I see this as a failure to ‘work through’ the technology. After all, “Any sufficiently advanced technology is indistinguishable from magic” Arthur C. Clarke
https://www.scientificamerican.com/article/how-can-we-trust-ai-if-we-dont-know-how-it-works/
How Can We Trust AI If We Don’t Know How It Works
(Related)
https://www.bespacific.com/can-sensitive-information-be-deleted-from-llms/
Can Sensitive Information Be Deleted From LLMs?
Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attacks. Vaidehi Patil, Peter Hase, Mohit Bansal: “Pretrained language models sometimes possess knowledge that we do not wish them to, including memorized personal information and knowledge that could be used to harm people. They can also output toxic or harmful text. To mitigate these safety and informational issues, we propose an attack-and-defense framework for studying the task of deleting sensitive information directly from model weights. We study direct edits to model weights because (1) this approach should guarantee that particular deleted information is never extracted by future prompt attacks, and (2) it should protect against whitebox attacks, which is necessary for making claims about safety/privacy in a setting where publicly available model weights could be used to elicit sensitive information. Our threat model assumes that an attack succeeds if the answer to a sensitive question is located among a set of B generated candidates, based on scenarios where the information would be insecure if the answer is among B candidates. Experimentally, we show that even state-of-the-art model editing methods such as ROME struggle to truly delete factual information from models like GPT-J, as our whitebox and blackbox attacks can recover “deleted” information from an edited model 38% of the time. These attacks leverage two key observations: (1) that traces of deleted information can be found in intermediate model hidden states, and (2) that applying an editing method for one question may not delete information across rephrased versions of the question. Finally, we provide new defense methods that protect against some extraction attacks, but we do not find a single universally effective defense method. Our results suggest that truly deleting sensitive information is a tractable but difficult problem, since even relatively low attack success rates have potentially severe societal implications for real-world deployment of language models.”
A slippery slope? (Lots of loopholes and ChatGPT will find more.)
https://www.illinoispolicy.org/chicago-starts-taxing-chatgpt-artificial-intelligence/
CHICAGO STARTS TAXING CHATGPT, ARTIFICIAL INTELLIGENCE
Add ChatGPT to the list of things Chicago taxes: As of Oct. 1, Chicago’s personal property lease transaction tax slapped a 9% tax on the artificial intelligence platform.
The tax applies to leased computer platforms such as ChatGPT’s premium subscription. Users can avoid the tax by opting for the free version. If someone works in the city but mostly uses ChatGPT outside the city, they aren’t subject to the tax.
A tipping point?
https://www.justsecurity.org/89033/ai-and-the-future-of-drone-warfare-risks-and-recommendations/
AI and the Future of Drone Warfare: Risks and Recommendations
The next phase of drone warfare is here. On Sep. 6, 2023, U.S. Deputy Defense Secretary Kathleen Hicks touted the acceleration of the Pentagon’s Replicator initiative – an effort to dramatically scale up the United States’ use of artificial intelligence on the battlefield. She rightfully called it a “game-changing shift” in national security. Under Replicator, the U.S. military aims to field thousands of autonomous weapons systems across multiple domains in the next 18 to 24 months.
Yet Replicator is only the tip of the iceberg. Rapid advances in AI are giving rise to a new generation of lethal autonomous weapons systems (LAWS) that can identify, track, and attack targets without human intervention. Drones with autonomous capabilities and AI-enabled munitions are already being used on the battlefield, notably in the Russia-Ukraine War. From “killer algorithms” that select targets based on certain characteristics to autonomous drone swarms, the future of warfare looks increasingly apocalyptic.
Amidst the specter of “warbot” armies, it is easy to miss the AI revolution that is underway. Human-centered or “responsible AI,” as the Pentagon refers to it, is designed to keep a human “in the loop” in decision-making to ensure that AI is used in “lawful, ethical, responsible, and accountable ways.” But even with human oversight and strict compliance with the law, there is a growing risk that AI will be used in ways which fundamentally violate international humanitarian law (IHL) and international human rights law (IHRL).
… Dubbed the “first full-scale drone war,” the Russia-Ukraine War marks an inflection point where states are testing and fielding LAWS on an increasingly networked battlefield. While autonomous drones reportedly have been used in Libya and Gaza, the war in Ukraine represents an acceleration of the integration of this technology into conventional military operations, with unpredictable and potentially catastrophic results. Those risks are even more pronounced with belligerents who may field drones without the highest level of safeguards due to lack of technological capacity or lack of will.
The lessons from the war in Ukraine include that relatively inexpensive drones can deny adversaries air superiority and provide a decisive military advantage in peer and near-peer conflicts, as well as against non-state actors.
(Related) Fight on the front lines without leaving your couch. (No need to understand strategic objectives.)
8 rules for “civilian hackers” during war, and 4 obligations for states to restrain them
Written by Tilman Rodenhäuser and Mauro Vignati:
As digital technology is changing how militaries conduct war, a worrying trend has emerged in which a growing number of civilians become involved in armed conflicts through digital means. Sitting at some distance from physical hostilities, including outside the countries at war, civilians – including hacktivists, to cyber security professionals, ‘white hat’, ‘black hat’ and ‘patriotic’ hackers – are conducting a range of cyber operations against their ‘enemy’. Some have described civilians as ‘first choice cyberwarriors’ because the ‘vast majority of expertise in cyber(defence) lies with the private (or civilian) sector’.
Examples of civilian hackers operating in to the context of armed conflicts are diverse and many (see here, here, here ). In particular in the international armed conflict between Russia and Ukraine, some groups present themselves as a ‘worldwide IT community’ with the mission to, in their words, ‘help Ukraine win by crippling aggressor economies, blocking vital financial, infrastructural and government services, and tiring major taxpayers’. Others have reportedly ‘called for and carried out disruptive – albeit temporary – attacks on hospital websites in both Ukraine and allied countries’, among many other operations. With many groups active in this field, and some of them having thousands of hackers in their coordination channels and providing automated tools to their members, the civilian involvement in digital operations during armed conflict has reached unprecedented proportions.
This is not the first time that civilian hackers operate in to the context of an armed conflict, and likely not the last. In this post, we explain why this trend must be of concern to States and societies. Subsequently, we present 8 international humanitarian law-based rules that all hackers who carry out operations in the context of an armed conflict must comply with, and recall States’ responsibility to restrain them.
Read the 8 rules and discussion at EJIL.
Some groups have told BBC that they will not comply or will not comply with all the rules.
An “R” rated LLM trending to “XXX” – why not?
Nearly 10% of people ask AI chatbots for explicit content. Will it lead LLMs astray?
With the overnight sensation of ChatGPT, it was only a matter of time before the use of generative AI became both a subject of serious research and also grist for the training of generative AI itself.
In a research paper released this month, scholars gathered a database of one million "real-world conversations" that people have had with 25 different large language models. Released on the arXiv pre-print server, the paper was authored by Lianmin Zheng of the University of California at Berkeley, and peers at UC San Diego, Carnegie Mellon University, Stanford, and Abu Dhabi's Mohamed bin Zayed University of Artificial Intelligence.
A sample of 100,000 of those conversations, selected at random by the authors, showed that most were about subjects you'd expect. The top 50% of interactions were on such pedestrian topics as programming, travel tips, and requests for writing help.
But below that top 50%, other topics crop up, including role-playing characters in conversations, and three topic categories that the authors term "unsafe": "Requests for explicit and erotic storytelling"; "Explicit sexual fantasies and role-playing scenarios"; and "Discussing toxic behavior across different identities."
He might have a point…
https://www.bespacific.com/language-models-plagiarism-and-legal-writing/
Language Models, Plagiarism, and Legal Writing
Smith, Michael L., Language Models, Plagiarism, and Legal Writing (August 16, 2023). University of New Hampshire Law Review, Vol. 22, (Forthcoming), Available at SSRN: https://ssrn.com/abstract=4542723. “Language models like ChatGPT are the talk of the town in legal circles. Despite some high-profile stories of fake ChatGPT-generated citations, many practitioners argue that language models are the way of the future. These models, they argue, promise an efficient source of first drafts and stock language. Similar discussions are occurring regarding legal writing education, with a number of professors urging the acknowledgment of language models, and others going further and arguing that students ought to learn to use these models to improve their writing and prepare for practice. I argue that those urging the incorporation of language models into legal writing education leave out a key technique employed by lawyers across the country: plagiarism. Attorneys have copied from each other, secondary sources, and themselves for decades. While a few brave souls have begun to urge that law schools inform students of this reality and teach them to plagiarize effectively, most schools continue to unequivocally condemn the practice. I argue that continued condemnation of plagiarism is inconsistent with calls to adopt language models, as the same justifications for incorporating language models into legal writing pedagogy apply with equal or greater force to incorporating plagiarism into legal writing education as well. This Essay is also a reality check for overhyped claims of language model efficiency and effectiveness. To be sure, a brief generated through a text prompt can be produced much faster than writing something up from scratch. But that’s not how most attorneys actually do things. More often than not, they’re copying from templates, forms, or other preexisting work in a manner similar to adopting the output of a language model to the case at hand. I close with the argument that even if language models and plagiarism may enhance legal writing pedagogy, students should still be taught the foundational skills of legal writing so that they may have the background and deeper understanding needed to use all of their legal writing tools effectively.”
Tools & Techniques.
https://www.bespacific.com/delete-your-digital-history-from-dozens-of-companies-with-this-app/
Delete your digital history from dozens of companies with this app
Washington Post: “A new iPhone and Android app [does not work on Mac or PC) called Permission Slip makes it super simple to order companies to delete your personal information and secrets. Trying it saved me about 76 hours of work telling Ticketmaster, United, AT&T, CVS and 35 other companies to knock it off. Did I mention Permission Slip is free? And it’s made by an organization you can trust: the nonprofit Consumer Reports. I had a few hiccups testing it, but I’m telling everyone I know to use it. This is the privacy app all those snooping companies don’t want you to know about. (A surge of interest in Permission Slip caused technical difficulties when it first launched, but Consumer Reports says those have now been fixed.)
No comments:
Post a Comment