AI Assistants

Whether you are seeking new approaches to cancer research, aiming to predict protein structures or searching for improved research options: artificial intelligence (AI) is playing an ever-increasing role in science. This is because AI tools help simplify the daily work of researchers on many levels. Here, we introduce five tried-and-tested smart solutions:

AlphaFold: folding proteins per mouse click

In 2020, AlphaFold impressively demonstrated how AI is capable of enhancing research. This deep-learning algorithm by Alphabet subsidiary Google DeepMind was able to calculate the three-dimensional structure of proteins at previously unachieved speeds. Proteins that had previously been measured manually by researchers served as the training data for AlphaFold. In the meantime, this tool has calculated the structure of almost all of the 200 million known proteins – with high accuracy. A giant leap for the life sciences, as those who know the structure of a protein will be able to find out what it does – and thus better understand processes and diseases within the body.

Researchers worldwide are now using AlphaFold to study antibiotic resistance and herpes viruses or investigate novel vaccines, earning the developers of AlphaFold to share the Nobel Prize in Chemistry in 2024. In the meantime, DeepMind has not been idle: the new version of AlphaFold 3 is able to not only calculate the shape of proteins, but it can also predict how they will interact with other molecules – with an accuracy that had not been achieved previously. With only a few mouse clicks, researchers can now model complex interactions between proteins and active agents. Thus, this new version of AI is even more useful to basic science and drug development research than its predecessor.

AIFS: more accurate weather forecasting

In times of climate change, weather forecasts are increasingly gaining importance: the more accurately heavy rain, tornadoes, droughts and other weather extremes can be predicted, the earlier governments will be able to initiate targeted countermeasures. At this time, weather forecasts are calculated by supercomputers using complex physics-based models. This takes a significant amount of time, and even more electricity. The new AI-supported forecast system AIFS by the European Centre for Medium-Range Weather Forecasts (ECMWF) is meant to carry out this task noticeably faster and more precisely, at only one thousandth of the energy required by traditional technology.

Educated using historic weather data, this AI identifies patterns which it then compares to current data measured by satellites and weather stations. The thus generated prognoses were 20 percent more accurate when it came to the prediction of tropical storms than modern physics-based models, states the ECMWF. But also, in the area of climate research, AI will likely capitalize on its prowess in pattern recognition. For example, Helmholtz researchers have recently developed a basic AI model which is intended to enable more long-term as well as more reliable climate prognoses without the need for a supercomputer.

Consensus, ScholarAI and SciSpace: AI conducts research

Every scientific publication begins with an overview of the current state of research. AI-supported scientific search engines like Consensus, ScholarAI and SciSpace can considerably speed up this process. First, one poses a research question such as “Can animals sense earthquakes?” or “Is a vegan diet healthier than an omnivorous one?” The AI tools then comb through millions of scientific papers and deliver compact summaries of the most relevant publications. In order to evaluate the relevance of each respective publication, Consensus provides additional information such as how often the work has been cited and on which method or sample size the study is based. A “Consensus Meter” further shows whether there is overall consensus within the research field on a given topic or whether there is controversy. The question of whether animals are able to sense earthquakes yielded the following results: eight percent of the scientific papers state that animals can sense earthquakes, eight percent said no, and 83 percent say that it is possible.

Wordvice AI, Jenni AI: digital editing

After the research phase comes the writing. Here, too, one may request assistance from supporting AI tools. Writing assistants such as Wordvice AI or Jenni AI provide suggestions for improving style or avoiding duplications – and they check the grammar of a number of languages. Jenni AI also serves as first aid in the case of writer’s block: once a sentence has been started, this AI makes suggestions on how it could continue. Wordvice AI, on the other hand, offers to check for plagiarism. This tool searches more than one billion texts in online sources worldwide to ensure that one, while staying focused on the work, does not accidentally “borrow” phrases and ideas from other research papers.

Transkribus: deciphering ancient texts

Historians spend much of their time in the archives – backs bent, poring over historic documents. Medieval codices, official documents from the Early Modern Period, or private letters from the 18th century would have much to tell if the process of deciphering them was not quite so cumbersome and time-consuming. AI tools such as Transkribus are available to facilitate work involving historic sources. This platform, developed by the University of Innsbruck within the framework of an EU-project, transcribes ancient scripts such as Sütterlin or Kurrent, but also Arabic and Hebrew, into a digital text. To date, the community has already generated more than 200 transcription models in over 75 languages. Researchers are able to train these neuronal networks for their own purposes in order to uncover historic treasures in a way that is gentle on their backs and eyes.

More articles from
"Off the Bench"

Subscribe newsletter