A Safe Collaborative Chatbot for Smart Home Assistants.

Smart home assistants, which enable users to control home appliances and can be used for holding entertaining conversations, have become an inseparable part of many people's homes. Recently, there have been many attempts to allow end-users to teach a home assistant new commands, responses, and rules, which can then be shared with a larger community. However, allowing end-users to teach an agent new responses, which are shared with a large community, opens the gate to malicious users, who can teach the agent inappropriate responses in order to promote their own business, products, or political views. In this paper, we present a platform that enables users to collaboratively teach a smart home assistant (or chatbot) responses using natural language. We present a method of collectively detecting malicious users and using the commands taught by the malicious users to further mitigate activity of future malicious users. We ran an experiment with 192 subjects and show the effectiveness of our platform.

Conversational Agents: Goals, Technologies, Vision and Challenges.

In recent years, conversational agents (CAs) have become ubiquitous and are a presence in our daily routines. It seems that the technology has finally ripened to advance the use of CAs in various domains, including commercial, healthcare, educational, political, industrial, and personal domains. In this study, the main areas in which CAs are successful are described along with the main technologies that enable the creation of CAs. Capable of conducting ongoing communication with humans, CAs are encountered in natural-language processing, deep learning, and technologies that integrate emotional aspects. The technologies used for the evaluation of CAs and publicly available datasets are outlined. In addition, several areas for future research are identified to address moral and security issues, given the current state of CA-related technological developments. The uniqueness of our review is that an overview of the concepts and building blocks of CAs is provided, and CAs are categorized according to their abilities and main application domains. In addition, the primary tools and datasets that may be useful for the development and evaluation of CAs of different categories are described. Finally, some thoughts and directions for future research are provided, and domains that may benefit from conversational agents are introduced.

Vision-Less Sensing for Autonomous Micro-Drones.

This work presents a concept of intelligent vision-less micro-drones, which are motivated by flying animals such as insects, birds, and bats. The presented micro-drone (named BAT: Blind Autonomous Tiny-drone) can perform bio-inspired complex tasks without the use of cameras. The BAT uses LIDARs and self-emitted optical-flow in order to perform obstacle avoiding and maze-solving. The controlling algorithms were implemented on an onboard micro-controller, allowing the BAT to be fully autonomous. We further present a method for using the information collected by the drone to generate a detailed mapping of the environment. A complete model of the BAT was implemented and tested using several scenarios both in simulation and field experiments, in which it was able to explore and map complex building autonomously even in total darkness.

Reactive stroma and trastuzumab resistance in HER2-positive early breast cancer.

We investigated the value of reactive stroma as a predictor for trastuzumab resistance in patients with early HER2-positive breast cancer receiving adjuvant therapy. The pathological reactive stroma and the mRNA gene signatures that reflect reactive stroma in 209 HER2-positive breast cancer samples from the FinHer adjuvant trial were evaluated. Levels of stromal gene signatures were determined as a continuous parameter, and pathological reactive stromal findings were defined as stromal predominant breast cancer (SPBC; ≥50% stromal) and correlated with distant disease-free survival. Gene signatures associated with reactive stroma in HER2-positive early breast cancer (N = 209) were significantly associated with trastuzumab resistance in estrogen receptor (ER)-negative tumors (hazard ratio [HR] = 1.27 p interaction = 0.014 [DCN], HR = 1.58, p interaction = 0.027 [PLAU], HR = 1.71, p interaction = 0.019 [HER2STROMA, novel HER2 stromal signature]), but not in ER-positive tumors (HR = 0.73 p interaction = 0.47 [DCN], HR = 0.71, p interaction = 0.73 [PLAU], HR = 0.84; p interaction = 0.36 [HER2STROMA]). Pathological evaluation of HER2-positive/ER-negative tumors suggested an association between SPBC and trastuzumab resistance. Reactive stroma did not correlate with tumor-infiltrating lymphocytes (TILs), and the expected benefit from trastuzumab in patients with high levels of TILs was pronounced only in tumors with low stromal reactivity (SPBC <50%). In conclusion, reactive stroma in HER2-positive/ER-negative early breast cancer tumors may predict resistance to adjuvant trastuzumab therapy.

SAIF: A Correction-Detection Deep-Learning Architecture for Personal Assistants.

Intelligent agents that can interact with users using natural language are becoming increasingly common. Sometimes an intelligent agent may not correctly understand a user command or may not perform it properly. In such cases, the user might try a second time by giving the agent another, slightly different command. Giving an agent the ability to detect such user corrections might help it fix its own mistakes and avoid making them in the future. In this work, we consider the problem of automatically detecting user corrections using deep learning. We develop a multimodal architecture called SAIF, which detects such user corrections, taking as inputs the user's voice commands as well as their transcripts. Voice inputs allow SAIF to take advantage of sound cues, such as tone, speed, and word emphasis. In addition to sound cues, our model uses transcripts to determine whether a command is a correction to the previous command. Our model also obtains internal input from the agent, indicating whether the previous command was executed successfully or not. Finally, we release a unique dataset in which users interacted with an intelligent agent assistant, by giving it commands. This dataset includes labels on pairs of consecutive commands, which indicate whether the latter command is in fact a correction of the former command. We show that SAIF outperforms current state-of-the-art methods on this dataset.

Does small talk with a medical provider affect ChatGPT's medical counsel? Performance of ChatGPT on USMLE with and without distractions.

Efforts are being made to improve the time effectiveness of healthcare providers. Artificial intelligence tools can help transcript and summarize physician-patient encounters and produce medical notes and medical recommendations. However, in addition to medical information, discussion between healthcare and patients includes small talk and other information irrelevant to medical concerns. As Large Language Models (LLMs) are predictive models building their response based on the words in the prompts, there is a risk that small talk and irrelevant information may alter the response and the suggestion given. Therefore, this study aims to investigate the impact of medical data mixed with small talk on the accuracy of medical advice provided by ChatGPT. USMLE step 3 questions were used as a model for relevant medical data. We use both multiple-choice and open-ended questions. First, we gathered small talk sentences from human participants using the Mechanical Turk platform. Second, both sets of USLME questions were arranged in a pattern where each sentence from the original questions was followed by a small talk sentence. ChatGPT 3.5 and 4 were asked to answer both sets of questions with and without the small talk sentences. Finally, a board-certified physician analyzed the answers by ChatGPT and compared them to the formal correct answer. The analysis results demonstrate that the ability of ChatGPT-3.5 to answer correctly was impaired when small talk was added to medical data (66.8% vs. 56.6%; p = 0.025). Specifically, for multiple-choice questions (72.1% vs. 68.9%; p = 0.67) and for the open questions (61.5% vs. 44.3%; p = 0.01), respectively. In contrast, small talk phrases did not impair ChatGPT-4 ability in both types of questions (83.6% and 66.2%, respectively). According to these results, ChatGPT-4 seems more accurate than the earlier 3.5 version, and it appears that small talk does not impair its capability to provide medical recommendations. Our results are an important first step in understanding the potential and limitations of utilizing ChatGPT and other LLMs for physician-patient interactions, which include casual conversations.