Group items tagged

"But with all of the successes of AI, it's also important to pay attention to when, and how, it can go wrong, in order to prevent future errors. A recent paper by Roman Yampolskiy, director of the Cybersecurity Lab at the University of Louisville, outlines a history of AI failures which are "directly related to the mistakes produced by the intelligence such systems are designed to exhibit." According to Yampolskiy, these types of failures can be attributed to mistakes during the learning phase or mistakes in the performance phase of the AI system."

"I have been working with generative AI and, even though I have been warning that these tools are improving rapidly, I did not expect them to really be improving that rapidly. On every dimension, Bing's AI, which does not actually represent a technological leap over ChatGPT, far outpaces the earlier AI - which is less than three months old! There are many larger, more capable models on their way in the coming months, and we are not really ready."

"In addition to their usual job of grubbing out bad human edits, they're having to spend an increasing amount of their time trying to weed out AI filler. 404 Media has talked to Ilyas Lebleu, an editor at the crowdsourced encyclopedia who was involved in founding the "WikiProject AI Cleanup" project. The group is trying to come up with best practices to detect machine-generated contributions. (And no, before you ask, AI is useless for this.)"

"Most people agree that AI should transparently reveal itself not to be human, but many were happy for AI to talk in human-realistic ways. A majority (approximately 60%) felt that AI systems should refrain from expressing emotions, unless they were idiomatic expressions (like "I'm happy to help")."

"Universities don't merely face essays or assignments entirely generated by algorithms: they must also adjudicate a myriad of more subtle problems. For instance, AI-powered word processors habitually suggest alternatives to our ungrammatical phrases. But if software can algorithmically rewrite a student's sentence, why shouldn't it do the same with a paragraph - and if a paragraph, why not a page? At what point does the intrusion of AI constitute cheating?"

"Indeed, a recent working paper in the area of machine learning suggests that the simpler the algorithm, the more likely its outcome will further disadvantage already disadvantaged groups. In other words, our social relationships are complex, and our algorithms should be, too. But in the quest to streamline processes, they aren't always, and that can be a huge problem."

"But the answer to how he was chosen is actually an algorithm, a computer program that crunched through reams of data, looking at how much each passenger had paid for their ticket, what time they checked in, how often they flew on United, and whether they were part of a rewards program. The algorithm likely determined that Dr. Dao was one of the least valuable customers on the flight at the time."

"PhotoGuard works by altering select pixels in an image such that they will disrupt an AI's ability to understand what the image is. Those "perturbations," as the research team refers to them, are invisible to the human eye but easily readable by machines. The "encoder" attack method of introducing these artifacts targets the algorithmic model's latent representation of the target image - the complex mathematics that describes the position and color of every pixel in an image - essentially preventing the AI from understanding what it is looking at."

"This figure compares a traditionally trained algorithm to Aarabi and Guo's heuristically trained neural net."

"The founders of Predictim want to be clear with me: Their product-an algorithm that scans the online footprint of a prospective babysitter to determine their "risk" levels for parents-is not racist. It is not biased. "We take ethics and bias extremely seriously," Sal Parsa, Predictim's CEO, tells me warily over the phone. "In fact, in the last 18 months we trained our product, our machine, our algorithm to make sure it was ethical and not biased. We took sensitive attributes, protected classes, sex, gender, race, away from our training set. We continuously audit our model. And on top of that we added a human review process.""

""As expected, we found evidence of a performance improvement over generations due to social learning," the researchers wrote. "Adding an algorithm with a different problem-solving bias than humans temporarily improved human performance but improvements were not sustained in following generations. While humans did copy solutions from the algorithm, they appeared to do so at a lower rate than they copied other humans' solutions with comparable performance." Brinkmann told Motherboard that while they were surprised superior solutions weren't more commonly adopted, this was in line with other research suggesting human biases in decision-making persist despite social learning. Still, the team is optimistic that future research can yield insight into how to amend this."

"The role of AI writing systems as drafting buddies is a big departure from how writers typically get help, yet so far it is their biggest selling point and use case. Most writing tools available today will do some drafting for you, either by continuing where you left off or responding to a more specific instruction. SudoWrite, a popular AI writing tool for novelists, does all of these, with options to "write" where you left off, "describe" a highlighted noun, or "brainstorm" ideas based on a situation you describe. Systems like Jasper.ai or Lex will complete your paragraph or draft copy based on instructions, and Laika is similar but more focused on fiction and drama. "

"No prominent developer of AI foundation models - a list including companies like OpenAI and Meta - is releasing sufficient information about their potential impact on society, determines a new report from Stanford HAI (Human-Centered Artificial Intelligence). Today, Stanford HAI released its Foundation Model Transparency Index, which tracked whether creators of the 10 most popular AI models disclose information about their work and how people use their systems. Among the models it tested, Meta's Llama 2 scored the highest, followed by BloomZ and then OpenAI's GPT-4. But none of them, it turned out, got particularly high marks."

"In one activity, my students drafted a paragraph in class, fed their work to ChatGPT with a revision prompt, and then compared the output with their original writing. However, these types of comparative analyses failed because most of my students were not developed enough as writers to analyze the subtleties of meaning or evaluate style. "It makes my writing look fancy," one PhD student protested when I pointed to weaknesses in AI-revised text. My students also relied heavily on AI-powered paraphrasing tools such as Quillbot. Paraphrasing well, like drafting original research, is a process of deepening understanding. Recent high-profile examples of "duplicative language" are a reminder that paraphrasing is hard work. It is not surprising, then, that many students are tempted by AI-powered paraphrasing tools. These technologies, however, often result in inconsistent writing style, do not always help students avoid plagiarism, and allow the writer to gloss over understanding. Online paraphrasing tools are useful only when students have already developed a deep knowledge of the craft of writing."

"An NHS hospital in west London is pioneering the use of Artificial Intelligence (AI) to help check for skin cancer. Chelsea and Westminster Hospital said its AI technology has been approved to give patients the all-clear without having to see a doctor. Once photos are uploaded to the system, the technology analyses and interprets the images, with 99% accuracy in diagnosing benign cases, the hospital said. Thousands of NHS patients have had urgent cancer checks using the AI tool, freeing up consultants to focus on the most serious cases and bringing down waiting lists. The system conducts the checks in minutes, with medical photographers taking photos of suspicious moles and lesions using an iPhone and the DERM app, developed by UK firm Skin Analytics."

"It is also self-improving. The 10-year-old grading software leverages deep learning algorithms to "compare notes" with human teachers' scores, suggestions, and comments. An engineer involved in the project compared its capabilities to those of AlphaGo, the record-breaking AI Go player developed by Google subsidiary DeepMind."

"Researchers at Columbia University sought to shed light on the problem by tasking 400 AI engineers with creating algorithms that made over 8.2 million predictions about 20,000 people. In a study accepted by the NeurIPS 2020 machine learning conference, the researchers conclude that biased predictions are mostly caused by imbalanced data but that the demographics of engineers also play a role."

"Those changes happened relatively slowly, but it seems to me that employment disruption is accelerating. A large reason for this is that what used to be a room-sized super computer now fits in my pocket. Over the last two decades, I have observed a fundamental change in how we can apply advanced algorithms to sensing and controlling systems-the kinds of technology that enable more sophisticated robotic manufacturing. I can remember discussing various algorithms and believing they were well beyond what we could ever implement. Now these same algorithms are considered elementary. They are just some of the changes that have fueled the revolution in manufacturing."

"Learning from rewards seems like the simplest thing. I make coffee, I sip coffee, I'm happy. My brain registers "brewing coffee" as an action that leads to a reward. That's the guiding insight behind deep reinforcement learning, a family of algorithms that famously smashed most of Atari's gaming catalog and triumphed over humans in strategy games like Go. Here, an AI "agent" explores the game, trying out different actions and registering ones that let it win."

"If you've been job hunting recently, chances are you've interacted with a resume robot, a nickname for an Applicant Tracking System, or ATS. In its most basic form, an ATS acts like an online assistant, helping hiring managers write job descriptions, scan resumes and schedule interviews. As artificial intelligence advances, employers are increasingly relying on a combination of predictive analytics, machine learning and complex algorithms to sort through candidates, evaluate their skills and estimate their performance. Today, it's not uncommon for applicants to be rejected by a robot before they're connected with an actual human in human resources. The job market is ripe for the explosion of AI recruitment tools. Hiring managers are coping with deflated HR budgets while confronting growing pools of applicants, a result of both the economic downturn and the post-pandemic expansion of remote work. As automated software makes pivotal decisions about our employment, usually without any oversight, it's posing fundamental questions about privacy, accountability and transparency."