A cybersecurity researcher finds that 20% of software packages recommended by GPT-4 are fake, so he builds one that 15,000 code bases already depend on, to prevent some hacker from writing a malware version.
Disaster averted in this case, but there aren't enough fingers to plug all the AI-generated holes 😬
We updated our Library on our website about AI Literacy https://buff.ly/4at9yeq head over to see a list of papers, articles and posts that we are currently reading.
"How much do we want AI to be involved in farming? The time for that conversation is now, before these trends are irreversibly locked in. Now is the time to set reasonable ethical limits."
"For example, an EU working group proposed in 2019 that AI systems ‘should take into account the environment, including other living beings’, but this is so broad it implies no meaningful limits at all on the use of AI in farming. A review of 22 sets of AI ethics guidelines concluded – brutally – that AI ethics, so far, ‘mainly serves as a marketing strategy’."
"Our results suggest that between 6.5% and 16.9% of text submitted as peer reviews to these conferences could have been substantially modified by LLMs, i.e. beyond spell-checking or minor writing updates. The circumstances in which generated text occurs offer insight into user behavior: the estimated fraction of LLM-generated text is higher in reviews which report lower confidence, were submitted close to the deadline, and from reviewers who are less likely to respond to author rebuttals. We also observe corpus-level trends in generated text which may be too subtle to detect at the individual level, and discuss the implications of such trends on peer review. We call for future interdisciplinary work to examine how LLM use is changing our information and knowledge practices."
Generative AI can not generate its way out of prejudice
The concept of "generative" suggests that the tool can produce what it is asked to produce. In a study uncovering how stereotypical global health tropes are embedded in AI image generators, researchers found it challenging to generate images of Black doctors treating white children. They used Midjourney, a tool that after hundreds of attempts would not generate an output matching the prompt. I tried their experiment with Stable Diffusion's free web version and found it every bit as concerning as you might imagine.
Some days I wonder if, quietly, CC0 is actually a bigger success story than the other @creativecommons licenses put together. Not to slight the other licenses! But CC0 is increasingly catalytic in the library, museum, and data spaces.
@luis_in_brief@creativecommons@dajb I think CC0 was only created 10 years into CC's existence because @lessig is an academic, and giving credit (i.e., any CC license with BY) is an absolute scholarly value.
Which makes academia's embrace of genAI really weird to me, since it is based on the work of others w/o giving any credit. Why don't academics just viscerally reject genAI for that reason? Is it because this violation of a basic academic norm is happening "at scale"?
#AI#AIEthics#ResponsibleAI: "If we train artificial intelligence (AI) systems on biased data, they can in turn make biased judgments that affect hiring decisions, loan applications and welfare benefits — to name just a few real-world implications. With this fast-developing technology potentially causing life-changing consequences, how can we make sure that humans train AI systems on data that reflects sound ethical principles?
A multidisciplinary team of researchers at the National Institute of Standards and Technology (NIST) is suggesting that we already have a workable answer to this question: We should apply the same basic principles that scientists have used for decades to safeguard human subjects research. These three principles — summarized as “respect for persons, beneficence and justice” — are the core ideas of 1979’s watershed Belmont Report, a document that has influenced U.S. government policy on conducting research on human subjects.
The team has published its work in the February issue of IEEE’s Computer magazine, a peer-reviewed journal. While the paper is the authors’ own work and is not official NIST guidance, it dovetails with NIST’s larger effort to support the development of trustworthy and responsible AI.
“We looked at existing principles of human subjects research and explored how they could apply to AI,” said Kristen Greene, a NIST social scientist and one of the paper’s authors. “There’s no need to reinvent the wheel. We can apply an established paradigm to make sure we are being transparent with research participants, as their data may be used to train AI.”"
You want to take a step back and reflect on the regulation of Artificial Intelligence? I have a report for you! 🚀
❓What's in it?
As we navigate the evolving landscape of #ArtificialIntelligence, it is crucial to put democratic principles at the forefront. The report does just that and outlines key recommendations on four different topics: Design, Liability, Ethics, and Governance.
"Aftermarket" fixes applied after training, like injecting diversity terms into prompts, don't fix the underlying model and can even exacerbate harmful fabrications. If the training set is biased—and the Internet is—it's really hard to correct that after the fact.
The Ohio House have introduced legislation this month to outlaw the sharing of malicious “deepfakes”. Two proposals create civil offenses, one of the bills goes as far as to create criminal penalties for creators and sharers of such content.