A novel machine learning model called Temporal Autoencoders for Causal Inference (TACI) accurately detects changing cause-and-effect relationships in complex, time-varying systems like weather patterns and brain activity. By analyzing both synthetic and real data, TACI captures dynamic interactions and quantifies shifts in strength or direction over time. Tested on long-term weather data and brain imaging in monkeys, TACI successfully pinpointed when causal connections emerged, weakened, or reversed.

For the first time, a robot has been trained to perform surgical procedures by watching videos of expert surgeons, marking a leap forward in robotic surgery. This breakthrough in "imitation learning" means that robots can learn complex tasks without needing to be programmed for every individual movement. By training on surgical footage, the robot replicated procedures with skill comparable to human surgeons, demonstrating its ability to adapt and even correct its actions autonomously.

During my first semester as a computer science graduate student at Princeton, I took COS 402: Artificial Intelligence. Toward the end of the semester, there was a lecture about neural networks. This was in the fall of 2008, and I got the distinct impression—both from that lecture and the textbook—that neural networks had become a backwater.

Neural networks had delivered some impressive results in the late 1980s and early 1990s. But then progress stalled. By 2008, many researchers had moved on to mathematically elegant approaches such as support vector machines.

I didn’t know it at the time, but a team at Princeton—in the same computer science building where I was attending lectures—was working on a project that would upend the conventional wisdom and demonstrate the power of neural networks. That team, led by Prof. Fei-Fei Li, wasn’t working on a better version of neural networks. They were hardly thinking about neural networks at all.

Read full article

Comments

© Aurich Lawson | Getty Images

A new AI tool identified long COVID in 22.8% of patients, a much higher rate than previously diagnosed. By analyzing extensive health records from nearly 300,000 patients, the algorithm identifies long COVID by distinguishing symptoms linked specifically to SARS-CoV-2 infection rather than pre-existing conditions. This AI approach, known as "precision phenotyping," helps clinicians differentiate long COVID symptoms from other health issues and may improve diagnostic accuracy by about 3%.

A recent study questions if large language models (LLMs) truly form coherent world models, despite their accurate outputs in complex tasks like generating directions or playing games. Researchers found that while LLMs provide nearly flawless driving directions, they fail with unexpected changes, suggesting the models don't grasp underlying rules.

A new study reveals that while human-AI collaboration can be powerful, it depends on the task. Analysis of hundreds of studies found that AI outperformed human-AI teams in decision-making tasks, while collaborative teams excelled in creative tasks like content generation. This research suggests organizations may overestimate the benefits of human-AI synergy. Instead, strategic use of AI’s strengths in data processing and humans’ creativity may yield the best results.

Researchers have developed an AI system that predicts chemical compounds capable of targeting two proteins simultaneously, potentially creating more effective medications. By training the AI with a chemical language model, it was able to generate novel molecular structures with dual-target activity, an essential feature for treating complex diseases like cancer.

A new study introduces "System 0," a cognitive framework where artificial intelligence (AI) enhances human thinking by processing vast data, complementing our natural intuition (System 1) and analytical thinking (System 2). However, this external thinking system poses risks, such as over-reliance on AI and a potential loss of cognitive autonomy.

New research shows that stories generated by AI, such as ChatGPT, are almost as good as those written by humans. However, when people are told a story is AI-generated, they rate it more negatively, revealing a bias against AI-created content.

Researchers developed an AI algorithm, ARC-SV, to detect complex structural variants in the human genome that previous methods missed. Analyzing over 4,000 genomes, researchers discovered thousands of complex variants, many affecting brain-related genes and linked to schizophrenia and bipolar disorder.

People empathize with AI bots excluded from a virtual game, treating them like social beings in need of fairness. Participants favored giving the AI bot a fair chance in play, with older adults showing a stronger inclination to rectify the perceived unfairness.

A new study reveals a vulnerability in AI image recognition systems due to their exclusion of the alpha channel, which controls image transparency. Researchers developed "AlphaDog," an attack method that manipulates transparency in images, allowing hackers to distort visuals like road signs or medical scans in ways undetectable by AI. Tested across 100 AI models, AlphaDog exploits this transparency flaw, posing significant risks to road safety and healthcare diagnostics.

Researchers have developed a machine learning model that upgrades 3T MRI images to mimic the higher-resolution 7T MRI, providing enhanced detail for detecting brain abnormalities. The synthetic 7T images reveal finer features, such as white matter lesions and subcortical microbleeds, which are often difficult to see with standard MRI systems. This AI-driven approach could improve diagnostic accuracy for conditions like traumatic brain injury (TBI) and multiple sclerosis (MS), though clinical validation is needed before wider use.

A recent review explored how integrating machine learning with traditional statistical models can enhance disease risk prediction accuracy, a key tool in clinical decision-making. While traditional models like logistic regression are limited by certain assumptions, machine learning offers flexibility but has inconsistent results in some cases. The study revealed that combined models, especially stacking methods, outperform individual methods by harnessing each approach’s strengths and addressing their weaknesses.

Emotion recognition extends far beyond facial expressions, involving a rich interplay of context, physical attributes, and background knowledge. Researchers propose that recognizing emotion is part of forming an overall impression of a person, shaped by cues like clothing, perceived social roles, and personal history. For instance, a facial expression of fear might be interpreted as anger if background context suggests it.

Researchers tested artificial intelligence’s ability to solve abstract visual puzzles similar to human IQ tests, revealing gaps in AI’s reasoning skills. Open-source AI models struggled, while closed-source models like GPT-4V performed better, but far from perfectly.

Researchers have developed an AI-powered "electronic tongue" capable of distinguishing subtle differences in liquids, such as milk freshness, soda types, and coffee blends. By analyzing sensor data through a neural network, the device achieved over 95% accuracy in identifying liquid quality, authenticity, and potential safety issues. Interestingly, when the AI was allowed to select its own analysis parameters, it outperformed human-defined settings, showing how it holistically assessed subtle data.

A novel study study using AI has uncovered 161,979 new RNA viruses, significantly expanding our understanding of Earth's viral diversity. These discoveries were made by analyzing genetic data using a machine learning model, which identified previously unrecognized viruses hidden in public databases.

The 2024 Nobel Prize in Chemistry celebrates two groundbreaking achievements in protein science: designing novel proteins and predicting protein structures using AI. David Baker has pioneered techniques to construct entirely new proteins that could serve as pharmaceuticals, vaccines, and nanomaterials. Meanwhile, Demis Hassabis and John Jumper’s AI model, AlphaFold2, solved the decades-old challenge of predicting protein structures, allowing researchers to visualize nearly all known proteins. These advancements hold transformative potential for fields like medicine, biotechnology, and environmental science.