 |
|
|
|||||
Youve just finished a strenuous hike to the top of a mountain. Youre exhausted but elated. The view of the city below is gorgeous, and you want to capture the moment on camera. But its already quite dark, and youre not sure youll get a good shot. Fortunately, your phone has an AI-powered night mode that can take stunning photos even after sunset. Heres something you might not know: That night mode may have been trained on synthetic nighttime images, computer-generated scenes that were never actually photographed. As artificial intelligence researchers exhaust the supply of real data on the web and in digitized archives, they are increasingly turning to synthetic data, artificially generated examples that mimic real ones. But that creates a paradox. In science, making up data is a cardinal sin. Fake data and misinformation are already undermining trust in information online. So how can synthetic data possibly be good? Is it just a polite euphemism for deception? As a machine learning researcher, I think the answer lies in intent and transparency. Synthetic data is generally not created to manipulate results or mislead people. In fact, ethics may require AI companies to use synthetic data: Releasing real human face images, for example, can violate privacy, whereas synthetic faces can offer similar benefit with formal privacy guarantees. There are other reasons that help explain the growing use of synthetic data in training AI models. Some things are so scarce or rare that they are barely represented in real data. Rather than letting these gaps become an Achilles heel, researchers can simulate those situations instead. Another motivation is that collecting real data can be costly or even risky. Imagine collecting data for a self-driving car during storms or on unpaved roads. It is often much more efficient, and far safer, to generate such data virtually. Heres a quick take on what synthetic data is and why researchers and developers use it. How synthetic data is made Training an AI model requires large amounts of data. Like students and athletes, the more an AI is trained, the better its performance tends to be. Researchers have known for a long time that if data is in short supply, they can use a technique known as data augmentation. For example, a given image can be rotated or scaled to yield additional training data. Synthetic data is data augmentation on steroids. Instead of making small alterations to existing images, researchers create entirely new ones. But how do researchers create synthetic data? There are two main approaches. The first approach relies on rule-based or physics-based models. For example, the laws of optics can be used to simulate how a scene would appear given the positions and orientations of objects within it. The second approach uses generative AI to produce data. Modern generative models are trained on vast amounts of data and can now create remarkably realistic text, audio, images, and videos. Generative AI offers a flexible way to produce large and diverse datasets. Both approaches share a common principle: If data does not come directly from the real world, it must come from a realistic model of the world. Downsides and dangers It is also important to remember that while synthetic data can be useful, it is not a panacea. Synthetic data is only as reliable as the models of reality it comes from, and even the best scientific or generative models have weaknesses. Researchers have to be careful about potential biases and inaccuracies in the data they produce. For example, researchers may simulate the home-insurance ecosystem to help detect fraud, but those simulations could embed unfair assumptions about neighborhoods or property types. The benefits of such data must be weighed against risks to fairness and equity. Its also important to maintain a clear distinction between models and simulations on one hand and the real world on the other. Synthetic data is invaluable for training and testing AI systems, but when an AI model is deployed in the real world, its performance and safety should be proved with real, not simulated, data for both technical and ethical reasons. Future research on synthetic data in AI is likely to face many challenges. Some are ethical, some are scientific, and others are engineering problems. As synthetic data becomes more realistic, it will be more useful for training AI, but it will also be easier to misuse. For example, increasingly realistic synthetic images can be used to create convincing deepfake videos. I believe that researchers and AI companies should keep clear records to show which data is synthetic and why it was created. Clearly disclosing which parts of the training data are real and which are synthetic is a key aspect of responsibly producing AI models. Californias law, Generative artificial intelligence: training data transparency, set to take effect on January 1, 2026, requires AI developers to disclose if they used synthetic data in training their models. Researchers should also study how mistakes in simulations or models can lead to bad data. Careful work will help keep synthetic data transparent, trustworthy, and reliable. Keeping it real Most AI systems learn by finding patterns in data. Researchers can improve their ability to do this by adding synthtic data. But AI has no sense of what is real or true. The desire to stay in touch with reality and to seek truth belongs to people, not machines. Human judgment and oversight in the use of synthetic data will remain essential for the future. The next time you use a cool AI feature on your smartphone, think about whether synthetic data might have played a role. Our AIs may learn from synthetic data, but reality remains the ultimate source of our knowledge and the final judge of our creations. Ambuj Tewari is a professor of statistics at the University of Michigan. This article is republished from The Conversation under a Creative Commons license. Read the original article.
Category:
E-Commerce
President Trump recently promised to make America the “crypto capital of the world.” And his administration is working hard to make that pledge a reality. White House officials have established a working group on digital asset markets and directed federal agencies to craft a strategy to cement U.S. leadership. The president’s legislative team, meanwhile, helped push the GENIUS Act (Guiding and Establishing National Innovation for U.S. Stablecoins Act),through Congress earlier this summer, thus creating the first federal framework for stablecoins. And they’re working to pass the Clarity Act (Digital Asset Market Clarity Act), which would finally settle disputes over which regulator oversees digital assets. It’s refreshing to see our political leaders working to bring digital assets into the financial mainstream, especially after years of hostility from the prior administration. But the work is far from finishedand achieving universal legitimacy will require not just favorable laws and regulations, but also behavioral changes at leading crypto firms. Conflicting guidance For more than a decade, crypto innovators faced a patchwork of state regimes and conflicting federal guidance. The lack of clear regulation led to a proliferation of scams and bad actorsand kept many investors on the sidelines. Big banks and other legacy financial institutions hesitated to adopt cryptocurrencies and the underlying blockchain technology they’re based on, even as top financiers acknowledged blockchain’s potential to reshape the entire industry. The GENIUS Act represents Washington’s first serious attempt to genuinely regulaterather than ignore or suppressone of the leading forms of cryptocurrency. The new law requires stablecoin issuers to maintain dollar-for-dollar reserves and submit to audits. Far from rejecting this level of regulation, crypto leaders practically begged for it. They recognized that federal oversight and transparent standards are needed to transform what the public previously viewed as a speculative product into a reliable payment instrument. That’s why industry leaders are also working with the White House and Congress to finalize the Clarity Act, which would define the boundaries of authority between the Securities and Exchange Commission and the Commodity Futures Trading Commission, delivering the kind of predictability that underpins every functioning capital market. Cultural shift But better regulation alone won’t bring about the mainstream approval that industry leaders seek. Only an internal cultural shiftand rigorous self-policingcan deliver that. Every blockchain transaction depends on various forms of intellectual propertyfrom patents on mobile crypto wallets and bitcoin mining data centers to trade secrets in proprietary trading algorithms, and copyrights protecting exchange software to trademarks that build consumer trust. Coinbase, for instance, holds nearly 200 active patents. But most of the intellectual property powering today’s blockchain activity belongs to third parties outside the crypto industry. Yet even as leading platforms generate billions in revenue, the industry remains reluctant to acknowledge the legitimacy of IP rights. This reluctance is playing out in court. In May, Bancor’s nonprofit arm sued Uniswap, alleging that the leading decentralized exchange built its multibillion-dollar business on Bancor’s patented automated market maker technology without authorization. And earlier this summer, Malikie Innovations filed suits against Core Scientific and Marathon Digital, claiming their bitcoin mining operations infringe on Malikie’s patents for elliptic curve cryptography. Elliptic Curve Cryptography (ECC), a cryptographic technique developed and patented by Certicom years before crypto went mainstream, was licensed by companies like Cisco and Motorola as well as the National Security Agency. Cases like these highlight the tension: Crypto companies depend on IP to function, but too many are willing to disregard the IP rights of others, even as they clamor for legitimacy. Not how respectable companies operate This simply isn’t the way respectable companies in mature industries operate. Spotify and Apple Music wouldn’t enjoy their positive reputations if they refused to pay royalties to artists and record labels. Streaming platforms like Netflix and Hulu would be pariahs if they pirated films. Banks would be shunned by investors alike if they treated software licenses as optional. If leading crypto firms want to be seen as respectable, investable pillars of the global economy, they need to meet those same standards when it comes to intellectual property. Digital assets are here to stay. But universal legitimacy will come only from a combination of comprehensive regulation and a cultural shift within the industry itself.
Category:
E-Commerce
If you slip a tiny wearable device on your fingertip and slide it over a smooth surface like a touchscreen, you can feel digital textures like denim or mesh. The device, designed by researchers at Northwestern University, is the first of its kind to achieve human resolution, meaning that it can more accurately match the complex way a human fingertip senses the world. In previous attempts at haptic devices like this, once you compare them to real textures, you realize theres something still missing, says Sylvia Tan, a PhD student at Northwestern and one of the authors of a new study in Science Advances about the research. Its close, but not quite there. Our work is trying to just get that one step closer. [Photo: Northwestern University] The wearable, made from flexible, paper-thin latex, is embedded with tiny nodes that push into the skin in a precise way and can move up to 800 times per second. Past devices had low resolutionthe touch equivalent of a pixelated image or an early movie from the 1890s with so few frames that the movement looks jerky. Using nodes and arranging them in a particular density improves that resolution. [Photo: Northwestern University] Earlier devices were also bulky. The ultrathin new technology, which weighs less than a gram, is designed to be comfortable to wear. A big goal was to make it very lightweight so you arent distracted by it, Tan says. And [to make] something that we call ‘haptically transparent’that means that even when youre wearing it, you can still perceive the real world, so you can perform everyday tasks. [Photo: Northwestern University] In the study, users could identify fabrics like corduroy or leather with 81% accuracy. The technology is still in development, but in the future, it could make it possible to feel products as you shop online. It could also have more immediate uses for people who are visually impaired, like making it possible to feel a tactile map or translating text on a screen to braille without a large, expensive device. On devices like microwaves, where physical buttons have often been replaced by flat touchscreens, the wearable could help a visually impaired person know where to push. It could also help improve human-robot interfaces. “In the medical field, the Da Vinci robot has very good kinesthetic force feedback,” Tan says. “But getting a surgeon to feel exactly what’s happening at your fingertip as you move the angle of your finger is not quite there. And that’s very important for high-skill workers.”
Category:
E-Commerce
All news |
||||||||||||||||||
|
||||||||||||||||||