The past year has seen the emergence of generative artificial intelligence tools such as ChatGPT.GeminiOpenAI's video generation tool Sora — captured the public imagination.
Everything you need to start experimenting with AI An internet connection and a web browser. You can interact with the AI just like you would with a human assistant. You can talk to the AI, write to it, show it images and videos, or all of the above.
© Unsplash
Although this capability represents entirely new territory for the general public, scientists have been using AI as a tool for years.
But as public knowledge about AI increases, so too will public scrutiny of how scientists use it.
AI is already revolutionizing science. His 6% of all scientific research leverages AI, including not just computer science, but chemistry, physics, psychology, and environmental science.
Nature, one of the world's most prestigious scientific journals, has included ChatGPT in its 2023 Nature 10 list The most influential and until then exclusively human scientist in the world.
There are two aspects to the use of AI in science.
On one level, AI can improve the productivity of scientists.
Lawrence Berkeley Lab used AI when Google DeepMind released an AI-generated dataset of more than 380,000 novel material compounds Performing compound synthesis experiments on a scale orders of magnitude larger than what humans can achieve.
But AI has even greater potential, allowing scientists to make discoveries that simply would not be possible with any other method.
AI algorithm that first discovered signal patterns in brain activity data This indicates the onset of an epileptic seizure, a feat that even the most experienced human neurologists cannot replicate.
Early success stories of using AI in science have some envisioning a future where scientists collaborate with AI scientific assistants as part of their daily work.
That future is already here. CSIRO researchers are experimenting with his AI scientific agent and have developed a robot that can perform scientific tasks during fieldwork following spoken language instructions.
Modern AI systems, especially so-called artificial general intelligence tools, are extremely powerful; Things like ChatGPT and Gemini also have their drawbacks.
Generative AI systems are susceptible to “hallucinations”” So they make up facts.
Or maybe you're biased. Google's Gemini depicts America's Founding Fathers as a diverse group This is an interesting case of overcorrecting for bias.
The risk of AI fabricating results is very real, and it's already happening.It is relatively easy to obtain a generative AI tool to cite non-existent publications.
Furthermore, many AI systems cannot explain why the outputs they produce are produced.
This isn't necessarily a problem. It doesn't hurt if AI generates new hypotheses that are tested using normal scientific methods.
However, depending on the application, the lack of description may be a problem.
Reproducing results is a fundamental tenet of science, but if the steps taken by AI to arrive at a conclusion remain opaque, replicating and verifying results becomes difficult, if not impossible.
And that can undermine public trust in the science that is produced.
Here we need to distinguish between general and narrow AI.
Narrow AI is AI that is trained to perform a specific task.
Narrow AI has already made great strides. Google DeepMind's AlphaFold This model has revolutionized the way scientists predict protein structures.
However, there are many other success stories that are less well known. For example, CSIRO is using AI to discover new galaxies in the night sky, and IBM Research is developing an AI that rediscovered Kepler's third law of planetary motion.or building an AI that was able to replicate the scientific breakthrough that won Samsung AI a Nobel Prize..
Confidence in limited AI applied to science remains high.
AI systems, especially those based on machine learning techniques, rarely achieve 100% accuracy on a given task. (In fact, machine learning systems outperform humans at some tasks, and humans outperform AI systems at many tasks. In general, humans using AI systems outperform humans working alone, and There is extensive scientific evidence for this fact, including this study.)
AI in collaboration with expert scientists who review and interpret the results is a completely legitimate and widely viewed method. This is likely to result in better performance than either human scientists or AI systems working alone.
General AI systems, on the other hand, are trained to perform a wide range of tasks, rather than being limited to a specific domain or use case.
For example, ChatGPT can compose a Shakespearean sonnet, suggest a recipe for dinner, summarize a body of academic literature, or generate a scientific hypothesis.
When it comes to AI in general, the problems of hallucinations and bias are among the most serious and widespread. This doesn't mean that general AI isn't useful to scientists, but it should be used with caution.
This means that scientists need to understand and assess the risks of using AI in a given scenario and weigh them against the risks of not using AI.
Scientists now routinely use popular AI systems to assist them in writing their papers.assist in reviewing academic literature and also develop experimental plans.
When it comes to these scientific assistants, one danger could arise if human scientists take their achievements for granted.
Of course, a well-trained, hard-working scientist wouldn't do that. However, many scientists out there are simply trying to survive in a cut-throat industry where they must publish or perish.Scientific fraud is already on the riseeven without AI.
AI could lead to new levels of scientific misconduct, either through deliberate misuse of the technology or through sheer ignorance on the part of scientists who are unaware that the AI is making things up.
Both narrow and general AI have great potential to advance scientific discovery.
A typical scientific workflow conceptually consists of three phases. The idea is to understand which problem to focus on, run experiments related to that problem, and use the results for real-world impact.
AI can help in all three of these phases.
However, there is a big caveat. Current AI tools are not suitable for use directly in serious scientific research.
Public trust in both AI and science will only be gained and maintained if researchers responsibly design, build, and use the next generation of AI tools that support the scientific method.
There is value in getting this right. The possibilities for using AI to transform science are endless.
Famous words from Demis Hassabis, the iconic founder of Google DeepMind: “Building ever more powerful and general-purpose AI safely and responsibly requires solving some of the most difficult scientific and engineering challenges of our time.”
The opposite conclusion applies as well. Solving the toughest scientific challenges of our time will require building general-purpose AI that is more capable, safe, and responsible than ever before.
Australian scientists are studying it.
This article was first published 360 information Under Creative Commons License.read Original work.
Professor John Whittle is the Director of Data61 at CSIRO, Australia's National Research and Development Center for Data Science and Digital Technologies. He is co-author of the book Responsible AI: Best Practices for Creating Trustworthy AI Systems.
Dr Stefan Haller is Program Director for AI for Science at CSIRO's Data61, where he leads a global innovation, research and commercialization program aimed at accelerating scientific discovery through the use of AI. He is the author of the Lancet paper “It's not just about caution: The complex case for the ethical use of large-scale language models in healthcare and medicine.”
Stefan Harrer is an inventor with several US and international patents related to the use of AI in science.
