Does artificial intelligence manage the government? This is what we know

The Trump administration allows the Chatbots AI line.

Federal agencies such as Public Services Department and Social Security Administration Chatgpt-asque has launched their workers. the Old Warrior Affairs Department The artificial intelligence is used to write the code.

The American army has deployed Camogpt, The II Reed tool, to review documents to eliminate references to diversity, fairness and inclusion. More tools come down on the line. the Ministry of Education He suggested using the Importer IQ to answer questions from students and families about financial aid and payment of loans.

The Monochological Intelligence aims to automate the tasks of the previous government workers, with 300,000 discounts in jobs from the federal workforce by the end of the year.

Meg Young, researcher at Data & Society, says, Independent Research and Policy Institute in New York City.

“We are in the crazy noise cycle,” she says.

What do the artificial intelligence of the American government do?

Currently, government Chatbots aims to a large extent for public tasks, such as helping federal workers write emails and summarize documents. But you can expect government agencies to give them more responsibilities soon. In many cases, childish artificial intelligence is not at the level of the task.

For example, GSA He wants to use AI Tawylidi for purchasing tasks. Purchases are the legal and bureaucratic process by which the government buys goods and services from private companies. For example, the government will take the purchases to find a contractor when building a new office building.

The purchasing process includes lawyers from the government and the company that negotiates a contract that ensures that the company adheres to government regulations, such as the requirements of transparency or the requirements of the American Disabilities Law. The contract may also contain repairs to the company after delivery of the product.

It is unclear that the artificial intelligence will accelerate purchases, according to the Yong newspaper. It says, for example, it can make it easier for government employees to search and summarize the documents. But lawyers may find Amnesty International at the subject of error exposed to its use in many steps in the purchase, which involves negotiations on large sums of money. AI Tolidi may waste time.

Lawyers must carefully examine the language in these contracts. In many cases, they already agreed to the acceptable formulation.

“If you have Chatbot generates new conditions, it creates a lot of work and burns a lot of legal time,” says Young. “The most saving time is just copy and paste.”

Government workers also need to be vigilant when using artificial intelligence in legal issues, because they are not reliably accurate in legal thinking. A 2024 studies It was found that Chatbots is specially designed for legal research, which was released by LEXISNEXIS and Thomson Ruteers, committed realistic errors, or hallucinations, from 17 % to 33 % of the time.

Sorrani says that companies have issued new legal tools since then, but promotions have similar problems.

What are the types of errors made by artificial intelligence?

Types of widespread errors. The most prominent of which was in 2023, lawyers were approved on behalf of a customer who sues Avianca airlines after they were martyred in the unknown situation that was created by ChatGPT. In another example, Chatbot, the trainer, said on legal thinking that the Nebraska Supreme Court canceled the United States Supreme Court, says Fayez Surni, the co -author of the 2024 study.

“This is still mysterious to me,” he says. “Most high schools can tell you that this is not how the judicial system works in this country.”

Other types of errors can be more accurate. The study found that Chatbots has difficulty distinguishing between the court’s decision and the judicial argument. They also found examples as LLM cited under a law.

Surani also found that the chat keys sometimes fail to identify the inaccuracy of the claim itself. For example, when he is asked to ask about the rulings of a fictional judge named Luther A. Welgar, Chatbot answered in real condition.

Legal logic is very difficult for artificial intelligence because the courts fluctuate legislative cases and bodies cancel laws. Sorrani says that this system makes it expressions about the law, “it can be 100 % correct at some point, then it stops immediately to be completely correct.”

This is explained in the context of a technique known as a generation equipped, which is usually used by legal conditions a year ago. In this technology, the system first combines some relevant cases from a database in response to a claim and creates its results based on these cases.

But this method still produces errors often, I found a 2024 study. Wade and Planthood V. Casey, for example, say yes. But it will be wrong, as Roe was rejected by Dobbs V. Jackson Women Health Organization.

In addition, the law itself can be mysterious. For example, the tax law Not always clear What you can delete as medical expenses, so that the courts can consider individual cases.

“The courts have differences all the time, so the answer, even if it appears to be a simple question, can be somewhat unclear,” says Leg Osovsky, a law professor at North Carolina University, Chaplin Hill.

Are your taxes delivered to Chatbot?

Although the internal revenue service is not currently providing the Chatbot program, which works on behalf of artificial intelligence for public use, a 2024 Report of the Tax Authority He recommended more investment in the abilities of artificial intelligence for such a chat.

Obstest artificial intelligence can certainly be useful in the government. A Experimental program In Pennsylvania, in partnership with Openai, for example, it showed that the use of Chatgpt has saved people 95 minutes a day in administrative tasks such as writing emails and summarizing documents.

Young notes that researchers who manage the program did this in a measured manner, by allowing 175 employees to explore how ChatGPT can fit into the current workflow.

But the Trump administration did not follow a similar seizure.

“This process that they follow shows that they do not care if artificial intelligence is working for its declared purpose,” says Young. “It is very fast. It is not designed in the functioning of certain people. It is not carefully published for narrow purposes.”

The GSAI administration has released an accelerated timetable to 13,000 people.

In 2022, Osovsky I conducted a study Government legal guidance, including Chatbots. Chatbots did not use artificial intelligence. Their studies make several recommendations to the government on Chatbots for general use, such as those proposed by the Ministry of Education.

They recommend the presence of Chatbots with the evacuation of the responsibility that informs users that they do not speak to a person. Chatbot should also explain that its removal is not legally binding.

Currently, if Chatbot tells you that you are allowed to deduct specific work expenses, but the Tax Authority does not agree, you cannot force the Tax Authority to follow the Chatbot response, and Chatbot should say that in its production.

Government agencies also need to adopt a “clear series of leadership” that shows who is responsible for creating this chat and maintenance.

During their study, they often found that people who were developing chat were technology experts who were somewhat wearing other employees in the department. When the agency’s approach changed in legal guidance, it was not always clear how developers should update their chat groups.

With the government intensifying the use of obstetric artificial intelligence, it is important to remember that technology is still in its infancy. You may trust her to reach the recipes and write your condolences, but the governance is completely different.

Young says technology companies do not yet know what cases of use of artificial intelligence will be useful. Openai, HotHROPIC and Google are actively looking for these cases of use by Partnership with governments.

“We are still in the first days of assessing what artificial intelligence is and not useful in governments,” says Young.