This is part four in a five-part series called "The Limits of Accelerating Returns" that focuses on the limitations of Ray Kurzweil's Law of Accelerating Returns when applied to molecular biology and biomedical technology, including longevity treatments. The other articles in the series are "The Limits of Accelerating Returns," "Biology is not Digital," "Garbage In, Garbage Out," and "Implications of Fixed Returns."
In the previous post in this series, I suggested that, in 40 years, we will be able to run complex molecular simulations of the entire human body if the limiting factor is computational power. Unfortunately, that’s just not the case. We can’t generate the requisite data fast enough for the computers to crunch it. Not everything is amenable to accelerating returns.
Take, for example, a clinical trial. It takes about 8 years for a new drug to go from the first clinical trial to regulatory approval. No amount of computing power is going to change that. It’s a function of the regulatory process. Of course, clinical trials aren’t really going to get in the way of our human simulator.
What slows down the development of the simulator is the rate at which the requisite data is generated. X-ray crystallography, for example, is used to determine the exact shape proteins and other important biological macromolecules. The shape is critical for predicting how such molecules will interact with new drugs and therapies.
The problem is that crystallizing macromolecules is long, slow, frustrating, painstaking work. It’s not an information science, and it’s not amenable to exponential growth. We just have to wait for scientists to do the work, and that will take a very long time.
Similarly, there’s not anything that can be done to make cells grow faster than they do. There’s only so fast that bacteria, liver cells, or any other type of tissue can grow, and we’ve been at that limit for decades. Three-dimensional tissue culture and printed organs have dramatically pushed the envelope of what’s possible, but this is due more to sheer human ingenuity than exponential growth of some information process. Plus, you still have to wait for the cells to grow before you can print a new heart.
The rapidly growing field of bioinformatics is turning biology into more of an information science, and many bioinformatics projects are generating more data than anyone knows what to do with. That’s the rub: raw data is growing exponentially, but useful knowledge is much more constrained. Part of the reason is that the output of a bioinformatics experiment is not useful in itself; it just provides guidance for more detailed experiments. The in-depth experiments are often time-consuming and don’t scale.
Consider genome sequencing. The Human Genome Project cost billions of dollars to produce one sequence. Last year, two more people had their genomes sequenced, at a cost of $1 million each. The $10,000 genome is in sight, and the thousand-dollar genome is not far behind. But just because we know where all the little letters go, doesn’t mean we know what they do. Why is there so much of the genome that doesn’t appear to do anything? What genes are active in a healthy kidney cell? What changes happen with that cell becomes cancerous?
These are just a few of the questions that need to be answered before we can build a human simulator detailed enough to supplant direct experimentation. Getting all of the answers we need is going to take longer than 40 years.
A lot longer.
This has implications beyond our ability to accurately simulate biology. Here’s a standard, Kurzweilian argument for what will happen once someone perfects artificial general intelligence (source):
If anyone can build a full-up, human level AGI, that system will be able to invent new knowledge by itself—new technology, new medicines, and, of course, new types of AGI that can function more quickly than the original AGI. If you can build one AGI that is as smart as the best medical researcher on the planet, you can duplicate that machine with all of its knowledge intact … something that has never been possible with human experts. And if you use AGI systems to develop better AGI systems, you could produce new systems that generate new discoveries much faster than we do. If, for example, the new systems function at one thousand times the speed of a human researcher, this would mean that new discoveries would start arriving at a rate of one thousand years of new science and technology per year.
If it were possible to discover new knowledge by simply thinking about it, then modern scientists wouldn’t need labs. So let’s assume that our army of superintelligent computers has invented themselves up some nice laboratory robots. They still can’t make cells grow faster. They can’t make protein crystallization any easier. It’s true that they remove some barriers to the problem—specifically, there is now an essentially unlimited number of grad students available to throw at the problem.
Except that there are limits to that number as well. Who is building the robotic lab assistants? More robots? Who’s building those robots? What are they building those robots from? How long will those resources last?
More importantly, who’s paying for all of this?
So there are limits to how fast even superintelligent computers can make new discoveries. They may be able to do it faster and cheaper, but they can’t cure cancer overnight. Even if, as given in the example above, they’re a thousand times faster than human researchers, they’re not going to generate a thousand years of knowledge annually. There are other limits—limits that can’t be overcome by being smarter—that will slow them down.
