Four Ways AI Has Already Changed Science

Most of the world is fixated on how artificial intelligence (AI) will change things in the future, for both good and ill. But the fact is AI has already made some dramatic changes to the present, even if they are not blatantly obvious yet.

The medical and scientific research communities, naturally, are at the forefront of AI deployment, and they have already made significant findings that will lead to new products and even alter the way we view matter and the universe.

In this article, we discuss the four game-changing discoveries that AI has made in the past year alone.

Four Ways AI Has Changed Science

1. AI-Accelerated Medical Research

Perhaps the most significant advancements are being made in the field of biology. Earlier this year, the University of Michigan announced it had developed a new model that can autonomously perform up to 10,000 experiments on bacteria per day. The BacterAI platform represents a vast increase in the scope and speed of medical research, potentially leading to a rapid uptick in the development of life-enhancing, even life-saving, techniques.

The human body is home to trillions of bacteria, both inside and out, some of which are harmful, but many are vital to our everyday bodily functions. By better understanding how they live, grow and interact with other bacteria and tissue, the scientific community could be on the cusp of uncovering new ways to cure a wide range of infections and also create the means to extend human life and overcome the physical and cognitive issues that accompany aging.

What’s more, BacterAI can begin its research with no prior knowledge of any particular species of bacteria, 90 percent of which have yet to be studied. The model creates its own data sets, designs its own lab experiments, interprets the results, and then distills its findings in ways humans can understand and examine for further testing.

In a recent demonstration, the platform analyzed the metabolism of two common oral bacteria by testing combinations of amino acids, which can number into the millions. It completed the task in nine days with 90 percent accuracy – a task that normally would have taken weeks, months, or even years.

2. New Medicines

Drug discovery is also an area that stands to benefit greatly from a generous influx of AI. MIT Technology Review’s Will Douglas Heaven notes that AI not only has the potential to increase the pace and lower the cost of drug research – which can take up to 10 years and cost billions of dollars – it can also jump-start efforts to cure rare diseases that often get overlooked due to their limited demand.

Many cancer victims, for example, can benefit greatly from customized drug cocktails tailored to their specific DNA and other markers, but these are difficult and expensive to create – usually through a lot of trial and error. AI can dramatically improve this process by examining minute changes in actual patient tissue to see what works and what doesn’t faster and more accurately than traditional methods.

But perhaps the more profound discoveries being propelled by AI fall within the more esoteric, but no less important, realm of pure scientific research. This is where some of the most important advancements in knowledge have taken place in the past (F=ma, E=MC2), and AI can excel in this capacity by crunching large amounts of data and finding hidden patterns in numbers that tend to elude human observation and analysis.

3. Reimagining the Physical Universe

Researchers at Columbia University, for example, recently uncovered a new set of variables that govern the movement of particles, ushering in an entirely new understanding of physics. The team developed an AI model that could analyze video of objects in motion – everything from a simple double-pendulum to flames in a fireplace – and predict future movement by identifying variables like momentum and angular velocity. Most of the variables it detected were well-known, but a few were not, and these proved crucial to the accuracy of the model’s predictions.

Even more interesting, when the researchers ran the experiments multiple times, the number of variables remained the same, but the specific ones used to make the predictions changed. This leads to the possibility that there are multiple ways to describe the universe and that some of the thornier issues that have bedeviled physicists over the decades can be solved by re-considering the variables. Time travel, anyone?

4. Material Gains

A little more down to earth is the work being done at the Brookhaven National Laboratory that resulted in three new self-assembling nano-structures using the same kind of AI-driven “self-experimentation” that is enhancing medical research. Normally, working out all the structural combinations of existing materials to create new ones is tedious and expensive. Brookhaven’s model can autonomously define and perform experiments from scratch, using data from spectral analysis, ultra-bright x-rays, and other techniques to work out the most promising combinations.

One potentially significant find was a unique nano-scale “ladder” that emerged after only six hours of experimentation – a process that would normally take a month or longer. Potential applications for all three new materials range from microelectronics to clean energy solutions.

The Bottom Line

Every great era of scientific discovery was prompted in large part by the development of new tools and techniques – the telescope, the microscope, radio, and digital computing. Each time, knowledge was expanded by enhancing our own senses to see, hear and otherwise perceive the universe. AI has the potential to take that a step further by perfecting the ability to ingest data and form conclusions – essentially enhancing the brain’s ability to think.

Right now, AI remains largely in the hands of leading scientific institutions. As it scales and becomes more ubiquitous, however, there is no telling what truths it will unveil next.