Can AI be used for Science Communication?
“Bzzzz”… Can you hear it? Yep, that’s the buzz of everyone talking about AI, and more specifically AI for science communication. Can AI be used to communicate science? Can AI replace a science communicator’s creativity or expertise in translating complex topics? And on the flip side, how will AI in science communication be positive? Will it give scientists more tools to communicate their research to a broader audience? Today we consider some of these questions as SayoStudio’s founder Nicolle Fuller shares her first impressions using AI image generators DALL-E , MidJourney, and Adobe’s Firefly.
Science Illustrator Nicolle Fuller’s first impressions of AI image generation for science
For some time now, our studio’s designers and animators have been discussing AI’s implications for science visuals, and more largely, science. Thus far, as SayoStudio’s owner and director, I’ve largely sat back and listened. Honestly, I knew the impacts would be huge, but I’ve been shocked to see how advanced many of these tools are. I’ve finally jumped in to try out many of these tools myself. Today, I’ll share my initial reactions to AI and discuss its current capabilities for communicating science.
AI tools in the context of previous digital art innovations
After two decades of continually learning new tools to visualize science, I expected AI’s impact to be like the many other technological innovations I’ve witnessed. A slow roll. I’m remembering the introduction of digital art tools like Photoshop’s clone stamp, or the early integration of 3D software (yes, I’ve been doing this for awhile ;-)). These tools were all a bit threatening to those of us originally trained to use pencil and paper, but ultimately they became part of a larger toolkit.
Like any good tool, they still required creators to understand the principles of good composition, information flow, and an understanding of science. In many ways, 3d software opened up the field for more ex-scientists to join the ranks of science communicators, only enhancing the field. Contrary to what many feared, those with hand-drawn painting skills retained a niche at creating pieces with that unique ‘je ne sais quoi’ that comes from the micro-differences.
Will AI be yet another tool, or will it completely reshape our approach?
Curious yet? Reviewing AI science images produced by AI
Let’s jump in and see what AI is capable of. AI-generated art is undeniably impressive in its ability to create visually appealing images, incredibly quickly. Although AI gets a bad rap for producing images that all look the same, I’ve found AI’s strength is in variating style. I wonder if some of the tendency toward AI-generated art-similarity may stem from users not having the artistic vocabulary to change different prompts and different platforms. Initially, I tried Adobe Firefly and DALL-E, and thought that perhaps it was a dataset problem. Adobe’s Firefly seemed to provide more unique stylistic interpretations to my prompts, compared to DALL-E. But then, I tried Midjourney, and it’s a powerhouse at creating different looks. Perhaps not coincidentally, MidJourney is also the AI image generator that gives me the most pause when it comes to ethical implications or artists and copyright infringement (for a future post!)
ChatGPT’s Dall-e AI-generated science-inspired images:
Midjourney AI generated science-inspired images
Adobe Firefly AI generated science-inspired images
As you see in these examples, errors abound. There are a ton of problems with scale. Bacteria are larger than macrophages, and cell organelles are only suggested, not depicted in detail. Overall, there were some beautiful images. How can we make use of this? What all of these programs DO well, is presenting different styles and directions for more editorial art.
AI for Science Visual Style Experimentation
AI isn’t going to cut it for the accuracy needed for science figures and diagrams. Yet, where we see promise is using AI as a tool to experiment and play visually. AI can generate initial concepts to be used as inspiration for the hard work of integrating accurate science information.
For example, AI can quickly produce a range of visual styles for a scientific concept, allowing illustrators to experiment with different approaches before settling on the final design. This can save time, spark creativity and provide a broader palette of ideas to choose from. AI can also help visualize abstract or complex concepts in novel ways, offering fresh perspectives that might not have been considered otherwise.
Furthermore, this may have great potential to facilitate better collaboration with scientists. AI-generated visuals can serve as a starting point for discussions, allowing clients to play with different styles to see what resonates with them. This interactive process can help clients articulate their preferences more clearly, leading to more effective and satisfying collaborations. We used this approach recently, discussed in our post on AI for cover art.
The Future of AI in Science Communication: Accuracy and Beyond
As we move forward, it will be fascinating to see how quickly AI’s accuracy improves. Advances in machine learning and data processing are continually enhancing AI’s capabilities. It is likely that future iterations will address many of the current inaccuracies. However, even as AI becomes more accurate, the role of human judgment and expertise will remain crucial.
Science communicators and visualizers will need to act as curators, using their knowledge and experience to select, refine, and validate AI-generated content. Our role will evolve to focus more on ensuring the accuracy and appropriateness of the visuals we create. This involves not only correcting errors but also interpreting scientific data in ways that are both visually appealing and scientifically accurate.
Balancing Innovation and Responsibility
The integration of AI into science communication presents a delicate balance between embracing innovation and maintaining responsibility. While AI offers exciting possibilities, we must be vigilant in ensuring these tools are used appropriately. This means rigorously testing AI-generated content for accuracy, while being transparent about the limitations and potential biases of these tools. Moreover, as science communicators, we have a responsibility to educate our audiences about the role of AI in the creation of scientific visuals. By being open about the use of AI, and the steps taken to ensure accuracy, we can build trust and maintain the integrity of our work.
In future posts, we hope to consider in more detail the following risks and challenges:
- Accuracy Problems: Clearly, we’re just not there yet for accuracy. Which isn’t that surprising, seeing as AI has such a hard time with human hands. How can it be expected to accurately depict a molecule’s chirality? We’ll delve into this further, but for now, we point you to the infamous mouse testicles AI figure.
- Ethical Implications: Artists and other creators, ourselves included, have serious reservations about AI art. It exists from mining images from the internet, from artists whose copyrighted work was used without permission (read up on our Copyright Basics). In my Youtube video, I ask AI to “create an image like Nicolle Fuller’s NSF work”. It was shocking to see it create something that clearly pulled from my work (to some degree). Lawsuits are in process, but the law is slow to catch up to technological advances like this. In all truth, I’d strongly prefer NOT to have had my work data mined. But now that the genie is out of the box, is there any going back? We’d love to hear your thoughts, as we continue discussing.
- Legal and Business concerns: Similar to point 2, if you’re a scientist or a business hiring an artist, you need to be aware of how they’re creating it. There are serious risks to a company who uses AI art, as the laws governing AI usage is very much a state of the wild-wild west. This is yet another area that we’re keeping up on, and we hope to share what we learn in the months and years to come.
AI for science, is it useful? Will it be useful?
AI is poised to revolutionize the field of science communication. It aims to offer new tools for idea generation, style experimentation, and client collaboration. However, current limitations in accuracy highlight a need for caution and critical oversight. As science illustrators, our expertise and judgment are critical in ensuring the visuals we create are not only compelling but scientifically accurate.
As we navigate this evolving landscape, our commitment to accuracy, responsibility, and innovation will guide us in making the most of AI’s potential while upholding the highest standards of scientific integrity.
Sign up for our newsletters below and get our FREE AI Guide delivered to your inbox.