Retrieval-augmented generation (RAG) utilizes retrieved texts to enhance large language models (LLMs). However, studies show that RAG is not consistently effective and can even mislead LLMs due to noisy or incorrect retrieved texts. This suggests that RAG possesses a duality including both benefit and detriment. Although many existing methods attempt to address this issue, they lack a theoretical explanation for the duality in RAG. The benefit and detriment within this duality remain a black box that cannot be quantified or compared in an explainable manner. This paper takes the first step in theoretically giving the essential explanation of benefit and detriment in RAG by: (1) decoupling and formalizing them from RAG prediction, (2) approximating the gap between their values by representation similarity and (3) establishing the trade-off mechanism between them, to make them explainable, quantifiable, and comparable. We demonstrate that the distribution difference between retrieved texts and LLMs' knowledge acts as double-edged sword, bringing both benefit and detriment. We also prove that the actual effect of RAG can be predicted at token level. Based on our theory, we propose a practical novel method, X-RAG, which achieves collaborative generation between pure LLM and RAG at token level to preserve benefit and avoid detriment. Experiments in real-world tasks based on LLMs including OPT, LLaMA-2, and Mistral show the effectiveness of our method and support our theoretical results.

Current models on Explainable Artificial Intelligence (XAI) have shown an evident and quantified lack of reliability for measuring feature-relevance when statistically entangled features are proposed for training deep classifiers. There has been an increase in the application of Deep Learning in clinical trials to predict early diagnosis of neuro-developmental disorders, such as Autism Spectrum Disorder (ASD). However, the inclusion of more reliable saliency-maps to obtain more trustworthy and interpretable metrics using neural activity features is still insufficiently mature for practical applications in diagnostics or clinical trials. Moreover, in ASD research the inclusion of deep classifiers that use neural measures to predict viewed facial emotions is relatively unexplored. Therefore, in this study we propose the evaluation of a Convolutional Neural Network (CNN) for electroencephalography (EEG)-based facial emotion recognition decoding complemented with a novel RemOve-And-Retrain (ROAR) methodology to recover highly relevant features used in the classifier. Specifically, we compare well-known relevance maps such as Layer-Wise Relevance Propagation (LRP), PatternNet, Pattern-Attribution, and Smooth-Grad Squared. This study is the first to consolidate a more transparent feature-relevance calculation for a successful EEG-based facial emotion recognition using a within-subject-trained CNN in typically-developed and ASD individuals.

This month, we take a look at all the things we are thankful for in the AI community. Joining the discussion this time are: Tom Dietterich (Oregon State University), Sabine Hauert (University of Bristol), Holger Hoos (Leiden University), Sarit Kraus (Bar-Ilan University), Michael Littman (Brown University) and Carles Sierra (Artificial Intelligence Research Institute of the Spanish National Research Council). Holger Hoos: I think one can be really grateful that progress in AI has come at a point where we really need it. I think we've maneuvered ourselves as humankind into a situation where the limitations of our own natural intelligence make it very likely that we're going to drive ourselves against the wall. Issues such as climate change are simply too complex for us to figure out, even if you bring lots of smart people together and give them lots of resources.

There's no question artificial intelligence has brought many benefits to both people and businesses. From helping consumers shop and answer questions online to boosting productivity and customer service to treating illness and tracking the spread of COVID-19, AI has been a tremendous boon to society in many ways. Completely dependent on the humans who build and train it, AI can be infused with the biases of its creators. And other negative consequences can come with the increasing dominance of AI that aren't obvious to the average layperson. While AI is an incredibly useful tool, it's important for governments, businesses and the public at large to be aware of what can happen if AI is used indiscriminately or without diligent human oversight.

According to an article published in the AMA Journal of Ethics, recent advances in artificial intelligence (AI) technology should result in an overhaul of medical school curricula to incorporate the effective use of AI, communication, and empathy. The article authors cited the recent focus on the deteriorating mental health of medical students, highlighting the demanding learning environment that contributes to learners' poor mental health. The current information overload crisis has resulted in the need for physicians to manage and use AI applications that aggregate data collaboratively. Based on this, the researchers recommend that medical education be reformed to focus on knowledge management, including effective use of AI technology. In the near future, the skills required of practicing physicians will involve collaborating with AI applications that manage data.

If they haven't done so already, cyber attackers may soon be arming themselves with artificial intelligence and machine learning (ML) strategies and algorithms. Before long, it may not be a fair fight if defenders remain naive to what AI and ML can do on both sides of the battle. So suggests a new report by IEEE and the Canadian tech consulting firm Syntegrity. The report -- stemming from a three-day intensive meeting last October of cybersecurity experts from government, the military, and industry -- aggregates the group's findings into what it calls the six "dimensions" at the intersection of AI, ML, and cybersecurity. First, the report advocates ways to keep cybersecurity regulations and laws up to speed with the latest developments in the field.

I'm aboard the Nautilus, a derelict vessel in orbit around Jupiter's frozen moon of Europa, and the only thing I've come in contact with so far is the ship's artificial intelligence. It even has a friendly name--Kaizen, which translated into English means "Change for better." Kaizen's waiting for me to respond, there in the friendly blink of a command prompt on a terminal straight out of the 1970s. But I've seen plenty of science fiction films, and so I still hold my breath as I ask Kaizen to open the airlock doors, prepared for an "I can't do that, Dave," and the slow strains of "The Blue Danube" to accompany my floating corpse into the blackness of space. Kubrick's classic science fiction film is a clear inspiration, seen in everything from the pseudo-'70s retrofuturism of the Nautilus to its all-seeing security cameras--black, with a cherry dot in the center.

Data mining allows the exploration of sequences of phenomena, whereas one usually tends to focus on isolated phenomena or on the relation between two phenomena. It offers invaluable tools for theoretical analyses and exploration of the structure of sentences, texts, dialogues, and speech. We report here the results of an attempt at using it for inspecting sequences of verbs from French accounts of road accidents. This analysis comes from an original approach of unsupervised training allowing the discovery of the structure of sequential data. The entries of the analyzer were only made of the verbs appearing in the sentences. It provided a classification of the links between two successive verbs into four distinct clusters, allowing thus text segmentation. We give here an interpretation of these clusters by comparing the statistical distribution of independent semantic annotations.