First, banks migrated software applications to clouds hosted by Amazon, Google and Microsoft. Now, they are using large language models that in many cases come from these same companies, sometimes called hyperscalers.
JPMorgan Chase
As they experiment with large language models, banks' vendor shortlists feature these companies and OpenAI, in which Microsoft has invested $13 billion and is entitled to 49% of the profits.
"Most banks — excluding perhaps the biggest tier-one banks with substantial research and development in-house — will be much more focused on the technology providers they work with to build these capabilities, in other words the IBMs, Microsofts, Googles, Oracle and AWSs of the world," said Gilles Ubaghs, a strategic advisor at Datos Insights, in an interview last month.
Some worry that having so much of the financial system running on the servers of a few tech giants increases concentration risk. What if one of these providers has an outage? What if
Add to this the
"Google Cloud supports openness, multi cloud, and the ability for financial firms to freely choose which services and providers best meet their needs," a Google Cloud spokesperson said in a statement. Microsoft and Amazon did not respond to requests for comment by deadline.
Some bankers are thinking about this concentration risk.
"We are constantly exposed to concentration risk due to our reliance on foundational banking technology," said Ryan Hildebrand, chief innovation officer at Bankwell Financial Group in New Canaan, Connecticut, which has $3.2 billion of assets. "Our clients and our operations are dependent on cloud-based platforms such as core banking systems, online banking, loan operating systems, data lakes, API platforms and customer contact solutions."
These systems are prone to disruptions, outages and errors, he noted.
"As we continue to embark on our AI journey, this challenge will become even more significant, particularly in the context of the models that underpin our core operating businesses," Hildebrand said.
Misleading everyone all at once
One concentration risk that is unique to large language models is, could multiple banks be misled by the models' errors, hallucinations or bias along the same lines?
"There's certainly a risk for sure," said Bhavesh Dayalji, chief AI officer for S&P Global and CEO of Kensho, noting that the consolidation of power in AI is still happening. For instance, in March, Microsoft paid Inflection AI $620 million for the nonexclusive right to sell access to the Inflection AI model through its Azure Cloud.
But Dayalji is also encouraged by the work being done in the open source community.
"The pace of this change is really enabling smaller, more dynamic companies to help ensure that there isn't just going to be a few of these killer model providers and you have to be locked into one of those ecosystems, otherwise you can't do anything," Dayalji said.
Chris Tanner, head of research and development at Kensho and lecturer at MIT, pointed out that the same idea of an industry being led in a wrong direction by a widely used model could be applied to search engines.
"Does it make sense for Google to be the de facto search engine for everything?" he said. "Clearly there's some risk with Google being the one main source that everybody gets information from."
The efforts, which respond to President Biden's October executive order on AI, aim to improve synthetic content detection and improve the trustworthiness of AI models.
Like Dayalji, Tanner sees hope in the form of open source communities like Hugging Face.
"Hugging Face has been the one main organization that has allowed so much innovation because of its open source democratization," Tanner said. "I have no reason to believe that that won't continue."
But the popularity of a few open-source models may not be enough to curtail the cloud giants' power.
"Unless open-source AI is pursued to a reckless degree, there will necessarily be concentration of the most advanced AI models to a small number of companies," said Liron Shapira, spokesman for PauseAI, a nonprofit set up earlier this year to mitigate the most catastrophic risks of AI.
PauseAI would like U.S. and international regulators to closely monitor the progress of foundational models like OpenAI's ChatGPT.
"There is an understanding that such models pose a risk of becoming uncontrollable to humanity," Shapira said. "Since we are already used to centralization in the supply chain of computing hardware and cloud providers, centralization of cutting-edge AI models will probably be a familiar experience for customers such as banks."
Concentration of power
Daniel Susskind worries about the impact the cloud monopolies have on democracy. For instance, "the way social media determines who gets a platform and who doesn't, what information we receive, what information we don't receive."
There are basic issues of liberty, says Susskind, who is an economist at Oxford University and author of the book,
"There are various ways in which the technology constrains us, from a driverless car that doesn't go above a certain speed to an electric bike that won't cycle on particular roads, even if you are in an emergency and try to rush to the emergency room for some catastrophe that's befallen your family," Susskind said.
And there are issues of social justice.
"These systems already help determine who can get a loan, who gets a job offer or an interview, who gets parole," Susskind said. "So these technologies in lots of different ways impede on our lives."
One way banks can avoid being controlled by the big providers is by working with several of them.
"Since the beginning of the modern cloud earlier this century, I've always said don't put all your eggs in any cloud provider's basket," said Steve Rubinow, a professor in the Department of Information Technology and Management, College of Computing, at the Illinois Institute of Technology. "Give yourself the flexibility to move around or at least do some of the load here and do some of the load there. And I think that that's exactly what's needed today."
Discover has a hybrid cloud strategy in addition to its own data centers.
"We build all of our applications in a way that gives us the choice as to how we want to deploy them," said Angel Diaz, vice president of technology capabilities and innovation at Discover Financial Services. "Because of that, we're able to build the resiliency, service continuity, disaster recovery, all the things we need to manage that."
The concentration risk of cloud and AI model providers is not the technology itself, argues Yanis Varoufakis, former finance minister of Greece.
"The problem is who owns it and what they use it for," Varoufakis said in a
In Varoufakis' view, governments need to step in and create laws governing these providers. For instance, he recommends a cloud tax of about 5% on the largest cloud companies.
Financial institutions can't strive against the "tsunami" of big cloud providers. "The responsibility, as far as I'm concerned, lies with politics, with our elected representatives," he said.
