Artificial intelligence is advancing rapidly—with serious implications. For example, biological attacks that failed in the past because of a lack of knowledge might succeed in a world in which bad actors can use AI tools to bridge the information gap.
A new preliminary study by RAND researchers explores this risk, focusing particularly on how large language models, or LLMs, could aid in carrying out a biological attack. The study employs red teams—experts pretending to be malicious actors—to scrutinize LLMs across various fictional scenarios.
In one scenario, the red teams asked an LLM about how to obtain and distribute specimens infected with Yersinia pestis, the bacteria that causes plague. In another, they asked the LLM for a cover story that could be used to help acquire Clostridium botulinum, which causes botulism.
Initial findings show that LLMs can produce concerning outputs that could assist in planning a biological attack. However, it remains to be seen whether the capabilities of existing LLMs represent a new level of threat beyond the harmful information that is readily available online.
A new RAND analysis shows how technology has affected the U.S. workforce in the past—and provides clues about how AI could affect jobs in the future. The good news: While a lot of occupations will be affected by AI, many jobs are unlikely to go away. Instead, they'll morph into different roles that meet employers' needs.
Do people judge algorithmic decisionmaking more harshly than human decisionmaking? In the future, will people be willing to hold algorithms accountable through legal channels for unfair, incorrect, or problematic decisions? New RAND survey results offer some insights.
A new RAND report explores the housing cost burden that U.S. veteran households face. The findings reveal that veteran households are generally better off than comparable nonveteran households. But that difference goes away among low-income households. Also, some veteran populations, such as women and renters, are more likely to experience housing cost burden. This may suggest areas to focus on moving forward. Read more »
More than one million people earn short-term certificates in U.S. colleges each year. Many students can then “stack” these credentials and enroll in higher-level coursework, leading to gains in earnings (especially if students go on to earn a degree). RAND researchers have been assessing what states and colleges are doing to make stackable credentials available to more students—and building evidence on whether these programs are achieving their goals. Read more »
RAND Recommends
In the Washington Post, RAND's Gian Gentile was quoted about Israel's anticipated offensive. The scope will be “much larger” than those of its recent operations, he said.
You already get the latest insights from RAND in your inbox. Why not your earbuds? Policy Currents is available as a weekly podcast. New episodes every Friday.
