Jamie Dimon’s annual letter to shareholders, published this week, would indicate that if the JPMorgan Chase CEO wasn’t a believer in the potential for artificial intelligence and machine learning to improve banking, he is now.

“The power of artificial intelligence and machine learning is real,” he declared.

Noting that the bank recently sent a senior team to China “to study what’s being achieved there with artificial intelligence (AI) and fintech,” he said it was hard not to be impressed and worried about the progress made. “It made our management team even more motivated to move quickly,” he said.

Dimon said the bank is “all in” on cloud and AI, but he also admitted partial responsibility for the bank’s slow adoption of cloud as his “early thinking about the cloud was that it was just another term for outsourcing.” Still, he touted the potential for cloud to give JPMorgan an ability to achieve scale and “elasticity of computing power exponentially” beyond the bank’s existing capacity, particularly when it comes to AI and ML.

“These technologies already are helping us reduce risk and fraud, upgrade customer service, improve underwriting and enhance marketing across the firm,” he said. “And this is just the beginning. As our management teams get better at understanding the power of AI and machine learning, these tools are rapidly being deployed across virtually everything we do. We can also use artificial intelligence to try to achieve certain desired outcomes, such as making mortgages even more available to minorities.”

Among other examples of how JPMorgan is using AI and ML, Dimon said the corporate and investment bank has a product called DeepX, which “leverages machine learning to assist our equities algorithms globally to execute transactions across 1,300 stocks a day.” DeepX is being rolled out to new countries, he added.

He also said the bank will be deploying AI-driven virtual assistants across the board to handle tasks such as maintaining internal help desks, tracking down errors and routing inquiries.

Additionally, Dimon said initial results from ML fraud applications “are expected to drive approximately $150 million of annual benefits and countless efficiencies,” and that, over time, AI will also “dramatically improve Anti-Money Laundering/Bank Secrecy Act protocols and processes as well as other complex compliance requirements.”

As a footnote to all of this, he said the bank will “try to retrain and redeploy our workforce as AI reduces certain types of jobs.”

“We are evaluating all of our jobs to determine which are most susceptible to being lost through AI,” Dimon continued. “We will plan ahead so we can retrain or deploy our employees both for other roles inside the company and, if necessary, outside the company.”

JPMorgan is spending about $11.5 billion on technology annually and employs about 50,000 IT workers. The bank also recently awarded 47 financial grants to university faculty and PhD students for AI research.

In the meantime, AI champions, fence-straddlers and skeptics abound in financial services.

Farhan Ahmad, founder and CEO of B2B payment solutions firm Bento for Business, recently told Bank Innovation that “nobody is doing AI, everyone is doing machine learning and calling it AI, and that’s fine.”

The two terms are similar but not exactly interchangeable. Machine learning is generally considered to be when a computer uses data to make predictions and generate insights. AI, in the meantime, is really the broader concept of machines being able to carry out tasks in a way people might consider “smart,” and without the need for the underlying data used in machine learning.

“We do some machine learning, too,” Ahmad said. “I’m very realistic about what is AI and what is machine learning, but data is very predictive and always has been.”

He said traditional card firms, for instance, are really data firms and have long been able to predict (with a fair amount of accuracy) what people are going to buy and when.

Ahmad said AI is certainly on Bento’s roadmap, as more connection points are made and more data streams become available. He said AI will eventually help companies solve problems with insights specific to their individual business rather than applying general industry insights, and without human intervention.

Gal Krubiner, co-founder and CEO of Pagaya, a fintech specializing in AI-powered asset management, recently emphasized to Bank Innovation that he is against combining AI and human beings. “We’re not a big believer in AI enabling people or in any sort of combination of the human being with machines because it’s understood that they are 10 times stronger than us when they are being used right,” he said.

Krubiner pointed to the number of parameters and data points that go into the typical loan application. “No matter how much time I give you as a credit officer, your ability to price that risk, with 1,500 parameters and data points, is almost meaningless because the machine can do it very well,” he said.

“These companies are trying to adopt AI as an add-on and don’t understand that this is from scratch,” he said, calling this attitude toward AI the biggest hurdle for the technology in banking.

Additionally, Krubiner said there is little to no institutional expertise as “basic” AI is transforming into a more “robust, deep-dive” AI. “If you’re good at tuning a car, that doesn’t mean you’re good at tuning an airplane, even though you’re both mechanics,” he said.