A big part of the appeal of the new technology for consumers, who have started using many generative AI tools like ChatGPT, is its ease of use.
However, many companies need to use generative AI tools such as these carefully to avoid regulatory and compliance issues.
“There are certain management functions where it makes sense to involve generative AI, and there are certain management functions with more deliberate decision-making, more discriminatory decision-making…sometimes we want to be more conscious.” said. Her Regina Sam Penti, a partner at Ropes & Gray law firm, said: MIT Technology ReviewEmTech Digital 2023 conference on May 2nd.
Highly regulated industries such as finance and healthcare have long exercised restraint when using AI tools and technologies.
Organizations using AI technology have also recently dealt with heightened regulatory activity by the Federal Trade Commission (FTC). Improper use of AI.
While most existing laws and regulations do not explicitly address AI biases and other issues, the FTC has identified biased algorithms and unexplained practices in handling consumer credit, jobs, housing, and insurance. It targets some organizations that use algorithms that do not exist.
Representatives from JP Morgan Chase and law firm Ropes & Gray discuss the challenges of generative AI.
Challenge to regulated industry
Given the tightening regulatory environment, it is not surprising that companies in highly regulated industries have been slow to incorporate generative AI tools such as Microsoft partner OpenAI’s ChatGPT and Google’s Bard into their systems. That’s it.
However, even medical and financial institutions that work with large volumes of sensitive data cannot ignore the excitement and popularity of these large-scale language models (LLMs).
For example, ChatGPT has already reached over 100 million active users, surpassing other innovative AI technologies to date. Some people may want to adopt technology quickly, but faster is not always better because technology has potential problems.
Organizations in highly regulated industries should consider using the tool in a controlled environment so they can better mitigate risk, said Penti.
Still, there are many issues with generative AI that have prevented certain industries from implementing it in some way immediately.
For example, models tend to be non-deterministic and are likely to produce different results for different questions or queries, which can put risk in the decision-making process. The model is also updated only sporadically. For example, ChatGPT has only been updated for 2021 yet.
Additionally, generative AI systems are often wrong and can spew out false information. Many users have tried to address these issues with rapid engineering (attempts to fine-tune data inputs to questions and models), but hackers are already exploiting it. Finally, cost is an important factor for all businesses and organizations, not just highly regulated industries.
For example, ChatGPT is available on Microsoft Azure, so those considering it would need to integrate with Azure, which can be costly.
JP Morgan Chase
The approach of large multinational financial services companies such as JPMorgan Chase & Co., the world’s largest bank, is to slowly implement LLM in areas where the risks are less pressing.
We are 100% crawling, and we may not be crawling yet.
Brian MaHead of Product, Enterprise AI and ML Platform at JPMorgan Chase & Co.
“Our approach is just ‘crawl, walk, run,’” said Brian Maher, product lead for JPMorgan Chase’s company-wide AI and machine learning (ML) platform, during the same EmTech panel discussion. increase.
“We are 100% crawling and we may not be crawling yet,” he said.
Financial giants look for applications where LLM can be used with low risk and minimal impact on customers and the company. Also consider whether the data your LLM is using is low-risk or internal sensitive data, Maher said.
“We have a safe learning environment,” he said. “We don’t have enough knowledge to publish this as a publicly available tool.”
Maher said JPMorgan Chase is taking a slow-paced approach to addressing the challenges of using generative AI. In the meantime, the company is keeping a close eye on its technology.
“If we use this technology in our company, it must be highly registered,” says Maher. “It has to be monitored.”
This oversight means putting people in the loop to constantly give feedback on whether the model is working.
The need for such a high level of oversight is due to the nature of generative AI models compared to traditional AI models. Maher says the risk of generative AI models lies in how they are used, not how they are built.
And with little visibility into how the model works, regulators like the FTC could pose problems for companies like JPMorgan Chase.
“I think the monitoring of the models that we need to do on a regular basis will be greatly enhanced,” Maher said. “There are so many unknowns because it’s really hard to explain.”
As an example of upcoming regulatory changes, the U.S. Securities and Exchange Commission (SEC) has commented on new rules prohibiting investment advisors from outsourcing certain services or functions without conducting due diligence and oversight of service providers. I am considering.
“I’m not necessarily naming AI, but it’s mixed,” Penty said. “These are some of the most specific rules I have seen…the SEC requires asset managers and investment advisors to impose certain conditions to ensure that your provider provides services. We’ve come up with it in terms of requiring it to be included in the contract, which is necessary to meet the customer’s request.”
Dealing with such rules and regulations means a cautious approach for the financial industry.
Another way JPMorgan Chase should pay attention is to ensure that AI models are explainable, Maher said.
“At JPMorgan Chase, as a financial provider, we have a responsibility to be transparent with our regulators, stakeholders, shareholders and customers about how we are doing this,” he said.
The two-day conference was held in person and virtually.
Esther Ajao is a news writer covering artificial intelligence software and systems.
