photobank.kiev.ua - Fotolia
Shawn Edwards, global chief technology officer (CTO) at Bloomberg, became enthralled by science and technology as a boy when he received a copy of Cosmos by astronomer Carl Sagan. But in adulthood, it is software, not astronomy, that pays the bills.
While Edwards started out stargazing, the secrets of the universe, through quantum computing, are set to change the world of computing. “Understanding the basic physics of quantum and talking to enough people, I think that when it does work, quantum computing will be a game-changer,” he says.
However, it is not practical today. “It is something we keep an eye on and it is something that will be real at some point. There is enough basic science behind it – and that’s exciting,” says Edwards.
During most conversations with a CTO, at some point the question of artificial intelligence (AI) is raised. This time is no different, as Edwards describes how and his team are looking into how AI can be applied at Bloomberg.
“Bloomberg has vast amounts of data and we have built incredibly powerful models,” he says. “These are really powerful for extracting data out of financial documents, but would not help in analysing consumer sentiment on an e-commerce site.”
Asked whether today is the age of AI, Edwards says: “There is a big debate on whether we are hitting another plateau. Are we going beyond neural nets? For years there was a lot of science and a lot of belief, but unfortunately there was also a lot of hype.
“We see a lot of practical value in AI. Today, you are still challenged to manually label data. But people can do things now that they couldn’t do before.”
Google, Amazon and the like have been pushing the limits of mainstream machine learning and AI, but one company that has been surfing the AI wave is Nvidia, the graphics card company associated with rendering life-like scenes in computer games.
If quantum computing is set to change computing in the future, Nvidia rewrote the rulebook a decade ago with a way to run supercomputer workloads on graphics chips. This approach is cementing the company’s future in the age of AI to power new machine learning algorithms for applications such as self-driving cars.
Shawn Edwards, Bloomberg
Discussing how Nvidia’s share price has skyrocketed recently, Edwards describes his first experiences of graphics processing units (GPUs), the brains inside a graphics card: “About 10 years ago, our quants [quantitative analysts] came up with a new model for calculating asset-backed security covering risk and pricing on a mortgage or car loan. It required running lot more Monte Carlo simulations, which would require 10 times as many Linux boxes.”
A member of Edwards’ team proposed using a GPU to solve the problem. “I gave him a budget and he went off and worked with Nvidia, and we got one of the first double precision floating point GPU cards,” says Edwards.
“He actually built a prototype that blew us away and we ended up building one of the largest GPU farms of the time just to price asset-backed securities. Now GPUs are everywhere and we use GPU farms for machine learning and we actually use Nvidia products.”
With GPUs, Edwards says people are now building more customised architectures, such as the work Google is doing in its datacentres. “When you look at the IoT [internet of things] and speak to the Googles of this world, they push this technology to the phone. The GPU on the phone is really quite powerful, and it may do some processing before pushing up to the cloud.”
Role of a CTO
Just as in the Nvidia example, Edwards says the role of Bloomberg’s CTO Office is to assess new and cutting-edge technologies. “We are a small group of researchers who focus on a couple of strategic areas, where we feel there is a real need for investment or experimentation,” he says.
The CTO team works closely with Bloomberg’s engineering and product teams, says Edwards. “We carve out little teams who build proof of concepts. The idea is to try things out, build stuff in the lab, and work with academia, the open source community and vendors to build a proof of concept that either dies quickly or is a good idea.”
As and when something becomes a good idea, the role of the CTO’s Office is then to socialise it, says Edwards. “We say this is the technology that will help us solve a problem – this is where we build it out. We then switch hats and become the product owner of that new technology until it becomes mainstream.”
As an example, Edwards says he expects GPU platforms to evolve to become more optimised for machine learning.
As for blockchain, he says: “That is a field with a lot of promise, but with very little delivery so far. I have a few experts and we are observing. We talk to tons of startups and a number of consortiums to assess what is the application of a blockchain.”
This assessment involves boiling blockchain down to its fundamental technologies, says Edwards. “Blockchain has certain qualities. It is decentralised; you work in a non-trusted environment, which means you need verification and proof. Where do I need those combinations of properties? The answer is where I would use blockchain.
“It should not be used in any old case just to say you are using blockchain, because relational databases and cloud services work really well.”
Read more CTO interviews
- Cyber security technology innovator and veteran Steve Grobman shares his views on adversarial artificial intelligence, post-quantum cryptography and security for next-generation technology.
- The head of Hewlett-Packard Labs discusses a new era of computing, where memory is no longer a constrained resource.
It is the trajectory of computer science that technology evolves over time to become more abstracted and easier to use. It is no longer necessary for GPU programmers to master Cuda, the programming libraries needed to make the most of the Nvidia hardware. Instead, they can use higher-level programming libraries. This raises the question of whether low-level coding is still relevant. Edwards believes it is.
“You will still need someone to create the next algorithm, like Tensorflow [Google’s object recognition library], but I love the fact that people are working to make these things easier to use, which makes the technology accessible to more people,” he says. “There are powerful abstractions, so we don’t have to spend time learning low-level code.
“Someone can take public APIs [application programming interfaces] and focus on the business problems we are trying to solve by building something that is novel or useful to the world, leveraging the low-level code.”
Experts are still needed, especially people who understand the mathematics to build really complex machine learning algorithms, to understand how to identify bad data and how to tune data models, he says.
Beta testing in production
According to Edwards, Bloomberg tries to be first to market, even if the products it releases are not complete.
“While you can’t give a trading application wrong numbers, if you have a new search or analytics capability, you may not have shipped all the analysis or all the questions that may be asked of that instrument, but you’ll have the first set,” he says.
“Let’s get that to market, let’s get some use and then continue to build. We try not to make being perfect the enemy of being good. Bloomberg has long tried to get things out that are useful – not bad, not wrong – and then polish them.”
Edwards is also a strong believer in trying out a new concept and proving it. “There is a lot of hype in the technology industry,” he says. “Part of our job is saying no and figuring out what’s not useful. The other half is about proving something is useful.”
Finding new technologies that could be deployed at Bloomberg is the main role of the CTO Office, and this requires a certain mindset.
Bloomberg has a unique culture, says Edwards. “Not everyone will thrive here. Not everyone belongs here. But I pride myself in building really great teams and I want to surround myself with people smarter than me in all the different technology fields,” he says. “My job is to help them understand the bigger picture and setting the direction.”
For Edwards, it is not only important to hire the right people, but also to mentor and coach them. “We have more good ideas than we can work on,” he says. “You may think you have the best idea, but for one reason or another, it doesn’t make sense.
“So sometimes, someone’s favourite idea gets pushed down near the bottom and we have to accept that. But I think people love to work here because the ideas they do get to work on have a huge impact on the company and our clients.”