Quantum Leaps

Eric Ostby, PhD (MS ’04, PhD ’09) on the current state of quantum computing—and the shape of things to come

 

by Maureen Harmon

Photograph by Nadia Tyson

Though Eric Ostby, PhD (MS ’04, PhD ’09) was aware of the concept of quantum computing in his undergrad years and in his years at Caltech, it wasn’t a field he planned to pursue. “I thought quantum computing was a fascinating field for the future,” says Ostby, “but not something that was applicable today.”

He changed his mind when he joined Google as a product manager in its quantum computing group. It was there that he became part of a seminal moment in the field when his team announced in 2019 that they had achieved “quantum supremacy,” meaning that they had created a machine that could do a well-defined task much faster than a classical computer.

Today, Ostby is vice president for product development at Rigetti Computing, a company that develops systems and hardware to support quantum computing—helping usher in a tech revolution that once looked fantastical. In this conversation with Techer, he discusses the innovations that quantum computing could make possible, and what it will take to get there.

How do you explain your job to your young children?

Eric Ostby: I tell them it all just comes down to math. A classical computer is really a tiny processor with the capacity to do relatively basic math—and that math allows my daughter to watch videos on a phone. A quantum computer can do much more complex math, and it can do it fast. Let’s say, for example, you want to figure out how many square feet you have in a room. One option is to do a lot of addition, I might tell the kids. You could take a one-foot square tile and go around the room and add up all the squares it would take to fill the room. Think of that as classical computing.

But we can solve that square-foot problem much more quickly with an advanced form of math: multiplication. Four tiles down the room and six across is 24 feet.

Quantum computing is using advanced math so we can take a shortcut to the answer. A quantum computer could help us figure out the best way to get to Mars or it could help us design a drug to address some types of cancer. I tell my kids that I’m working to build those machines.

You talk a lot about the quantum computers’ ability to solve problems that don’t seem solvable today. Give us an example.

EO: Today we can’t simulate the chemistry of even small molecules. We can do methane, we can do helium, but we can’t simulate larger molecules over 50 valence electrons. The classical techniques break down when the molecules get more complex. So chemical companies today have to do a lot of approximate calculations and thousands of experiments every day on a small scale. They’re constantly testing because they don’t have efficient ways to get to the answers.

But with quantum computing, there’s promise that you could actually do those simulations and computations without needing to run thousands of experiments. So, for biopharma, quantum computing holds the promise for accelerated drug development.

“Classical computation is like a coin. You have heads or tails. (Think ones and zeros.) And a coin sitting on a table is a very stable thing. You put a coin heads up on a table, and it stays there; it doesn’t change. But take that same coin and spin it on the table: that’s quantum computing.”

Eric Ostby in front of ORNL’s Summit supercomputer while at Rigetti.

‍What’s one of the biggest challenges we face with quantum computing?

EO: Classical computation is like a coin. You have heads or tails. (Think ones and zeros.) And a coin sitting on a table is a very stable thing. You put a coin heads up on a table and it stays there; it doesn’t change. But take that same coin and spin it on the table: that’s quantum computing. It’s fast, and it’s a mixture of heads and tails (ones and zeros) existing at the same time. The problem is, the coin will eventually stop spinning—it can’t hold that state forever.

So one of the challenges we face is creating more stable states for qubits—the unit of information in quantum computing that is similar to a bit in classical computing—so we can run longer computations to solve more challenging problems. Think of it this way: Classical computers are like the family car. They can run around a racetrack for hours and days without breaking down. A quantum computer is like a very fast race car, a super car, but it can only make it halfway around the track before it breaks down. We want to get that super car to be able to go around the track as long as the family car—and then we can really go far in that race.

Everyone seems to be talking about quantum computing, but we’re still early in the field’s history.

EO: It’s not early days, exactly; I’d say we’re getting into the Middle Ages. We have functional machines with limitations, and now we’re working to improve their performance. If the field were a 20-chapter book, I think we’re at least on chapter five.

How do the worlds of AI and quantum computing overlap?

EO: I’m actually getting that question a lot, even from my family. Technologies like ChatGPT have created models that are trained on an enormous amount of data. For example, ChatGPT basically downloaded all of Twitter for a couple years—all those conversations and comments. Then it scanned books, newspaper articles, and encyclopedias. It’s trained on more information than you or I could read in our lifetimes and it’s able to predict text by matching patterns that it has seen in the past. It can seem very human because it’s basically regurgitating written human speech. But quantum computers are not large-data solutions. This is a common misconception in the field.

Even so, quantum computers could be useful for AI in generating new data that doesn’t exist. One of the things that I’m working on at Rigetti is called generative quantum machine learning. This is an area of quantum computing where we can look at problems that have sparse data sets which we consider “rare events.” One example might be the meteorite that wiped out the dinosaurs. We don’t have a lot of data from that event, so machine learning isn’t going to work very well because it has very limited data for training. But a quantum computer can generate missing data based on what we already know.

There are some theories in the field that a quantum computer works more like the human brain, which is essentially a learning model. As humans, we have experiences and we have memories, but we also form connections between those experiences and memories. This is a neural net and the idea is that a quantum computer could function as a neural net. I think it’s an interesting area of long-term research to say, could we build a neural net with a quantum computer that’s more human-like than what we have now? We haven’t been very successful at producing something that models the human brain with classical computing. Tesla’s autopilot is good, but it’s nothing compared to a human driver.