Best AI research assistant to analyze interviews 2024

How can AI enhance academic research?

In recent years, and especially since the launch of ChatGPT, many tools have been developed that use AI to support scientific work and are becoming more widely known in the academic community.

Although the academic community is divided on the importance of AI for research, the number of those who believe that it will help to unlock the potential of research is growing. Many academics are aware of and generally interested in the development of AI, but rarely use AI in their day-to-day work (with the exception of increasingly common technologies such as automated translation or grammar correction).

This may be changing.

Large language models clearly have much to offer academic research. Their ability to process human language effectively, to understand complex questions, to find relevant information in large bodies of text, and to generate coherent text based on specific instructions make them suitable for automating some of the most repetitive and tedious tasks in research, and allowing academics to focus on understanding problems and developing ideas.

Why we built aureka as a research assistant for audio and video

These advances are mostly focused on textual formats. If you search for “AI research assistant”, you will find that most of the tools reviewed refer to tools that help with tasks such as searching for textual references, obtaining literature summaries, "chatting" with a document or brainstorming through conversation with an AI.

However, audio and video represent a very rich source of information for research. We produce and interact with more audiovisual content every day. Most qualitative research projects require interviews to gain in-depth insights, and TV and radio are among the most influential media of the twentieth century.

Either if it is for academic research drawing on interview data, consultancy, market research or focus groups, working with interview data can be very time consuming, especially if we don’t count with adequate tools.

We developed aureka to fill this gap.

Improving the quality of automatic transcription opened the door to a new way of working with audio (visual) material. But transcription is only the first step.

aureka is the AI research assistant for audio and video. It transcribes automatically and uses the transcript as the basis for various forms of automation useful for research work: full-text search synchronised with the audio/video content, automatic translation, synchronisation of your annotations with the recording, automatic identification of people, places and organisations mentioned, visualisations and much more.

AI assistants for audio and video

There are several ways to use AI to support your work with audio and video.

• AI-powered transcription software: the many AI-powered transcription software out there (such as Sonix, Otter, Amberscript or Trint, to name a few) will help automate the first – and probably most time-consuming – step of analysing interviews. You no longer need to manually transcribe the audio. This is great! However, most transcription software doesn’t seem to have the researcher’s use case in mind, and doesn’t offer useful features such as AI powered searching based on semantic similarity, or user friendly ways of keeping an overview of one's own annotations for further use in publications. 
• Qualitative research software can also be used to analyse interviews or other audio/video recordings, provided that we upload the transcript (Atlas.ti), or use an integrated automatic transcription service for an extra cost, which is often more expensive that most translation tools (NVivo or MAXQDA). All in all, there is a high cost in terms of time and/or money.

Some of these tools have introduced AI features powered by OpenAI.

• Some transcription tools, such as Sonix for example, offer AI analysis tools for extra cost. They provide summarization and extraction of topics and sentiments, which is very usefull to get a quick overview of the transcript's content. However, it might not be suited to work with predefined (e.g theory-oriented) categories with a deductive approach. 
• In the case of MAXQDA, automatic summaries of texts that share a code are generated, which is very useful for keeping track of coded content and having notes ready for the writing process. However, the task of coding is still done manually or with a simple automation (without AI) that codes text fragments that contain a keyword.
• Atlas.ti uses OpenAI to provide automated coding. AI processes the selected document and applies inductive coding to the data. The user then has to review each code and can edit or delete it. Although this can be very useful if inductive coding is what we want, it may not be suitable if we have a coding system in place beforehand. Its "intentional AI Coding" can help if there is a coding system given beforehand, as it formulates questions based on the coding intent, and these then guide the coding process. The assigned codes should then be reviewed at the end, which might very time-consuming in the case of a large corpus. 

aureka’s approach to AI-based interview analysis

aureka automates qualitative analysis, enabling both inductive and deductive coding, while keeping you in full control.

Our system is designed with a human-in-the-loop approach. This means that we combine AI automation with human intervention at the right moment to refine the coding system and thus reduce the need of reviewing each individual code at the end.

aureka’s automated AI coding includes the following steps

• aureka processes the transcripts and creates a summary of each document using generative AI. It extracts descriptive keywords that summarise the topics covered – even if the terms don’t appear literally in the text, allowing for a higher level of abstraction.
• If you have a coding system, you can enter it and aureka will compare the generated keywords with the codes in your system.
• It will then return the codes found and the generated terms that did not match as a suggestion for extending your coding system. This is where human control comes in, and you can decide which new terms to include or not.
• With a refined coding system, aureka will match text fragments to the codes based on semantic similarity. You can check a few to check the accuracy, but if it is high enough, there is no need to check every code.

In this way, aureka’s approach to automated coding saves a lot of time without compromising accuracy.