AI tools for research
We've put together an overview of AI tools that offer new ways of searching for and working with academic information. Although these tools currently cannot be regarded as fully reliable, the AISE (artificial intelligence science and engineering) field is developing rapidly, while necessarily involving ethical questions related to academic integrity and the collection of personal data.
The list below is a selection of AI search tools that we are aware of at present and which can help you navigate academic literature.
As with traditional search tools, results should be approached critically and ideally verified by other means. If you would appreciate more advice on how to understand your search results, schedule a free consultation.
The Literature Review function provides a list of sources based on a prompt or keywords (and operators). It generates a short summary of each paper, as well as compiling the five most relevant ones into a paragraph covering the topic. The tool can also be used for other purposes.
Elicit uses the Semantic Scholar database. Like SciSpace, it attempts to find the most relevant sources in order to answer your question, summarizing the content in a similar way. Note that using keywords and operators instead of questions yields less precise results.
Searching either the Semantic Scholar or the PubMed database, Research Rabbit makes suggestions based on sources that you place into a collection. Mapping connections between the papers you select, it recommends additional results of similar research carried out in the field, as well as a list of relevant authors. Collections can be shared and items in them annotated.
Litmaps uses Semantic Scholar and open access metadata from Crossref and OpenAlex. The Seed function maps references and citations of a selected article.Discover connects multiple papers and suggests other similar sources. Map/Visualize creates graphs based on publication dates and citation counts. Visualizations can be adjusted using the mouse, exported, or shared. Additional functions are available in the Litmaps Pro plan.
Inciteful searches several databases to map interconnections between selected papers. The Paper Discovery function allows you to create a collection and then suggests similar sources. You can also narrow down your results using filters and keywords (this is a subscription feature in Litmaps). The Literature Connector function links two publications of your choice based on their citations.
Scite_ focuses on how a given paper has been cited, making use of a vast range of databases. It analyzes the citations in a selected publication, labeling them as merely mentioning, supporting, or contesting ("contrasting") the selected item, as well as counting the number of self-citations. To provide context, it shows a section of the text that surrounds each citation. Scite_ also informs you if a paper has been amended or retracted after publication. A browser extension and a chatbot function are available. Scite_ is mostly available with a subscription.
AI chatbots generate text based on your prompts. They are built on large language models (LLMs) and usually search the internet. Be aware that they may "hallucinate" (i.e., provide wrong information) and tend to work best in English.
We do not recommend relying on information in chatbot-generated answers. Ideally, read and cite the source that the chatbot has referenced (e.g., an academic paper). If you decide to use chatbot-generated text, make sure that this will not violate any institutional rules, especially concerning plagiarism and citing. For more information, feel free to contact us.
This chatbot answers your prompts by searching the internet and generating text with references to the sources. It also suggests further questions that you might want to ask. The Copilot function, using the GPT-4 large language model from OpenAI, allows access to more precise answers, but its use for free is limited. A mobile app is available.
Previously Bing Chat. This Microsoft product functions as a Bing search assistant. Built on the GPT-4 language model, it generates text based on its search of the internet, referring to the sources it finds. You can choose between Creative, Balanced, and Precise modes, the last being best suited to information searches. A limited number of turns per conversation and chats is allowed per day.
Previously Bard. A Google product that uses its own language model PaLM 2 (and the LaMDA model prior to that), made fully available as the last chatbot in this selection, in summer 2023. It searches the internet but does not reference its sources. Bard is integrated with other Google products, allowing users to export answers and save them as a document on Google Drive or to email them as text from a Gmail account.
AI chatbot from OpenAI, available either as a free version that uses the less evolved GPT-3.5 language model (since autumn 2022) or as part of a Plus subscription plan with access to the GPT-4 model. ChatGPT primarily relies on its own database, with a data cut-off in 2021 (free version) and 2023 (paid version). Internet search is now possible in the latter version.
The following analyzers summarize text and generate answers to questions.
Using text analyzers may save you time, but it is best to approach the results as approximate, at present. If you would like to discuss how to read academic papers efficiently, you can schedule a consultation.
Next to the text summary, the Copilot function can explain (paraphrase) a selected section of an article and search for sources that might be relevant to it. SciSpace reads pdf files or open access articles found through its search engine by typing in keywords or a question. A browser extension is available. SciSpace can also be used for other purposes.
ChatPDF provides a summary of any text uploaded as a pdf file or of an open access paper (accessed via its URL). It answers your questions based on the content and generates additional ones. This AI analyzer uses the same large language model as ChatGPT. Only three tasks are allowed per day without the ChatPDF Plus subscription plan.
Humata summarizes any uploaded pdf text file and uses its content to generate answers to your questions, as well as offering some of its own. It also highlights the relevant sections of the text and provides page numbers with hyperlinks. Free use is limited to sixty pages in total.
Below is a selection of tools whose purpose is to analyze text and determine whether it has been generated with the use of an AI tool (especially a chatbot). Most of them currently analyze only text in English. It can be inserted directly into a text box or uploaded as a file.
Results are estimations and may vary between individual AI detectors. Do not hesitate to contact us if you have any questions.
SciSpace detects chatbots primarily built on the GPT-3.5 and GPT-4 language models and the writing assistant Jasper. It also differentiates between scientific and non-scientific text. For each sentence, SciSpace determines whether it has been generated with the use of AI, after which it assigns the whole text a corresponding percentage and provides a brief commentary. A maximum of a 1,000 words can be analyzed per task.
GPTZero has been trained on various types of data to identify the Bard chatbot and tools built on the GPT or LLaMA (Meta) language models. It uses percentage and word commentary to mark the text as AI-generated, written by a human or both. The free version analyzes the maximum of 5,000 characters per task. Up to three pdf files can be uploaded at once. A browser extension is available and, as a subscription feature, a plagiarism detection function.
GPTKit detects text generated by ChatGPT, determining the extent to which it has been written by a human (labeled as "real") or by an AI tool (labeled as "fake"). The percentage is provided on an opposite scale to, for instance, SciSpace or GPTZero. GPTKit allows up to five analyses per minute (with the maximum of 2,048 characters permitted per task) in the free version.
Winston AI can be used to analyze text in English, French and Spanish. It detects the Bard chatbot, tools built on the GPT models and other AI tools. The result is provided in percentage and, as with GPTKit, it is an estimate of authenticity (i.e., human work). The maximum of 2,000 words can be analyzed overall, after which only the paid version is available (offering also the plagiarism detection function).