Nature (Nature feature article) June 5, 2023 Artificial intelligence chatbots, such as ChatGPT, have impressive abilities. Yet for all their apparent sentience, chatbots are not intelligent — and they must be used with caution. Researchers who have become adept with the tool offer advice for scientists on how to avoid the pitfalls – “treat this AI as a summer intern” — hard-working and eager to please, but also inexperienced and error-prone. In short, ChatGPT and related tools based on large language models (LLMs), which include Microsoft Bing and GitHub Copilot, are incredibly powerful programming aids, but must be used with […]
Category Archives: Artificial Intelligence
China has started a grand experiment in AI education. It could reshape how the world learns.
MIT Technology Review August 2, 2019 According to one estimate, China led the way investing over $1 billion globally last year in AI education. Tech giants, startups, and education incumbents have all jumped in. Tens of millions of students now use some form of AI to learn. Three things have fueled China’s AI education boom. The first is tax breaks and other incentives for AI ventures, academic competition in China is fierce and Chinese entrepreneurs have masses of data at their disposal to train and refine their algorithms. Squirrel, one of the largest AI education companies in China, also opened […]
An AI for generating fake news could also help detect it
MIT Technology Review March 12, 2019 To detect fake news researchers at MIT and Harvard based their experiments on the hypothesis that language models produce sentences by predicting the next word in a sequence of text. So, if they can easily predict most of the words in a given passage, it’s likely it was written by one of their own. They tested this idea by building an interactive tool based on OpenAI’s GPT-2 and fed it both machine and human generated text. The tool generally correctly identified the machine generated and human generated texts. When it was fed text generated […]
Causal disentanglement is the next frontier in AI
Phys.org February 20, 2019 Complex behaviour emerges from interactions between objects produced by different generating mechanisms. Researchers in Sweden introduce a universal, unsupervised and parameter-free model-oriented approach, based on the seminal concept and the first principles of algorithmic probability, to decompose an observation into its most likely algorithmic generative models. They demonstrated its ability to deconvolve interacting mechanisms regardless of whether the resultant objects are bit strings, space–time evolution diagrams, images or networks. Although this is mostly a conceptual contribution and an algorithmic framework, they have provided numerical evidence evaluating the ability of the methods to extract models from data […]
MIT has just announced a $1 billion plan to create a new college for AI
MIT Technology Review October 15, 2018 The new college of computing is being built with $350 million in funding from Stephen A. Schwarzman, the CEO and cofounder of a private equity firm. The school will open next September. Under the recent initiative, the Quest for Intelligence, it aims to make breakthroughs in AI by bringing together researchers from cognitive science and neuroscience as well as computer science. According to the president of MIT the new approach was necessary because of the way computing, data, and AI are “reshaping the world”…read more.
Invention of ionic decision-maker capable of self-learning
Eurekalert October 15, 2018 Researchers in Japan have developed a decision-making ionic device capable of operating using electrochemical phenomena induced by the movement of protons within a solid electrolyte. When the device makes a correct decision, ions migrate toward the electrode associated with the decision. They applied this mechanism to a congested radio communication network and succeeded in demonstrating that the device can select an optimum communication channel to be assigned for a given transmission to achieve the most effective overall channel utilization in relation to changing congestion situations and more complex decisions. The invention may lead to the development […]
AI for cybersecurity is a hot new thing—and a dangerous gamble
MIT Technology Review August 11, 2018 Many firms are now rolling out machine-learning-based products to get an audience with customers who have bought into the AI hype cycle. According to experts many products being rolled out involve supervised learning. The training information they use has not been thoroughly scrubbed of anomalous data points which could lead to the algorithm missing some attacks. Other concerns include difficulty of figuring out why some very complex algorithms spit out certain answers and overreliance on a single, master algorithm to drive a security system. Experts emphasize the importance of monitoring and minimizing the risks… […]
The US military is funding an effort to catch deepfakes and other AI trickery
MIT News May 23, 2018 This summer, under a project funded DARPA, the world’s leading digital forensics experts will gather for an AI fakery contest. They will compete to generate the most convincing AI-generated fake video, imagery, and audio—and they will also try to develop tools that can catch these counterfeits automatically. The contest will include so-called “deepfakes,” videos in which one person’s face is stitched onto another person’s body… read more.
Artificial intelligence helps soldiers learn many times faster in combat
Eurekalert April 27, 2018 Stochastic Gradient Descent (SGD) is widely used for Collaborative Filtering, a well-known machine learning technique for recommender systems. A team of researchers in the US (ARL, University of Southern California) has developed an FPGA-based accelerator, FASTCF, to accelerate the SGD-based CF algorithm consisting of parallel, pipelined processing units which concurrently process distinct user ratings by accessing a shared on-chip buffer. Compared with non-optimized baseline designs, the hierarchical partitioning approach they used results in up to 60x data dependency reduction, 4.2x bank conflict reduction, and 15.4x speedup… read more. TECHNICAL ARTICLE
Image Inpainting for Irregular Holes Using Partial Convolutions
Arxiv April 20, 2018 Existing deep learning-based image inpainting methods often lead to artifacts such as color discrepancy and blurriness. Researchers in the US (NVIDIA Corporation) propose the use of partial convolutions, where the convolution is masked and renormalized to be conditioned on only valid pixels. They include a mechanism to automatically generate an updated mask for the next layer as part of the forward pass. They have demonstrated that their model outperforms other methods for irregular masks. They have shown qualitative and quantitative comparisons with other methods to validate their approach… read more. Open Access TECHNICAL ARTICLE