AI, famously, will either save humanity or destroy it, depending on who you ask. Recently the media’s focus has mainly been on the pessimistic outlook, which critics point out hinges on  technology that does not currently exist, and is unlikely to exist any time soon. What of the other side’s arguments, though? We are told that AI is going to save humanity through advanced energy solutions and improved medical outcomes, but are these outcomes any more likely than killer robots?

Behind the chatbot hype and doomer headlines, a quiet revolution is continuing in science. It’s quiet because machine learning techniques have been used in certain areas for decades, going back to the AI pioneers of the 1950s, but as the technology has matured it has also spread widely, and now informs almost all areas of science. This can only be good news; as a paper published this week in the research journal Nature puts it, in almost every field AI is "optimizing parameters and functions, automating procedures to collect, visualize, and process data, exploring vast spaces of candidate hypotheses to form theories, and generating hypotheses and estimating their uncertainty to suggest relevant experiments." But what do these improvements mean for humanity’s future? Let’s look at some key areas to see what’s coming!
‍

DRUG DISCOVERY

The advances being made by AI in this field alone could fill many articles, so we’ll focus on one remarkable example of what’s happening. One of the key battlegrounds in drug discovery is tackling increasingly antibiotic-resistant bacteria, and we’re beginning to see real progress in that area thanks to AI. For example, in May 2023, scientists from MIT and MacMaster published a paper detailing their search for an antibiotic to tackle acinetobacter baumannii, a superbug that attacks the immune system, and one of three pathogens identified by the World Health Organization in 2017 as a critical level threat to human health.

‍

This was a two-stage operation; first, the AI was employed to quickly scan 7,500 molecules to assess their ability to inhibit A. baumanii’s growth. This data was then fed into a neural network, which used the information to make predictions about what an effective antibiotic would look like, and this led the team to a compound called abaucin. This wasn’t a discovery, as such; abaucin was already known to scientists, but not in this context. Nobody had examined its antibiotic properties before the tip-off from the AI! 

‍

‍

CLEAN ENERGY 

Another field where there is far too much activity to cover. The search for clean energy solutions is extremely broad, and includes areas where humans cannot tread, but AIs can fearlessly operate. One such is nuclear fusion, where the challenge facing scientists is one of efficiency; they can generate energy from nuclear fusion, but doing so requires careful manipulation of plasma that is literally hotter than the sun, and that causes problems. 

Humans have been building fusion reactors since the 1950s, but we’ve yet to create one that generates more energy than is needed to maintain it. The most famous modern attempt - the International Thermonuclear Experimental Reactor (ITER) -  hopes to launch in 2025, but is over budget and behind schedule. ITER was designed in 2007, using what’s called a tokamak design, where the plasma needed for the fusion reaction is suspended in magnetic fields. It remains to be seen whether ITER will ever generate energy, but the next iteration of ITER-like reactors is likely to be greatly superior thanks to AI. A 2022 paper from Google’s DeepMind division has shown how reinforcement learning can be used to better control the magnetic fields used to manipulate the plasma, and react appropriately to changes in the plasma’s condition, making it easier to create the necessary conditions for fusion. The AI has even been able to separate the plasma into separate “droplets”, controlling multiple plasmas at once within the tokamak! 

‍

GENOMICS

AIs are complex beasts, but so are actual beasts (including humans!), and one of the big challenges facing medical researchers these days is that there is simply too much genomic data to be processed. As the cost of sequencing decreases, the amount of data balloons, and the National Human Genome Research Institute estimates that by 2025, there will be 40 exabytes of storage needed to store all the data collated on human genomes alone. An exabyte is a billion gigabytes, and 40 of them is roughly equivalent to eight times the storage needed to store every word uttered in human history, or 2.6 million Library of Congresses. It’s a lot. 

Analyzing that volume of data would be all but impossible without modern machine learning techniques, and even with them there are still many bottlenecks in the process that slow the rate of analysis down. Ultimately, though, AIs are bringing us far closer to goals such as “precision medicine”, where a patient’s genetic variabilities are factored into their treatment, as well as broader objectives such as improved drug discovery or better understanding - and prevention of - disease. Also, as alluded to earlier, it’s not just human genomics being analyzed; the plant and animal kingdoms are also being carefully studied, which is leading to advanced agriculture and animal husbandry techniques, as well as further discoveries that can improve human health. 
‍