated by ChatGPT，Not to mention to check the moral risk of them。We do not lack technologies to protect stakeholders such as students and educators，For example，Digital watermarks that protect intellectual property rights help detect plagiarism of the model。The real problem is，When information technology companies can actively take the responsibility for governance of artificial intelligence，Framework of perfecting moral decision -making。

Innovation of information technology is inseparable from the creativity and courage of technical producers，But also requires a humble attitude and the spirit of courage to take responsibility。Roycu thinks，Regulations on information technology are essential，But the law cannot cover every moral decision。The company must discuss the ethical and moral impact of development and use of artificial intelligence technology，Let ethical analysis and moral development develop into the second "nature" of information technology companies。

　　Combining moral accountability with scientific and technological innovation

School of Public and International Affairs of Princeton University、Professor Steven Kelts, a professor at the Human Value Research Center, said，During the process of chasing profit targets，Information Technology Company changed the life of the user，promoted the bet365 live casino games connection of people，It even promoted the emergence of the development model of the new community，It also affects the moral concept of society。This can be seen，Information technology products and services contain the development team's idea of ​​improving human well -being，At the same time, the moral outlook on employees also exported to the society。

But，Cacques proposed，Information Technology Company's future project management structure，Let the employees bear the challenges for their moral responsibility。Japanese management scientist Nakakaka Nakamo's "Knowledge Creation Theory" has shaped the innovative mode of contemporary information technology company。Nakagaka Yujiro pointed out in the book "Knowledge Creation Company"，Individuals and organizations have "hidden knowledge" and "explicit knowledge"。"Hidden Knowledge" refers to the personal experience that has not been formally cheified and unable to be specific skills and intuition。"Expressive Knowledge" is an objective knowledge that can be expressed in formal and institutionalized language and images。Manager matching employees with different opinions and functions work together，Encourage employees to dig deep into the hidden knowledge accumulated on weekdays，The idea has not yet been developed or has not been improved，and transform into an original product，Occupy market opportunity。

Calz said，The concept of responsibility of a person is retrospective。The concept of responsibility requires people to make mistakes that have appeared，or the error that can be foresee now，Affordable moral responsibility。For example，Employees of information technology companies must also have foreseeable responsibility for innovative technology products。This means that employees must constantly clarify the use of existing knowledge in the future，What are the well -being of information technology innovation in wide prediction of information technology innovation，instead of meeting the needs and wishes of users。

Calz Call for，Employees of information technology companies must constantly reflect on and discuss bet365 live casino games a question in the process of development technology，That is, "Is our product conducive to the positive development of human beings?" Only when organization and individual recognize their prejudice and limitations，can actively take on moral responsibility，Make more moral decisions，Developed artificial intelligence technology to maintain moral standards。