The System Can Give AI Moral Awareness — Sometimes

Delphi finds results in recent advances in AI and languages. Feeding a lot of text to algorithms that use neural network mathematics has made amazing progress.

In June 2020, researchers at OpenAI, a company that specializes in cutting down AI weapons, developed a software called GPT-3 that can predict, summarize, and create word-for-word. and what is often seen as a unique skill, though they will also spit on the words of prejudice and hatred learned from the words he read.

Researchers behind Delphi also asked questions about GPT-3. He found that his responses were consistent with those of more than 50 percent of the staff at the time — slightly better than the monetary unit.

Improving its management system like Delphi will require a variety of AI methods, which may include some that allow the machine to express its ideas and express them when in conflict.

The idea of ​​giving machines a code of ethics dates back decades to academic research and scientific fiction. The famous Isaac Asimov Three Rules of Robotics widespread the idea that machines can imitate human behavior, although short articles that explored the concept underscore the contradiction of such simple reasoning.

Choi says Delphi should not be seen as providing a definitive answer to any cultural questions. The most advanced type may show uncertainty, due to various assumptions in its studies. He says: “Life is very difficult. “No two people can fully agree, and there is no way the AI ​​program can conform to human judgments.”

Some machine learning machines have shown their own blind dots. In 2016, Microsoft released a chatbot called Tay for learning from online chat. The program was done quickly being destroyed and being taught to say offensive and disgusting things.

Attempts to explore the best ideas for AI have also highlighted the difficulties of such work. Work founded in 2018 by researchers at MIT and elsewhere investigated how people view social problems that can be encountered by self-driving cars. He asked people to decide, for example, whether it would be right for a car to hit an older person, a child, or a criminal. This work revealed different perspectives in different countries and groups of people. Those from the US and Western Europe were more likely than those who were asked in other countries not to harm the child than an adult.

Some AI developers are eager to participate in ethical issues. “I think people are right to point out the flaws and shortcomings of the model,” says Nick Frost, CEO of Cohere, a startup that has made a great example of a language that is available to others through the API. “They are a reflection of a bigger, bigger problem.”

Cohere developed strategies to improve the output of its algorithms, which are now being tested by other businesses. It controls the content of the algorithm and trains algorithms to learn how to deal with racist or hate speech.

Frost says the controversy surrounding Delphi highlights a key question facing the professional industry – how to create the right technology. Often, he says, when it comes to correct the content, false, and algorithmic bias, companies are trying to wash the hands of the problem by saying that all technology can be used for good and for bad.

When it comes to ethics, “there is no truth, and sometimes professional companies give up because there is no truth,” says Frost. “The best way is to experiment.”

Updated, 10-28-21, 11:40 am ET: The first article in this series stated that Mirco Musolesi is a professor of philosophy.

Some of the Best WIRED Stories

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *