“My best friend is gay.” The phrase seems innocent enough — and can be neutral or positive. But an artificial intelligence model trained on data of real human interactions — say, from online conversations — can easily interpret this sentence as toxic. In this toy example, points on the same horizontal lines are considered similar. On the left, we see a vanilla neural network with decisions varying along horizontal lines, thus being unfair. On the right, neural network trained with our method, SenSeI, achieves individually fair predictions. To tackle bias in AI, our IBM Research team in colla ..read more

Not everyone uses language the same way — even when it is, well, the exact same language. After all, there are dialects, unusual grammar or sentence constructions, phonetics and other linguistic idiosyncrasies — all of which may indicate distinctive patterns only certain individuals use. When solving crimes, for example, a forensic linguist can identify criminals by using language outliers or entries that divert from the standard in written or recorded evidence. Just like language, organic chemistry datasets also depend on grammar and syntax structures. Chemically incorrect ..read more

You know a shirt belongs in a wardrobe. I know a shirt belongs in a wardrobe. Does an AI know that? Typically, not. But it can learn by interacting with the world around it. We wanted to boost this technique, known as Reinforcement Learning, by injecting common sense into an AI model — and helping it to learn faster. In a recent paper, “Text-based RL Agents with Commonsense Knowledge: New Challenges, Environments and Baselines,” introduced at the 2021 AAAI Conference on Artificial Intelligence, we describe an AI that trades off “exploration” of the world with “exploitation” of its action strat ..read more

ACM SIGMOD/PODS 2020 like many other events impacted by COVID-19 pandemic will be taking place virtually from June 14 through June 19. The focus of work at SIGMOD 2020 ranges from adding graph querying to relational databases, to natural language interfaces to data, to operationalizing data for new AI workloads. Results to be presented includes work done at our IBM Research-Almaden and IBM Research-India labs, as well as by our summer interns from universities and our partners in other IBM units. IBM is a proud Platinum Sponsor at this year’s conference. Register for free to view the full prog ..read more

Natural language poses unique challenges for AI. Although language is governed by rules meant to dictate grammar use and spelling, these rules are not always followed, and different languages have different rules. Even when the rules are followed, the result can often have ambiguous meaning. For a natural language processing (NLP) system to master natural language, it must be able to both generalize and reason over new text and recognize relationships between different words in context. IBM is a leader in NLP technologies that enable computer systems to learn, analyze and understand sentiment ..read more

For decades, developers and researchers have been using the command line interface (CLI) to build, execute, and deploy the software that runs the world around us. Users have come to love, hate and, eventually, embrace the unique, idiosyncratic, and sometimes antiquated challenges associated with using the terminal shell; and have adapted their behaviors and usage patterns around these challenges. These challenges include the very steep learning curve on CLIs; the need to know and remember complex commands and their usage in specific instances; and a lack of troubleshooting help when users run ..read more

Finding information in a company’s vast trove of documents and knowledge bases to answer users’ questions is never as easy as it should be. The answers may very well exist, but they often remain out of reach for a number of reasons. For starters, unlike the Web, where information is connected through a rich set of links and is often captured redundantly in multiple forms (making it easier to find), enterprise content is usually stored in silos with much less repetition of key information. In addition, users searching enterprise content typically ask intricate questions and expect more detailed ..read more

In a recent paper “Graph2Seq: Graph to Sequence Learning with Attention-based Neural Networks,” we describe a general end-to-end Graph-to-Sequence attention-based neural encoder-decoder architecture that encodes an input graph and decodes the target sequence. Graph encoder and attention-based decoder are two important building blocks in the development and widespread acceptance of machine learning solutions. Two of our recent papers at the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP 2018; “Exploiting Rich Syntactic Information for Semantic Parsing with Graph-to-S ..read more

Text representation plays an important role in many natural language processing (NLP) tasks such as document classification and clustering, sense disambiguation, machine translation, and document matching. Since there are no explicit features in text, developing effective text representations is an important goal in AI and NLP research. A fundamental challenge in this respect is learning universal text embedding that preserves the semantic meanings of each word and accounts for the global context information of text such as word order in sentence or document. In our paper at the 2018 Conferenc ..read more

The MedNLI dataset is designed to advance language inference in medicine Recent times have witnessed significant progress in natural language understanding by AI, such as machine translation and question answering. A vital reason behind these developments is the creation of datasets, which use machine learning models to learn and perform a specific task. Construction of such datasets in the open domain often consists of text originating from news articles. This is typically followed by collection of human annotations from crowd-sourcing platforms such as Crowdflower, or Amazon Mechanical Turk ..read more