Generative AI: From the Stars to the Superbugs, and a Darker Side

Advances in computing tech­nology opened the gateway to processing complex soft­ware applications. Genera­tive AI (Gen AI) is not a new term; it has been part of technological advancement for decades. However, with increased computing power and advancements in hardware devices such as GPUs, it has become much easier to adopt AI & GenAI in the emerging frontiers of Sci­ence & Technology. ChatGPT, Gemini, Cloude are some of the popular GenAI models that are becoming essential for tasks related to scientific analysis, re­search, and knowledge generation and interpretation. A brief timeline of GenAI is depicted as follows and shown in the figure 1.

1 GEN AI APPLICATIONS TO ENHANCE THE QUALITY OF ASTRONOMICAL OBSERVATIONS

1.a. Morpheus, an AI tool

It is an AI model to classify astronomical images captured by JWT. It was trained by the NVIDIA GPU-enabled super­computer, lux, at the University of Cali­fornia, Santa Cruz. It analyses captured images pixel by pixel to identify which pixel belongs to star, galaxy or any as­tronomical object. Morpheus observes galaxy images and finds out which class it belongs to, e.g. spherical, elliptical, spiral, disk and irregular. Morpheus is useful to process Webb’s massive image datasets much faster than manually in­specting everything.

All Images Courtesy: Authors

1.b. Reconstruction of galaxy images: Diffusion Model

A useful extraction from galaxy images depends on its quality, which is mea­sured in terms of pixel density. Earlier methods use semi-analytical models that depend on assumptions, experimenta­tion and parameter tuning. In contrast, data-driven generative models learn from observational data and are able to generate high-quality images of gal­axies. In the diffusion model, noise is gradually added to the original image to learn how noise degrades the quality of the image. Further, noise is removed step by step to learn how data quality degrades. Overall, the diffusion model develops high-quality images by learn­ing how to remove noise from random images. Using the diffusion model, noise is added to high-resolution images and low-resolution images are enhanced. The Dark Energy Spectroscopic In­strument (DESI) and Sloan Digital Sky Survey (SDSS) were applied to train the model. A diffusion model, GalCatDiff, enhances galaxy image features and as­trophysical properties to generate better quality images, as shown in Figure 2.

1.c. Galaxy image simplification using Generative AI

A GenAI method has been developed to simplify distant galaxy images. The method generated publicly available 125,000 simplified galaxy images. Gal­axy10 DECaLS and Illustris were used to develop a GenAI model for the pro­cessing of galaxy images across ultravio­let, visible and infrared bands.

1.d. Generate visualisations of astronomical phenomena

Generative AI can be applied to create visual representations of astronomi­cal phenomena e.g., galaxy formation, galaxy collision, black hole merger and space-time visualisation. Visual­ising them accurately requires solving Einstein’s field equations, a task that has traditionally been computation­ally intensive. GenAI simplifies this by training on existing simulations and observational data to generate realistic representations. GenAI-based models can produce images that help research­ers, academics, and students for better understanding of complex concepts. For example, a black hole merger cannot be photographed directly in the traditional sense, but scientists can model it using physics equations and gravitational-wave data. Generative AI can turn phys­ics equations and gravitational-wave data into visual scenes to model black hole mergers. Black holes spiralling inward, warping spacetime, and com­bining into one larger black hole, such astrophysical events can be visualised. Such AI-generated visualisations are not exact photographs (Figure 2); they are interpretive representations based on available scientific data. It helps the sci­entific community to explore phenom­ena beyond normal human observation and supports science communication, education, and public outreach.

2. GENERATIVE AI AGAINST TWO CONVERGING THREATS: CLIMATE CHANGE AND SUPERBUG EVOLUTION

The world is quietly stumbling into the overlap of two crises it was never quite prepared for. The first is antimicrobial resistance: the stubborn, relentless abili­ty of bacteria to evolve around the drugs we throw at them. It already kills more than 1.2 million people a year. That number, on its own, should be enough to command front pages. It rarely does. The second crisis is more familiar: cli­mate change, which is not just warming the planet but redrawing the biologi­cal map where pathogens flourish, how quickly they mutate, how far they travel. Taken separately, each is a serious prob­lem. Taken together, they are beginning to look like something we genuinely do not have the tools to handle.

Here is what makes the situation so uncomfortable. Developing a new antibiotic, the old-fashioned way, takes somewhere between ten and fifteen years. It costs well over a billion dol­lars. And by the time the drug finally reaches the patients who need it after trials, approvals, manufacturing, and distribution, and there is a real chance the bacteria it was designed to kill have already shifted. Resistance does not wait for paperwork. Climate change makes this worse. Warmer temperatures do not just make summers unpleasant; they ac­celerate the rate at which bacteria swap genetic material with each other, which is the primary mechanism through which resistance spreads from one spe­cies to another. Floods, increasingly se­vere and increasingly common, carry hospital wastewater and agricultural runoff, both laced with resistance genes into rivers, groundwater, and eventually the food supply.

One can already see where this is heading. Candida auris, a fungal infec­tion that kills somewhere around half the patients it hospitalises, emerged simultaneously on several continents within the span of a decade. That is not a coincidence. Researchers tracing its spread have pointed directly at rising environmental temperatures as the se­lective pressure that allowed a thermo­tolerant, drug-resistant strain to thrive where it previously could not. It is, in a very uncomfortable sense, a dress re­hearsal. The resistant organisms of the next decade are not being engineered in some rogue laboratory. They are evolv­ing right now, in warming floodplains and stressed soils, on a timeline nobody controls (Figure 3).

So where does that leave us? Hon­estly, in a better position than we were five years ago, and that is largely because of AI. Machine learning models can now scan millions of chemical structures in the time it once took a research team to run a single assay. They are not just fast; they are finding patterns in molecular data that human chemists, working through conventional intuition and trial and error, would simply never have noticed. Generative AI goes a step further, and it is worth pausing on this because it represents something genu­inely new. These models do not just rank existing compounds. They invent, given a bacterial target like a protein the patho­gen cannot survive without, and a generative model can de­sign a molecule from scratch, shaped specifically to disable that target, built from chemi­cal logic rather than chemical precedent. Studies have already produced real candidate antibi­otics this way, including com­pounds active against Acineto­bacter baumannii, a pathogen so dangerous that the World Health Organization keeps it on its critical priority list.

What is perhaps most strik­ing is how this research is be­ginning to absorb climate data directly. Scientists are building surveillance systems that pull together wastewater samples, environ­mental genomics, and atmospheric data to spot emerging resistance patterns before they become outbreaks. Large language models are reading and syn­thesising the global AMR literature, tens of thousands of papers surfacing connec­tions that no single research group would have the bandwidth to find on their own. Bacteria have been evolving for billions of years, but what is different now is that the tools are, for the first time, beginning to move at biological speed.

3. BEYOND RESEARCH & INNOVATION, THE DARK SIDE OF GENAI: DEEPFAKES, FRAUD, AND MISINFORMATION

The contribution of GenAI to science definitely leads to more dimensions. It not only enhances human research ability but also opens new pathways to explore the unsolved mysteries. Nature explora­tion has always been the prime interest of humans from early civilisation, e.g. ancient Indian and Sumerian, to the pres­ent era. Using GenAI, we can expedite the search from a vast knowledge base, solve complex equations and enhance the visualisation of complex processes. It can generate content on demand and has the potential to produce high-quality, cost-effective results. It enhances creativity by enabling quick decision-making and improving accessibility.

Apart from these benefits, GenAI has significant negative impacts also. These include hallucinations, inap­propriate outputs, bias, and potential threats to security and privacy. Such shortcomings mislead sci­entific innovations and discoveries. The deep­fake information is one of the problems that spreads rapidly and may harm the reputation of an individual or society’s image. The possibility of cybercrimes related to financial fraud and scams may also in­crease due to the misuse of GenAI.

Deepfake Threat Land­scape: The term ‘deepfake’ combines the deep learn­ing concept with something fake.

3.1. Financial fraud using Identity theft

Financial frauds are in­creasing day by day; when integrating with GenAI, its impacts are becoming severe. Generative AI may increase the risk of identity theft by making imper­sonation easier and more convincing. Cybercriminals can use it to generate fake voices, images, videos, and messages that mimic real in­dividuals. It allows them to trick people into sharing per­sonal information, financial details, or login credentials. As GenAI tools become more accessible, identity theft threats are growing more sophis­ticated and harder to detect. Imagine a well-known financial person giving some money-making techniques, offer­ing, etc. Anyone can be trapped and may become the victim of such scams.

3.2. Fake news and Synthetic media

Several instances of fake news are gen­erated by AI tools, and fake videos of Middle East conflicts are circulating on the internet. It is hard to find for a common man to identify the difference between fake videos and the original ones. Such videos may cause civil unrest, political instability, financial losses, and severe damage to a state.

3.3. Hoax images

GenAI can generate hoax images that look highly realistic, making it hard to determine whether they are real or not. Such AI-generated images spread with false information, create panic, dam­age reputations, or manipulate public opinion. Such hoax images are often shared on social media and may be used for scams or propaganda. As GenAI im­proves, verifying image authenticity has become increasingly important.

3.4. How to detect deepfakes

A careful observation of visual details in a video or image is required to spot a deepfake. In such videos, the eye move­ment or eyes do not focus properly, and it is a warning sign. Irregular or less frequent eyelid blinking is another common clue in AI-generated faces. The facial expressions may be unnatural or may not match the situation. An odd shift in face positioning, or it seems slightly misaligned with the head. A mis­match in lip movements with the spoken words, showing a lack of syncing. The body shape or posture sometimes looks distorted or inconsistent. A mismatch hairstyle that may appear blurry, un­even, or change unnaturally between frames. Skin colour may look patchy, too smooth, or change tones unexpect­edly. Paying attention to these signs can help identify manipulated content.

Figure 4 shows a GenAI-suggested way to spot deepfakes.

CONCLUSION

AI and generative AI are, like every transformative technology before them, a double-edged sword. They can map the farthest galaxies, accelerate the hunt for new antibiotics, and help us under­stand a climate in crisis. They can, with equal ease, manufacture falsehoods and sow chaos at scale. The difference be­tween those two outcomes is not written in any algorithm or programming; rath­er, it is written in the choices humanity makes about how to govern, direct, and restrain what it has built. The technol­ogy belongs to this generation. So does the responsibility.

*Dr Varun Barthwal is an Assistant Professor in the Department of Infor­mation Technology at Hemvati Nan­dan Bahuguna Garhwal University, Srinagar Garhwal, Uttarakhand. He has multiple years of expertise in Data analysis, AI & ML and Computer Pro­gramming. (varuncsed1@hnbgu.ac.in)

Dr Digar Singh is an Assistant Professor in Microbiology at HNB Garhwal University, and has expertise in metabolomics, microbial interac­tions and systems biology. (singhdi­gar1986@hnbgu.ac.in)

Prof Hemwati Nandan is a Profes­sor of Physics, and Director, Research & Development Cell, HNB Garhwal University. He can be reached at hem­wati.nandan@hnbgu.ac.in.