AI and Quantum Computers: Protein Folding, Prions, and the Limits of Machine Learning
Book: Quantum Supremacy: How the Quantum Computer Revolution Will Change Everything Author: Michio Kaku ISBN: 978-0385548366
Chapter 12 is where Kaku connects two big threads from the rest of the book: artificial intelligence and quantum computing. The most interesting part is not the AI history lesson though. It is how both technologies come together to tackle protein folding and brain diseases that we still cannot cure.
AI: From Hype to AI Winter and Back
Kaku starts with the 1956 Dartmouth Conference, where scientists predicted they could crack artificial intelligence in one summer. Many summers later, we are still working on it.
Early AI was built on a top-down approach. You program all the rules of chess, walking, algebra into software and hope the machine becomes smart. Marvin Minsky, the “Father of AI,” was at the center of this. Kaku interviewed him, and Minsky admitted he stopped making predictions about when machines would match humans. Too many times, enthusiasm ran away with reality.
Rodney Brooks from MIT took a different angle. He looked at a fly. It navigates rooms, avoids obstacles, finds food, all with a brain the size of a pinpoint. Our best robots could barely cross a room without falling over. Brooks realized nature does not program creatures with physics equations. Animals learn by bumping into things. Bottom-up, not top-down.
This led to neural networks and deep learning. You show a computer thousands of cat pictures and let it figure out patterns on its own instead of defining cat features mathematically. Google’s AlphaGo beat the world Go champion this way, playing against itself millions of times.
The Commonsense Problem
Deep learning has a serious limitation though. Kaku calls it the “commonsense problem.” Things every four-year-old knows are beyond our best computers:
- Water is wet, not dry
- Mothers are older than their daughters
- Strings can pull but cannot push
You can write down hundreds of these obvious facts in an afternoon. Computers do not understand them because they never experienced the physical world. Children learn these things by touching, falling, bumping into stuff. A computer is a clean slate with zero physical intuition.
Kaku argues the merger of AI and quantum computers could help here. AI brings the ability to learn. Quantum computers bring the raw computational power. Neither is enough alone. A quantum computer does not learn from mistakes, and an AI system on classical hardware runs out of processing power on complex problems. Together, they complement each other.
Protein Folding: Where It Gets Practical
This is the section that grabbed me most. Proteins are the actual workers in your body. DNA is just the blueprint. Proteins digest food, fight germs, build muscle, regulate everything. Their function is determined by their shape. “Function follows form,” as biologists say.
Take the Covid-19 virus. Those spike proteins on its surface are shaped like keys that fit specific locks on lung cells. That shape is why the world economy nearly crashed in 2020-22. Understanding protein shapes is not some abstract academic exercise. It has direct, real-world consequences.
The problem is that figuring out protein shapes by X-ray crystallography is painfully slow. You isolate the protein, crystallize it, shoot X-rays through it, and then try to decipher the resulting mess of dots on photographic film. Months or years per protein.
The CASP competition changed things. Scientists competed to predict protein shapes using computers. The approach borrows from Feynman’s principle of least action: you start with a rough model of amino acids connected in a string, then twist and rearrange bonds to find the lowest energy configuration. Like descending a mountain, taking small steps in whichever direction lowers your height.
Early results were terrible. By 2021 though, DeepMind’s AlphaFold cracked it. They deciphered the rough structure of 350,000 proteins, including all 20,000 proteins in the Human Genome Project. They even announced plans for a database of over 100 million proteins. Despite heavy approximations in their calculations, the results matched X-ray crystallography data surprisingly well.
AlphaFold solved the “easy” part though. Knowing the shape is step one. Understanding how that shape determines what the protein actually does, and then designing new proteins with specific functions, requires full quantum mechanical simulation without approximations. Only quantum computers can do that.
Prions and Brain Diseases
This is where the chapter gets personal. Kaku writes about losing his mother to Alzheimer’s. The science here is genuinely sobering.
Prions are proteins that folded the wrong way. Stanley Prusiner discovered them in 1982 and won the Nobel Prize in 1997. A prion spreads by forcing normal proteins to misfold when they come in contact. No bacteria, no virus. Just a misshapen protein causing a chain reaction.
Scientists now believe prions may be behind Alzheimer’s, Parkinson’s, and ALS. For Alzheimer’s, sticky amyloid proteins (beta and tau) build up in the brain. In 2019, German scientists found that people with misfolded amyloid proteins in their blood were twenty-three times more likely to develop Alzheimer’s, detectable up to fourteen years before symptoms appeared. Prusiner himself called Alzheimer’s a “double-prion disorder.”
In 2021, researchers at UC found that “good” and “bad” amyloid proteins differ in handedness. Normal amyloid spirals left-handed. The Alzheimer’s-associated version spirals right-handed. If this holds up, quantum computers could model the exact atomic structure of both versions and figure out how the bad one propagates and damages neurons.
ALS follows a similar pattern. The SOD1 gene creates a protein that normally breaks down dangerous superoxide radicals. When that protein misfolds, motor neurons die. Stephen Hawking lived with ALS for decades, which was exceptional. Most people survive only two to five years after diagnosis.
Parkinson’s disease, affecting about one million Americans, also involves mutated proteins in the brain. Scientists can already locate where neurons overfire using brain scans and partially treat tremors by targeting those spots. No cure exists though. The hope is that quantum computers can model the defective proteins well enough to design molecules that fix or neutralize them.
My Take
Kaku covers a lot of ground in this chapter. The AI history section feels like a recap if you have read anything about machine learning in the last decade. The protein folding and prion sections are genuinely excellent though.
The key insight is practical. AlphaFold solved protein folding with approximations on classical hardware. To go from knowing shapes to actually designing drugs and cures for Alzheimer’s, ALS, and Parkinson’s, you need to simulate full quantum mechanics of molecules. That is the quantum computer use case that feels most real and most urgent.
Whether quantum computers will deliver on this promise in our lifetime is still an open question. The direction is clear though, and the stakes are as high as they get.