Quantum-centric supercomputers to soon be a reality: IBM’s Dario Gil

Do you see quantum computing accelerating development of more foundational (large language and multimodal) models like GPT-4, ChatGPT, etc.?

It’s a nascent field of research. Quantum computers will provide clever and sophisticated new ways to encode information that could be used as a basis to connect it to some form of a neural network structure. But it’s not going to be simple. A quantum computer is not a big data machine. But that doesn’t mean that quantum may not have a big impact on AI. You need to formulate the problem you are trying to solve. That said, AI is power-hungry, and quantum could help.

But what should governments do when regulating foundational models, given the pace at which technology is accelerating?

We’re seeing the elevation of technology as a category to the highest level of economic, social, and national security importance. For nations, it has assumed geopolitical importance. Technology is now co-equal with things like trade or military alliances. But governments don’t control technology. In fact, most of the technology is in the hands of the private sector, and the amount of sophistication and knowledge inside governments is limited.

So, what should the G7 or G20 meetings on technology look like?

Most certainly, they would not just involve government officials —you will need private companies and other institutions, too. That doesn’t currently exist—we will need to invent it.

What about people who are faced with this AI deluge?

The rate of progress on these technologies, like AI, is overwhelming the system, and people are not even able to see the process. So, when you hear a call for a pause (referring to some tech experts calling for a six-month moratorium on developing GPT-4 like language models), it’s because there’s a lack of transparency around the processes. Hence, we need a different level of governance around these technologies. And IBM feels very strongly about this too. Personally, and I’m not trying to speak on behalf of everybody in IBM, I don’t think we should be experimenting on the public. I think that’s a bad idea for technology as a general principle.

IBM unveiled the Osprey processor with 433 quantum bits (qubits) in November and plans to build one with over 4,000 qubits by 2025. What’s the quantum road map?

In classical (used in homes and offices) computers, the processors need to be connected to one another through a communication channel. We’re now building quantum (communication) channels. This entails a radical change in terms of architecture, based on the principle of modularity (combining multiple processors into a single system with a communication link), which will help in developing quantum-centric supercomputing just like we have seen (central processing unit) CPU-centric and (graphics processing unit) GPU-centric supercomputing. Modularity will help us scale these systems and build quantum-centric supercomputers with thousands and tens of thousands of qubits.

Give us some examples of real-world applications of quantum computer-centric supercomputers.

We currently have 225 institutions as members of the global IBM Quantum network. They range from large companies to startups, to national laboratories, and different universities, including IIT-Madras and a startup from India. Boeing, for example, is interested in using quantum computing to develop materials or better composite materials that are less prone to corrosion since they use them for aeroplane wings, etc. ExxonMobil is planning to use quantum computing to accelerate the development of new materials that help the company’s push to provide more renewable energy. Researchers at Daimler hope that quantum computers will help them design next-generation lithium-sulphur batteries.

What is IBM doing to build the global quantum ecosystem, and what are the ingredients of a successful programme?

When we launched the IBM Quantum Experience in 2016, a core part of the reason was to build a community. First was by giving them access to a resource they didn’t have before—a quantum computer on the cloud. But very quickly, we also developed IBM Qiskit, which is by far now the leading open-source environment to develop quantum software (available in four Indian languages). We also partner with countries to create national programmes like those we have done in Japan, Germany, South Korea, Canada, and Spain—and there’s more to come.

Our experience showed that you’ve got to have four dimensions to make a quantum programme successful. First, infrastructure in the form of quantum computers. Second, you need a research and development (R&D) agenda to advance the field and discover new things. Third, you need skills and education and training programmes and certification. And fourth is industrial programmes. We take those four quadrants and methodically implement them in each country. It has proven to be very successful.

Is that model being applied in India, too?

In India, I think we’re in the early stages around that. We’re excited to see that the government is now in its final stages around elevating sort of a quantum strategy for the country (the Indian government allocated ₹8,000 crore towards quantum computing in 2020). We hope we can contribute to making it successful. We’ve begun with elements of it, the partnership with IIT Madras being a good example. At present, the infrastructure piece is through remote access (cloud), which is okay. And we do not yet have an industrial programme. Our desire would be to sort of elevate those seeds and let them grow because we believe that India needs to be a global leader for quantum computing to succeed.

How close are we to fault-tolerant quantum computing?

The ultimate goal is, of course, to build a quantum computer that implements error correction or exhibits fault tolerance. However, the way to get there is like a continuum that is going to be expressed by having machines that progress with lower error rates and higher coherence. Fully implementing error correction is relatively new — it’s a field we’re pioneering right now and getting good traction around it. I think you will see it soon.

At a high level, the way the approach works is that you can characterise or discover the noise in your system. And by modifying the circuit that you are executing and adding extra gates, you are able to suppress the errors that are present. We’ve done that already with 60-plus qubits.

Our 100×100 Challenge has caught a lot of interest and attention, and we have high confidence that we will be able to deliver systems that will include all the software and error mitigation and so on, such that you will be able to write a circuit with 100 qubits and depth-100 gate operations that have been error mitigated. (Quantum algorithms work by applying quantum operations, called quantum gates, on subsets of qubits. Quantum gates can be likened to instructions in a classical computer program. A quantum algorithm represented using gates is called a quantum circuit.). It’s a community-based approach with a very formal challenge.

The concept of quantum teleportation is gaining traction. Are there any real-world applications so far?

Historically, the idea of quantum teleportation has been linked to communications with entangled photons (light particles), putting them over a channel, and they share this entangled property. And now you can actually use that as a means to exchange information. The actual value in terms of technology application, though, is currently debatable. Some people say it’s the basis for creating secure communication channels. We’re believers in using quantum-safe encryption as a mechanism to secure communication channels rather than doing that (quantum teleportation). That being said, an interesting aspect of this vision of quantum teleportation is bringing together the world of computation and communication.