12 great examples that show what artificial intelligence can do

Like all technical innovations, artificial intelligence has the potential to make our lives easier, for example by helping detect credit card fraud or developing new medicines more efficiently. Thanks to AI, it was also possible to develop vaccines against the coronavirus in record time with previously unimaginable effectiveness.

Students, on the other hand, love AI chatbots like ChatGPT… well… that’s why.

At the same time, AI can make our lives more difficult, for example when it is used for surveillance or military purposes: apparently a majority of Swiss companies are already using AI-enabled software to monitor and control their workforce – and the trend is increasing rapidly. Chinese authorities are even using facial recognition and artificial intelligence to search for Uighurs. Fake news created with generative AI and artificially generated photos and videos is another problem area that will concern us for a long time.

In this article we focus on the positive aspects that are sometimes overlooked in the jungle of negative headlines.

Spam filters against unwanted calls have been around for a long time. Google has gone one step further and has been equipping its Pixel smartphones with an AI-based ‘call filter’ for some time now. When the feature is activated, the Google Assistant records calls from suspicious or all unknown numbers and tries to find out who is calling and why. You can read the conversation between the Google AI and the caller and intervene at any time if you want to speak to the person yourself.

Weather forecasting involves processing enormous amounts of data to make predictions. AI models are also used for this. Deepmind, a subsidiary of Google, can now produce 10-day forecasts that deliver better results 90 percent of the time than the industry’s current “gold standard,” the European Center for Medium-Range Weather Forecasts (ECMWF). Deepmind fed its new AI model Graphcast with 40 years of historical weather data.

The ECMWF has confirmed the quality of the AI-powered forecasts. Not only does Graphcast deliver better and much faster predictions than traditional systems based on the power of supercomputers, the AI ​​also uses only one-thousandth of the energy. However, the AI ​​was no better than traditional forecasts when it came to Hurricane Otis, which developed unexpectedly quickly; Graphcast is also far from perfect.

In the near future, the AI-based model will not replace conventional weather forecasts, but complement them. Deepmind has released the source code of Graphcast and made the technology available to researchers worldwide.

Part of AI research is about making people’s lives healthier and longer. This is exciting for researchers and lucrative for companies. For example, the AI ​​system AlphaMissense developed by Deepmind can detect potentially pathogenic point mutations.

This is significant because previously it was unknown for 98 percent of all point mutations in our DNA whether they were harmful, beneficial or ineffective. Of the 216 million possible point mutations, 32 percent are potentially disease-causing, the AI ​​predicted. Point mutations arise, among other things, from copying errors during cell division, which can lead to genetic diseases.

Not least thanks to more powerful AI models, it was possible to develop several highly effective Covid-19 vaccines in record time. Using an AI model it was possible to predict the shape of proteins with great accuracy. This is crucial for quickly studying viruses and developing effective vaccines.

Artificial intelligence is also used in the development of medicines, which can reduce the time for research and development and therefore costs. If development becomes cheaper, the chance that previously economically uninteresting drugs will come onto the market increases.

For example, scientists have managed to develop an antibiotic with the help of AI. AI acted as a sniffer dog for active substances.

Artificial intelligence can increase the chance of surviving blood poisoning. Researchers from the Technical University of Vienna (TU) had an AI learn how to most successfully treat people with blood poisoning using data from intensive care units in various hospitals. To do this, the AI ​​had to treat virtual patients with sepsis. If they then get it right, the AI ​​is rewarded. If the condition worsens or death occurs, she is punished.

The cure rate of artificial intelligence “is now higher than for purely human decisions,” the researchers write. For legal reasons alone, an AI will not make medical decisions in the near future, but it can advise staff.

AI models have long supported doctors in cancer screening. AI finds breast cancer lumps in black and white ultrasound more reliable than human eyes.

A well-trained AI recognizes the subject we want to photograph and, as if by magic, chooses the right camera setting. But AI often supports us without us noticing. When the smartphone turns the shaky holiday video into a calm, almost perfect film, this is done thanks to AI. And when the smartphone automatically brightens a photo that is much too dark and optimizes the colors and contrast, the AI ​​also works.

Moreover, professional image editing used to be reserved for a few people. With AI, even absolute novices can now optimize their photos – without any knowledge of Photoshop.

Google’s “Magic Editor” for Pixel smartphones enables AI-assisted photo editing on the smartphone.

Audio AI models can be used to translate spoken text in videos, including lip movements, into other languages. In the example video below, the man initially speaks in English. Start-up Hey Gen’s AI uses it to create videos of it saying the same thing in French and German.

No matter how practical a universal interpreter may seem, the risks of misuse are just as great. Large technology companies such as Google and Meta, which are otherwise not known for their restraint, are therefore still holding back this technology.

AI chatbots such as ChatGPT can also recognize image content and answer user questions about the associated photos.

An example: The user takes a photo of his bike and his toolbox and ChatGPT explains step by step how to adjust the saddle and whether you have the right tools.

ChatGPT should also be able to help solve problems using photos. The following example video from ChatGPT developer company OpenAI shows how this works.

AI took text translation to a new level a few years ago with tools like DeepL. Companies like Spotify are now using AI to automatically and cheaply offer popular podcasts in other languages.

There are over five million podcast episodes on Spotify, many of which are currently only in English. Like DeepL, AI-translated podcasts aren’t perfect, but they make the content accessible to a much wider audience.

In 79 AD, historically valuable papyrus scrolls were buried under the ash fall from an eruption of Mount Vesuvius. Researchers were never able to read the recovered but charred scrolls because they immediately disintegrated when unrolled. In 2023, an AI managed to decipher individual, short text segments of the ancient scrolls from high-resolution CT images. For this purpose, two previously unopened rolls were examined by a particle accelerator.

Historians could now gain at least some insight into what was written on the original scrolls, which are about 2,000 years old. Until now, they had to make do with copies that were repeatedly copied and edited by monks and scribes.

With AI, researchers can more easily determine biodiversity. For this purpose, a research team in northern Ecuador installed recording equipment on pastures and cocoa plantations that were no longer managed and where forests are gradually re-emerging. The recordings of the various animal sounds were then analyzed using AI models.

Previous research has shown that the occurrence of animal species linked to deforestation or forest restoration can be monitored using soundscapes.

The AI ​​model is intended to help monitor reforestation projects to see whether a diverse forest with high biodiversity is actually created. Using traditional methods (camera traps or counting animals or their remains such as feces and tufts of hair), determining species diversity is often very time-consuming.

And finally this: Why programmers love ChatGPT 😉

Resolution of the skin mole image: On the left the malignant melanoma (black skin cancer), on the right an innocent mole.