Transcription:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.
Penny Crosman (00:01):
Hi, I am Penny Crossman, technology editor at American Banker, and I am here with Derek Waldron, who is Chief Analytics Officer at JPMorganChase. And our topic today is how banks can scale AI. JPMorganChase is obviously a very large bank and you have been deploying AI across the entire workplace. How many people are using it right now?
Derek Waldron (00:30):
So we started about 18 months ago rolling out our generative AI platform. I think that's what we'll speak about today. And over the course of that 18 months, we've gone from zero to about 230,000 users across JPMorgan, which is a little over two-thirds of the entire employee base using that tool today.
Penny Crosman (00:50):
How can you keep customer and bank data safe when you're using large language models across so much data and it's used by so many people?
Derek Waldron (01:01):
Yeah, I think that the safe use of LLM inside the enterprise is a very important topic, and it's important to think about both the enablement of the technology and then how it's used inside the enterprise with regards to enabling the technology these days, LLMs can come into the enterprise into many, many channels. There's the big model providers like OpenAI and anthropic. There are the cloud providers that offer solutions. There's SaaS that provides solutions. There's open source models, and each of these has different risks associated with IT, data security being one of them. So the most important thing to do is to be very aware and deliberate as to how the LLM technology is going to be enabled and brought into the enterprise. And that requires thinking very carefully about the data lineage, what happens to that data, make sure that it's not stored where it shouldn't be. Certainly make sure that models can't be trained on that. And once the enablement is, the enablement channels are determined, everything else should really be shut down. So it's those to make sure that data can't inadvertently go into an enabling technology. It shouldn't. Once the technology is enabled, then there's a question is how it's being used. There's other considerations then for safe and appropriate use inside the enterprise.
Penny Crosman (02:25):
Well, you mentioned not having models trained on your proprietary data, which I know is something all banks are interested in doing. How do you do that? Do you just tell the model builder, model operator not to do it? Or do you have to put a specific gateway in place to protect your data?
Derek Waldron (02:44):
Yeah, these days, many of the big, big model providers do have enterprise offerings, which has technology and contractual provisions to ensure that bank data is kept safe. In addition, the cloud providers have enabled these types of models in their infrastructure. Of course, much bank data goes to enterprise cloud offerings today, and we're used to using that. As long as the security is at the same level for large language technology, I think banks can get very comfortable with that.
Penny Crosman (03:17):
Anthropic's CEO recently said that AI is going to kill 50% of white collar jobs. What did you think of that prediction?
Derek Waldron (03:27):
Yeah, without a doubt, the next generation of AI is going to have widespread workforce implications. But I think specifically how it plays out is quite unclear. It's not the first time in banking that we've had very disruptive types of technologies. Think about the mainframe or the ATM or the internet. All of these were speculated to have wide ramifications on the headcount base, yet we see that employee bases of banks just generally keep getting larger and larger, and the cost income ratio is a little bit more efficient, but it hasn't played out quite the way that people thought it would. A great example is the ATM, which was thought to be eliminating branches, yet at JPMorganChase 25 years after the ATM came out at scale. We have more branch bankers than we did back then, albeit they're doing different things than they were before.
Penny Crosman (04:25):
That's an interesting analogy. And how do you use generative AI in your personal life?
Derek Waldron (04:31):
I try and be a power user in my personal life. I have young kids, so I use it to create stories for them. I use it to push my own thinking as my own coach, my own teacher, my own mentor. And increasingly over time I'm pushing the envelope of the things that I can do with it. So maybe a year and a half ago, it was very much just for synthesis summary and question and answer. Now I'm really using it as a thought partner to help me think through incomplete ideas, to help become a real sort of sparring partner to push my thinking.
Penny Crosman (05:06):
Alright. Well, Derek Waldron, thanks so much for joining us today.
Derek Waldron (05:09):
Thank you.
