Amanda Stent was researching natural language processing (NLP) at a time when there were only around 20 people in the world working on it – Stent now heads up all things AI at Bloomberg.
Stent completed a PhD in NLP at the University of Rochester in 2001, focusing on natural language generation “well before natural language generation or generative AI were popular topics”, they tell Computer Weekly.
“Someone quite notable once said to me, ‘You have been doing AI for a very, very, very long time’ – that’s three verys,” Stent jokes.
After completing a PhD, Stent moved into research at AT&T as part of a team focused on speech and language processing, spoken conversation systems and applications of AI.
Research at the time included applying AI to journalism – “computational journalism”, as Stent describes it. “This was early days before generative AI.”
Stent worked on a project that set out to feed structured data into a system, such as stock price changes, to see if it was possible to automatically construct a news story that follows standard journalistic templates, with appropriate visuals such as charts and tables.
After leaving AT&T, Stent took up an applied research role at Yahoo Labs, moving on in 2016, when Yahoo was sold to Verizon, to join Bloomberg as a product manager in the chief technology office, where Stent once again focused on natural language processing.
The AI head then spent three years away from Bloomberg at Colby College – a liberal arts college and undergraduate-focused institution in the US – where they helped found the country’s first undergraduate institute for AI.
Stent says part of their reason for going to Colby College and later returning to Bloomberg – and one of the lessons they learned from working in industry for several years – is that although the world needs AI engineers, it has a greater need for AI-informed product managers, journalists, salespeople and media experts. “It affects everybody today, and we should all be informed and critical users.”
AI affects everybody today, and we should all be informed and critical users Amanda Stent, Bloomberg
Discover, analyse, summarise
Today, Stent is responsible for AI strategy at Bloomberg, which they say has been “leaning into AI for a long time” and was far from a blank canvas when they joined in 2016, having been using the technology since 2009. Back then, the data giant was providing its clients with a sentiment model and sentiment feed. “It was market sentiment, so – not that am I happy or sad about this – it does indicate that the price of something will go up.”
At present, Stent is focusing heavily on generative AI (GenAI). “Our focus these days, like everybody else I think, is on GenAI and how we can make the most effective use of it to solve real client problems.”
Stent says Bloomberg works with three kinds of data: structured data, which is the time series of prices; unstructured data, which includes news, company results and research; and communications data.
“GenAI can really help clients derive actionable insights from those types of data.” Bloomberg is using GenAI today to help clients discover, analyse and summarise its data.
One of the earliest real applications of GenAI at Bloomberg was for earnings transcript summarisation. “We summarise earnings calls – clients don’t have to ask questions, they just get a summary that is done by technology but informed by the subject matter experts that we have at Bloomberg,” says Stent.
“Instead of being a generic summary, we have 13 categories of information that subject matter experts, such as analysts and portfolio managers, are interested in when they listen to an earnings call.
“The client company will see the bullets under the theme, and when they click on it, it will take them to exactly where in the earnings call that information appeared. “This is called transparent attribution, and we want users to check the information from GenAI,” says Stent.
Bloomberg also uses GenAI to provide bulleted summaries of news if the story is more than a certain length. Clients can also ask questions of documents and sets of documents to get information for themes they’re interested in.
Every customer that uses the Bloomberg terminal has access to GenAI, says Stent, and some are building their own solutions as well.
“There are clients who are very technically sophisticated and are really leaning deeply into GenAI and agentic AI, and partnering with us to develop solutions,” says Stent. “And there are clients who are less sophisticated or a little more risk averse, and are kind of waiting to see how it can be done effectively and responsibly.”
Navigating the risks
GenAI will have a profound impact on society, and it is for this reason that experts like Stent need to understand the risks. One such risk is AI hallucinations.
“You have to make sure that your model stays up to date, because in financial services, this is important. You need to make sure that you are identifying the source or the provenance of the information the GenAI is using, because a known risk of GenAI is that it will hallucinate – it will make stuff up and lie through its teeth.”
“You need to make sure you are identifying the source or the provenance of the information the GenAI Is using, because a known risk of GenAI is that it will hallucinate – it will make stuff up and lie through its teeth”
Amanda Stent, Bloomberg
In financial services, making sure the information used to make decisions is accurate is critical.
“Anytime you are making a decision, a human or a system is making a decision, which will affect another human being, I think it’s really important to be as objective as possible,” adds Stent. “Now, computer systems are consistent, but consistent and objective are not necessarily the same thing. Humans are less consistent, but both may be subjective rather than objective.”
The humans behind the machine
There are 350 AI engineers at Bloomberg, but in terms of developing AI, many more people are involved, from a variety of backgrounds, says Stent.
“We also have subject matter experts who provide the data input to AI, and we have AI-informed product managers. It’s a collective team across the company.”
In terms of the skills the organisation is looking for in those who work with AI, it is a mix of people with strong mathematical backgrounds and those who are good at thinking about clients’ problems and identifying solutions.
We can teach people how to be effective users of AI without needing to know all that maths Amanda Stent, Bloomberg
“Our AI engineers might have PhD degrees in physics, computer science and maths, while subject matter experts will probably have graduate degrees in financial services, and potentially backgrounds at a number of clients,” says Stent.
“Sales and product managers may also have backgrounds as Bloomberg clients, and they are really good at identifying why we’re building something and translating the client need to the people who are building it.”
All of these different types of people – such as providence managers, engineers, data scientists, subject matter experts, compliance, risk and legal experts – are “really important to developing solid AI solutions”, says Stent.
At Bloomberg, tech is core, and the creation of more business and jobs is an anticipated outcome of the increased use of AI, but what about wider society and industries?
Stent says: “There has been no revolution in history that has not led to job transformation and more jobs, and I think that’s true here, too. AI will augment a lot of people.”
While there are certain types of tasks people do currently that they won’t need to do anymore, there will be new types of tasks, such as prompt engineering, adds Stent. “We can teach people how to be effective users of AI without needing to know all that maths.”
Jonathan is a tech enthusiast and the mind behind Tech AI Verse. With a passion for artificial intelligence, consumer tech, and emerging innovations, he deliver clear, insightful content to keep readers informed. From cutting-edge gadgets to AI advancements and cryptocurrency trends, Jonathan breaks down complex topics to make technology accessible to all.
