China
Africa
Explore chapters and articles related to this topic
Technical Writing Basics
Published in Phillip A. Laplante, Technical Writing, 2018
Testing for correctness in any kind of writing is difficult. You can use grammar and spell checkers, but they are not 100% accurate, and they cannot check technical correctness or the correctness of ideas. No spell checker will flag “200 kilometers” as incorrect because there is no spelling error, only a logic error. Review by one or more persons, or via group reviews, can increase the correctness of any writing. I discuss review and revision further in the next chapter.
The role of domain expertise in trusting and following explainable AI decision support systems
Published in Journal of Decision Systems, 2022
Sarah Bayer, Henner Gimpel, Moritz Markgraf
When it comes to making a decision, AI-based DSS may outperform a human in terms of concentration or calculative power and thus may help the user to make a better decision (Agrawal & Prediction machines, 2018). A spell checker in text editing software, for instance, helps the user to write grammatically correct sentences, but the user still has the executive power to accept or reject the suggestion (i.e. the grammatical correction). This increases trust in the DSS (Mesbah et al., 2019). In this process, the user starts by manifesting their guess by typing the words, and the DSS assesses these. Nothing else happens in more advanced DSS, but those responsible in the decision-making scenario often do not manifest their guess even though they might have one. Nonetheless, there is a notable increase in user acceptance (Thagard, 1989) and perceived trustworthiness (McKnight et al., 1998) if the suggestion is consistent with prior beliefs. Since we aim to investigate how the justifying of the AI’s suggestions impacts the user’s trust, this poses a problem for us as uncontrolled accordance with the participant’s prior beliefs could lead to a bias in our experiment. To prevent this, the participants start by making a chess move on their own. Afterwards, the AI-based DSS acts in the form of a Supporting AI by suggesting a move that differs from the one by the participant, which forces the said participant to decide between the personal move and the suggested move by the Supporting AI.
Fuzzy String Matching with a Deep Neural Network
Published in Applied Artificial Intelligence, 2018
Daniel Shapiro, Nathalie Japkowicz, Mathieu Lemay, Miodrag Bolic
An initial OCR output correction system was developed for this work based upon the ideas in Bassil and Alwani (2012) to correct OCR output text using the Google search engine’s built-in spell-check correction. The OCR output text was submitted to the search engine, and if the “Showing results for” field appeared in the results page, the term that the search engine expected replaced the OCR output text. This system correctly classified 1721 keywords out of 2480 in TESTING (69.40% accuracy). A second OCR output correction system was developed for this work to further evaluate the effectiveness of correcting spelling mistakes in OCR output text. The spell-check module (McCallum, 2014–2016) selects the best candidate correction from a variety of ranked options including known misspellings of words, dictionaries of words, word lists generated by parsing natural language documents, and word snippets. The ranking is based on word use frequency in the reference document, preferring more common terms over less common ones. The autocorrect dictionary was updated to include the corpus of keywords from TRAINING into its database. After this upgrade, the system correctly classified 1993 keywords out of 2480 (80.36% accuracy) when processing TESTING.