In a recent announcement with President Trump, three powerful tech leaders and an investor from the Middle East have announced a plan to invest more than $700 billion in building data centers in an attempt to make the United States the world’s leader in artificial intelligence (AI).
Larry Ellison, the creator of the relational database system Oracle; Sam Altman, CEO of OpenAI; and Masayoshi Son, an IT entrepreneur from Japan, indicated that AI could revolutionize healthcare.
This is a very big project, not unlike the Manhattan Project during the Second World War. There are currently 20 data centers planned, each with half a million square feet. Smaller data centers are perhaps 5,000 square feet or less; mid-sized data centers run 5,000 to 50,000 square feet; and large data centers typically run from 50,000 to 100,000 square feet.
A data center of greater than 100,000 square feet is called a “hyperscale” data center. Each of these 20 planned centers would be five times larger than a basic “hyperscale” data center. Essentially, this project is calling for the largest data centers in the world.
Electricity Requirements
These giant AI complexes will need their own electricity generation, separate from the grid. Data processing requires a great deal of electrical energy. There are two principal uses. First, the processors use a lot of energy; second, data centers must be air-conditioned so the chips do not overheat.
For an AI system such as ChatGPT, each rack of processing units uses around 120 kilowatts. A “rack” is a shelving unit that holds processing units that are stacked together. Some call these “pizza boxes.” Each rack takes up a little more than 400 square feet. So, if a data center is 100,000 square feet, it would hold 240 racks.
Twenty data centers of half a million square feet apiece adds up to 10 million square feet, so at 400+ square feet per pizza stack, that would amount of 24,039 racks. This would amount to 2,885 megawatts of electricity, plus another 40 percent for cooling, or about 4,039 megawatts.
To keep this in perspective, the total production of Niagara Falls is only 2,600 megawatts.
Using Small Modular (Nuclear) Reactors
According to some estimates, growing use of AI will account for a quarter or more of electricity consumption in the United States within a few years. In order to avoid stressing the public electrical grid, which already is inadequate, as noted, this project intends to feature separate, dedicated power stations to serve these data centers. They will have their own supply of electrical power and thus be less vulnerable to disruption caused by a public blackout.
To do this, thought is being given to clean energy such as small modular nuclear reactors (SMRs). For example, NuScale Power Modules generate 50 megawatts each. At this rate, the proposed AI computing complex could be powered by 80 or so SMRs.
The United States has launched other electricity-intensive projects during its history. The giant secret facility built in Oak Ridge, Tennessee to enrich uranium for the atomic bomb used one-seventh of all the electricity produced in the United States, about 22 million megawatts. Things have changed. In the 1940s, the production of electricity in the United States was 153 billion kilowatts; today, the United States is producing around 4 billion megawatts.
AI and Medicine
Much early writing about AI and medicine focused on how it could replace doctors and make diagnostic decisions. The first RACmonitor article on AI was published in January 2019. There has been almost universal derision of this new technology by the medical guild, and initially no one believed that AI would “learn” how to be a doctor.
But by now, it is obvious that AI is much more advanced than expected and is getting even more so all the time.
The thinking about AI in medicine has moved far beyond the naive notion of replacing doctors. Let’s look at what is on the drawing boards.
Enhanced Medical Diagnostics and Imaging
AI is becoming efficient at suggesting medical diagnoses from imaging. Of course, humans should be able to see the same things, but an AI system can examine hundreds of thousands more images than humans can. For the time being, we can expect that the analysis done by the AI will be fed to doctors to confirm their diagnosis, but pinpointing a possible problem quickly will save much time.
Predictive and Personalized Medicine
Personalized medicine refers to treatments that are specifically calibrated for an individual patient, or their specific disease. Work is being done on making personalization work better, and on allowing AI to review all available data on a patient, which it can do in an instant, whereas for a doctor (or medical team), this is a time-consuming process. In addition, depending on the nature of the disease, the AI should be able to determine optimum doses, timing, and combinations of therapies for each individual patient.
Remote Monitoring and Tele-Health
Technologies today, such as the suite of health applications on the Apple Watch, already allow participation in scientific research and constant monitoring of a patient. This capability with AI can be extended to simultaneous monitoring of hundreds of thousands of patients, including those in nursing homes, far beyond the capacity of human healthcare providers.
Expect much more monitoring, but also ongoing coaching of patients, reminding them to exercise, eat the right things, or to take various medications. Yearly there are 34 million hospital admissions, 31 million health center patients, and 139 million emergency room visits.
That adds up to 204 million persons. It sounds like a lot of computation, but AI running in hyper scale data centers easily can monitor this number of patients and do all of its other work without losing a beat.
Clinical Decision Support Systems (CDSS)
In his presentation, Larry Ellison suggested that AI will become an integral member of any medical team and will use its vast computational power to analyze all of the available data on patients in seconds and make recommendations to the medical team. The doctor will be able to carry on a dialogue with the AI and even discuss different scenarios for treatment and obtain the probabilities of success for each.
And of course, depending on the types of questions the doctors ask, or the possible scenarios for treatment suggested, the AI will be able to gauge the skill of the doctor, and keep a record of that also.
In Silico Research Finding New Drugs
My prediction is that the most important use of AI in medicine will be in research and development, aimed at finding new compounds to treat illness.
There is no need to go into the lengthy process. It takes years to test and get a drug accepted and approved. And it takes a very long time to evaluate the ocean of new molecular entities that may hold promise.
Let’s keep that in mind as we list the areas where AI is going to play an essential role. This includes drug discovery, analysis of repurposing of drugs, identification of targets that might prove useful in curing an illness, predicting the toxicity of a compound (without having to run actual tests on humans (or human analogues), and understanding pharmaco-kinetics of a new molecule, including the ADME properties (absorption, distribution, metabolism, and excretion).
The key to this is synthetic data generation and so-called “in-silico” research. In this method, the new molecular entity being explored for medicine is modeled by software. AI then can be used to test interactions between different molecules. In the old way of research, these interactions would be tested in the laboratory, one by one.
With in silico testing, hundreds of thousands of compounds can be tested in seconds. In addition, AI will be able to design new molecular entities that are reasonably able to have a positive effect on mitigating or curing a disease.
The Isomorphic Labs
These notions of using AI for creating new drugs is not theoretical. The Isomorphic Labs have attracted several Noble Laureates, including Jennifer Doudna, who developed CRISPR-Cas9 for genome editing; David MacMillan, for asymmetric organocatalysis; Paul Nurse, for his work on control of the cell cycle; and Venki Ramakrishnan, for study of the structure and function of the ribosome.
The CEO of the labs, Demis Hassabis, who also runs the Google DeepMind project, says that in the past, drug discovery took 5 to 10 years, but this will be accelerated tenfold. He has emphasized protein structure prediction, and use of in-silico research to understand the interaction between various types of biomolecules.
Most importantly, he has predicted that drug design by AI will be in clinical trials by the end of 2025.
AI and Auditing
Of course, AI will also be used in auditing. Auditors will be able to turn out hundreds of times more audits or even audit every single provider.
The world might become an Orwellian nightmare of auditing. RACmonitor has covered this issue extensively. But we must look at the bright side.
Providers can use AI also; in particular, AI can be used to create automated responses to AI audits. It’s AI against AI.
Auditing will become a “war of the machines.” But that is another story.
EDITOR’S NOTE:
The opinions expressed in this article are solely those of the author and do not necessarily represent the views or opinions of MedLearn Media. We provide a platform for diverse perspectives, but the content and opinions expressed herein are the author’s own. MedLearn Media does not endorse or guarantee the accuracy of the information presented. Readers are encouraged to critically evaluate the content and conduct their own research. Any actions taken based on this article are at the reader’s own discretion.