Goldman Sachs GS AI Assistant
Courtesy: Goldman Sachs
Goldman Sachs is introducing a generative AI assistant to its bankers, traders and asset managers, the first phase in the evolution of a program that will eventually take on the traits of a seasoned Goldman employee, according to Chief Information Officer Marco Argenti.
The bank has so far released a program called GS AI Assistant to about 10,000 employees, with the goal that all of the company's knowledge workers will have it this year, Argenti told CNBC in an exclusive interview. It will initially help with tasks such as summarizing or proofreading emails or translating code from one language to another.
“Think about all the tasks you would like to perform across a variety of use cases for all those professions you now have at your fingertips,” Argenti said. The Goldman Assistant is a “very simple interface that gives you access to the latest and greatest models.”
Goldman's move means that, along with JPMorgan Chase And Morgan Stanleythe world's three largest investment banks have aggressively released generative AI tools to their workforces, a notable development since ChatGPT went viral about two years ago.
Wall Street has embraced generative artificial intelligence faster than any other disruptive technology in recent years, experts say, because of the adeptness of large language models at replicating aspects of human cognition.
Today it can respond to queries, write emails and summarize long documents, but expectations are high that future versions will exhibit so-called 'agentic' skills, meaning they can perform multi-step tasks with little human intervention.
Speaking to CNBC about his vision for artificial intelligence at the company, Argenti – who joined Amazon in 2019 – repeatedly compared the AI program to a new hire who will absorb the Goldman culture in the coming years.
Initially, the tool will mainly produce answers based on Goldman data fed into AI models from OpenAI's ChatGPT. Google's Gemini and Metas Lama, depending on the task, Argenti said. The bank is also looking at models from companies such as Anthropic, Mistral and Cohere, he added.
“The AI assistant really feels like talking to another GS employee,” Argenti said.
Learning the Goldman way
“As we move forward, the second step is when you start to exhibit this agent behavior, which is, 'I'm completing a task on behalf of a Goldman employee and I have to take a series of steps,'” he said. “That's where the model starts doing things like a Goldman employee, not just saying things like a Goldman employee.”
This helps explain why companies have banned employees from using ChatGPT for work, and instead started creating their own platforms to leverage the technology. It allows companies to not only keep their information safe, but also create AI platforms that increasingly resemble the best examples of their own workforce.
“It is extremely important that the AI has a very specific identity that reflects the principles, values, knowledge and way of thinking of the company,” Argenti said.
In practice, this means that just as an experienced Goldman employee has to double-check his work with multiple data sources or use a specific algorithm for a calculation, the AI will absorb those lessons, he said.
Marco Argenti, chief information officer of Goldman Sachs, joined the bank from Amazon in 2019.
Courtesy: Goldman Sachs
But Argenti says he's most excited about the prospect of what will happen later, perhaps three to five years from now, when AI models continue to blur the lines between human and machine thinking.
At this stage of AI at Goldman, the model would “actually do more reasoning and become more like the way a Goldman employee would think,” he said.
So instead of being given a runbook, which is tech industry parlance for a set of step-by-step instructions for completing tasks or responding to incidents, the AI could generate detailed plans “in the way an experienced Goldman employee would do,” Argenti said.
Risk of disruption
The prospects for that future — and the fact that Wall Street workers are helping to train a technology that could make some positions obsolete while expanding others and creating entirely new roles — could send a new wave of fear through the ranks of make the employees flow.
Like Goldman, other major investment banks are on track to give generative AI tools to their entire workforces in the coming months.
More than 200,000 JPMorgan employees currently have access to internal generative AI tools, according to a person with knowledge of that bank who declined to be identified speaking about internal matters. About 40,000 Morgan Stanley employees had access to it at the end of last year, the bank said in October.
Finance and technology are seen as sectors where workers are most susceptible to unrest due to generative AI, potentially allowing companies to generate billions of dollars in additional profits. Meta CEO Mark Zuckerberg told podcaster Joe Rogan earlier this month that the AI will be able to write code as well as mid-level software engineers this year.
According to a report from Bloomberg's research department, global investment banks could cut as many as 200,000 jobs over the next three to five years as the companies implement AI. The report, based on a survey of tech executives at major banks, says that support and operational functions, known as the back and middle office, are most at risk.
At Goldman, however, the official position is that AI will enable employees to do more, which won't necessarily result in the need for fewer people.
“The importance of a phenomenal human workforce will actually become even greater,” Argenti said.
“In my opinion, it always comes down to people,” he said. “People are going to make a difference because people are going to be the ones who actually develop the AI, train the AI, strengthen the AI and then take action.”