AI’s greatest promises and greatest dangers are the same in the civilian world as they are in the military. It is the possibility of revealing or covering up the truth. That power could mean the difference between fair and unfair elections, or life and death in combat.
On Tuesday, leading AI innovators appeared before lawmakers on how new tools will bring value to investors and manufacturers despite being used to disrupt and mislead the public. We discussed whether
On the one hand, new artificial intelligence tools are proving useful not only for writing essays, but also for a kind of foresight. A group of researchers from Harvard University and the Massachusetts Institute of Technology recently demonstrated that generative AI models can predict more accurately than polls how people will vote based on the media they consume. Clearspeed’s product, used by the US military, insurance companies and other customers, is an AI tool that uses sentiment analysis to determine whether interviewees are answering yes or no questions honestly. offers. Steve Wisotski, a former Navy SEAL who manages government relations at the company, said. defense one The technology was developed to help the U.S. and Afghan military detect insider threats. Wisotsky says that thanks to a lot of new data, the company has refined its model to accurately identify the “risk” of interviewees lying, based on fewer than 10 questions, with an accuracy of 97 percent or more. It is said that
But there is also a growing use of AI to obfuscate the truth. Neuroscientist, author, and entrepreneur Gary Marcus testified before lawmakers about the potential threat posed by the simple tools already available. “They can and will create persuasive lies on a scale mankind has never seen before. Democracy itself is threatened.Chatbots also covertly shape our opinions, potentially surpassing the capabilities of social media. “The choice of datasets that AI companies use has profound, unseen consequences: Those who choose the data make the rules that shape society in subtle but powerful ways.”
Sam Altman, founder of OpenAI, developer of the popular new tool ChatGPT, told lawmakers: “My biggest fear is that the tech industry is doing so much damage to the world… That’s why we founded the company.” That’s a big part of why I’m here today. ”
Some key military officials share these concerns about the reliability of AI tools, even as the military seeks to use AI in a wide range of areas.
Lisa Sanders, director of science and technology for the U.S. Special Operations Command, said: defense one, “You can’t ignore [tools like ChatGPT] Because… the genie came out of the bottle. ”
Operating effectively in the gray zone of the future means “discovering the ground truth sooner than the environment would like it, faster than I might be able to do. It would mean ‘such psychic ability to reach the truth’.Ordinary people,” Sanders said. defense one said in an exclusive interview during the Global SOF event in Tampa, Florida. .
SOCOM is committed to integrating sensor data, open source data and artificial intelligence to assist operators in gray zone environments. Wearing special glasses and headphones, SOF units can better understand the emotions of locals, understand language without an internet-connected translator, and basically anticipate various events. But if such tools are unreliable, they can turn assets into liabilities. Sanders said his main concern is “how do I validate my conclusions, or raw data sources?”
The task of verifying truth doesn’t just fall to someone in the country who wrote the software and structured the data before it was deployed. A portion thereof belongs to the Operator or its team at the moment of dispute or potential dispute. So how do you give your users enough confidence to use your product in those tense moments, without overconfident operators with tools that seem only perfect?
Special operators “need to understand and fine-tune these algorithms,” Sanders said. Especially since we know that many of them come from non-military, non-national security backgrounds. There will be an inherent prejudice. So I think it’s important to be able to see if those biases are leading us down a path that we need to adjust to, and to be able to identify the biases we’re emitting. How do we do that without having a scientist with us all the time? Is not it?”
But relying on these “non-military, non-national security background” companies to produce safe and well-vetted products is a question that Congress has asked how that model actually works. , is not a good idea unless you can force them to be honest about what data goes into their model. . Congress should also allow outside government agencies and independent researchers to evaluate them, experts testified Tuesday.
OpenAI’s Altman suggested that the United States push for the creation of an international oversight body, perhaps similar to the IAEA, to oversee the development of AI tools. He proposed that the US government establish a standards and approval agency. Companies that abuse AI tools or pose risks to the public may have their licenses to use or develop their products revoked.
Not everyone in the tech community liked the idea. Eric Schmidt, former Google CEO who now runs a venture fund focused on defense of secrecy, said Sunday: was suggested Enterprise-scale AI like Google or Microsoft is too complex to be effectively regulated by anyone outside the industry.
But another former Google employee, Meredith Whitaker, said: pushed back On Twitter, “The failure of the tech industry to police itself, despite regulators enabling it, is how we got where we are. As someone who worked at Google for over 13 years, I say this: don’t be fooled by those in power who make selfish statements in a rational tone.”
Neuroscientist Marcus voiced a similar plea. To avoid the worst AI abuses, the U.S. government urgently needs to regulate the explainability of data sources and algorithms, and provide general public and private oversight.
“The whole point of the plan that big tech companies like is, ‘Trust us. But why should we? The amount of money at risk would be mind-boggling,” he said. “Mission drifts. OpenAI’s original mission statement stated, ‘Our goal is not constrained by the need to generate economic returns, but to advance AI in the manner most likely to benefit humanity as a whole.’ It was declared that Seven years later, they’re largely indebted to Microsoft, and are part of an epic battle between search engines who routinely make things up. As a result, Alphabet rushed to launch products and de-emphasized safety. Humanity has fallen behind. “
