Recently, we’ve been looking at tech philosopher George Gilder’s new Gaming AI about what AI can—and can’t—do for us. It can’t do our thinking for us but it can do many jobs we don’t even try because no human being has enough time or patience to motor through all the calculations.
Which brings us to the massive complexity of the proteins that carry out our genetic instructions—better knowledge of which would help us battle many diseases.
Gilder notes that when DeepMind’s AlphaGo beat humans at the board game Go in 2016, it wasn’t just for the fun of winning a game. DeepMind cofounder Demis Hassabis (pictured in 2018) is more interested in real-life uses such as medical research (p. 11). The human body is very complex and a researcher can be confronted with thousands of possibilities. Which ones matter?
The area the DeepMind team decided to focus on is protein folding: Human DNA has 64 codons that program little machines in our cells (ribosomes) to create specific proteins out of the standard twenty amino acids. But, to do their jobs, the proteins fold themselves into many, many different shapes. Figuring it all out is a real problem for researchers and the DeepMind crew hope that AI will help:
Over the past five decades, researchers have been able to determine shapes of proteins in labs using experimental techniques like cryo-electron microscopy, nuclear magnetic resonance and X-ray crystallography, but each method depends on a lot of trial and error, which can take years of work, and cost tens or hundreds of thousands of dollars per protein structure. This is why biologists are turning to AI methods as an alternative to this long and laborious process for difficult proteins. The ability to predict a protein’s shape computationally from its genetic code alone—rather than determining it through costly experimentation—could help accelerate research.Andrew Senior et al., AlphaFold: Using AI for scientific discovery Deep Mind blog (January 15, 2020)
As Gilder recounts, the biotech industry conducts annual global protein-folding competitions among molecular biologists and in 2019 DeepMind defeated all teams of relatively unaided human rivals:
Advancing from the unaided human level of two or three correct protein configurations out of forty, DeepMind calculated some thirty-three correct solutions out of forty. This spectacular advance opens the way to major biotech gains in custom-built protein molecules adapted to particular people with particular needs or diseases. It is the most significant biotech invention since the complementary CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) method for using enzymes directly to edit strands of DNA.George Gilder, Gaming AI (p. 15)
But now that we have found a way to tackle one aspect of the immense complexity of human bodily existence, here’s an interesting problem to think about: We are told by many philosophers that life came to exist on Earth purely by chance. How likely is that, given the intricacy of the machinery that governs our bodies?
As we all know from probabilities, you can get lucky once, but not thousands of times…
As real data shows, the probability of finding a functional sequence for one average protein family is so low, there is virtually zero chance of obtaining it anywhere in this universe over its entire history — never mind finding thousands of protein families.Kirk Durston, “Computing the “Best Case” Probability of Proteins from Actual Data, and Falsifying a Prediction of Darwinism” at Evolution News and Science Today: (July 28, 2015)
Yet that’s what we have. All those protein families. As we learn more about the world we live in, we may find ourselves confronting more challenges like this: We had to invent a really complex machine to even begin to figure out protein folding in our bodies —and we know that the machine did not happen by chance. So why should we believe that our bodies happened that way? Probably not.
Note: While medicine may be the most important way AI can help us, it also helps us in other areas where huge numbers of calculations are essential for success. For example, it can help recover lost languages and interpret charred scrolls. It can continuously scan the skies, sparing astronomers for more human-friendly work like interpreting the results. It can restore blurred images and help with cold case files. As with anything, the trick is to take advantage of what it can really do. We don’t need the courtroom sentencing robot or the AI Jesus—but then we never did. As our information resources become larger and more complex, we do need some help with the sheer volume and that’s where AI is bound to succeed.
You may also enjoy:
Why AI geniuses haven’t created true thinking machines. The problems have been hinting at themselves all along. Quantum computers play by the same rules as digital ones: Meaningful information still requires an interpreter (observer) to relate the map to the territory.