First came surprise, then came fear.
The emergence of artificial intelligence, or sophisticated large-scale language learning models and other machine learning tools over the past year, has left policymakers wondering how these technologies will change society and the global economy in the years to come. I am desperate to see what it might bring.
While most of these models represent a clear leap in innovation, they are still not ready for mainstream or regular deployment and use in a wide range of industries.
They make up facts in a big way. They rely on the work of others, or datasets filled with copyrighted or trademarked intellectual property. Despite the efforts of their crafters, they are downright immoral and are just as happy to give you good recipes for making bombs as they are meals. OpenAI and other providers try to limit the ability of programs to do things like this, but often these protections are relatively easy to bypass for creative prompters.
As SC Media reported, experts in the world of cybersecurity continue to grapple with the impact of advanced AI systems in their field, with tools creating and detecting malicious code, fluent It has shown varying degrees of promise to create phishing emails, conduct intrusion reconnaissance, etc. To exploit known vulnerabilities and other offensive and defensive features.
Today, technology experts and policy makers in the United States and abroad are seriously thinking about how best to nurture artificial intelligence systems while preventing their worst abuses.
President Joe Biden recently warned that this technology could pose not only “dangers” but also great potential for society. Bill Gates said “we are all afraid” that AI could be used and abused by the bad guys, and did not believe regulators were up to their responsibilities, and said that early He added he feared the move would undermine the potential of the technology in the long run. Rob Joyce, the NSA’s head of cybersecurity, said the agency expects to see examples of cyberattacks that make heavy use of ChatGPT and other tools in the next year.
The National Cyber Strategy, released in March, does not directly touch on artificial intelligence, but Kemba Walden, Acting Director of the National Cyber Bureau, said that it will focus on the competition of tech companies from developing a workforce that can compete with the rest of the world in innovation. He said many efforts were being made to improve power. Better management of data and ensuring or limiting the semiconductor supply chain that provides the necessary computing power will also be relevant to AI and other technologies.
“If you break it down into pieces, you can at least address the AI cybersecurity infrastructure,” Walden told SC Media and other reporters last month.
Lawmakers are also beginning to consider the implications. A bipartisan bicameral group of lawmakers last month unveiled a bill aimed at ensuring that AI’s influence on decision-making about the use of nuclear weapons is limited and controlled by humans.
And those concerned that smaller but still serious threats, such as the use of “zero-click” spyware like the Pegasus software developed by NSO Group, could be exacerbated by the high degree of automation provided by artificial intelligence. There is also
“A leading expert told me that AI would greatly reduce the cost of launching such an attack. So what do you do? Beware of phishing attacks. If you want to train your employees to: [multifactor authentication] And now … none of that matters,” Rep. Eric Swalwell (D-California), a senior member of the House Homeland Security and Cyber Subcommittee, told reporters last week.
Protection must address morality and bias in AI systems
White House Advisor and Acting U.S. Chief Technology Officer Alexander McGillivray said this week that the government will shape and regulate AI in the future. We have listed five risk areas where
First, AI “may increase cybersecurity threats and facilitate scams that target seniors with online harassment,” raising safety and security concerns. They can “dismiss workers and unfairly concentrate economic gains” and wreak havoc on the economy. AI tools are “exacerbating stigma and discrimination in many areas already”, including criminal justice, lending, housing, recruitment and education, opening up new fronts in the fight for civil rights. They “could expose personal information in ways we haven’t yet experienced.” [and] It can further invade our privacy. The technology can also make deepfake videos, audio, images and texts more convincing, and in the long run can “endanger truth, trust and democracy itself.”
Many of these principles were included in the “AI Bill of Rights” blueprint the administration released last year, and last month officials from four federal agencies applied the principles of fairness, nondiscrimination and safety to AI. publicly committed to exercising all existing regulators to product.
Nonetheless, these efforts are not binding and governments have traditionally relied on engagement and cooperation with industry to promote common standards and protections.
Suboda Kumar, professor of statistics, operations and data science and director of the Center for Business Analytics and Disruptive Technologies at Temple University’s Fox School of Business, said the “first thing” regulators need to work on is clear mechanisms. He told SC Media that it is to ensure that She brings transparency to how AI tools are built and accountability for how they are used.
“At the moment it’s more of a black box model, which is causing a lot of people a lot of concern. [these tools] More and more over time,” he said.
Those who embed or integrate AI into their products publicly describe the “brains behind the algorithms”, how the system works, how they think, and what they use to train the system. You should be compelled to provide as much information as possible about the data you are working with. .
We need a direct way to publicly scrutinize AI company claims, ensure that their products are tested regularly, and lay the groundwork for fines and other regulatory action if they fail to meet. Kumar suggested that independent commissions or committees set up by the government could play such a role, and within the framework proposed by Senate Majority Leader Chuck Schumer (DN.Y.), AI technology He suggested that a similar independent committee would be set up to review and test the technology before it goes live. Republished.
Additionally, last week, after meeting with tech companies, the White House announced that at the upcoming DEF CON conference, many companies will allow hackers to test their systems for vulnerabilities and weaknesses. This is similar to the work the conference has done with voting systems and systems. other technology.
Regulation is coming. Regulations are here.
While discussions in the United States are still in their early stages, discussions on how to put guardrails around emerging technologies are moving faster in Europe and other parts of the world.
This week, the European Parliament overwhelmingly passed legislation to ban the use of AI algorithms in social scoring systems used in countries such as China. It would also impose a number of constraints around “high-risk” AI systems to ensure that their decisions are controlled by humans and not used for manipulative purposes, leading to discriminatory behavior.
European officials called the new rules, in the words of Italian lawmaker Brando Benifay, “a landmark piece of legislation not just for Europe but for the world” for what they see as “life-changing technology.” I hope to become
U.S. lobbyists have already expressed concern at the Information Technology Industry Council, of which many large tech companies are members, and have “thrown a broad brush” in setting limits on a wide range of AI products and technologies. There are.”
Countries such as Brazil, China and Canada are also moving to enact their own regulations.
While this could put pressure on companies and policy makers to develop their own frameworks, there are still no signs that a comprehensive legal solution like the EU’s AI law is on the horizon. The subject and industry are still new to many US lawmakers.
“There is some tension between promoting and allowing innovation, while ensuring that technology is developed and deployed in a responsible manner, and striking the right balance is a challenge for many. I think that’s what lawmakers are working on,” said one technologist. A lobbyist told SC Media.
The slower pace adopted by the US poses both risks and opportunities. Failure to act swiftly and decisively could continue to pave the way for other nations, friends and foes alike, to follow suit for U.S. companies and the rest of the world.
But it could also give lawmakers more time to mature and resolve many of the thorny issues at the heart of tech regulation debates.
Fearing that it could undermine their advantage in the fast-growing and highly competitive commercial AI arms race, companies are handing over proprietary algorithms and underlying source code and training datasets they rely on. They will be reluctant to publish it.
Additionally, large multinationals developing or working on AI tools may seek to influence the regulatory debate by favoring their own products or excluding less-funded entrants. .
Kumar said that the composition and structure of various AI tools on the market today varies greatly, and it is probably impossible to develop a single rule to govern them all. Furthermore, the metrics companies and regulators use to cross-check and scrutinize AI systems are likely to be “subjective,” but how companies explicitly demonstrate how their products are designed to mitigate harm. It may also be a deterrent to thinking about whether it is designed for
Ultimately, he said, the social risks of a slow approach are far greater and more real than the social risks associated with introducing protections too early and having to fix them later. I think.
“If we don’t act quickly, the damage could be much greater than what we have seen so far. You can always mitigate it.I think the problem is that if you migrate too late, it can cause problems…it’s hard to undo it,” Kumar said. I was.
