Is Generative AI Augmenting Our Jobs, or About to Take Them?
“Generative AI is one of the best augmentation means — not substitution, augmentation. Augmenting the professions that actually exist. Augmenting an auditor, augmenting a tax advisor, augmenting a legal expert, augmenting a strategic adviser.”
So says Nitin Mittal, who is leading a new generative AI practice within Deloitte Consulting, when I asked him how generative AI is impacting white collar workers the world over. His point was that AI isn’t replacing those jobs, but providing an augmentation layer.
Others, however, are starting to argue differently — most notably, AI pioneer Geoffrey Hinton, who told the New York Times that while AI “takes away the drudge work,” soon “it might take away more than that.”
Hinton has just left his executive job at Google in order to talk more about what he terms “the dangers of AI.” Another tech leader who recently hinted at the impact of AI on the job market is IBM CEO Arvind Krishna, who reportedly told Bloomberg that about 30% of his staff could “easily” be replaced by AI in five years.
To be fair, Mittal himself recognizes that AI might not be purely augmentative in the coming years.
“Over time,” he said, “is there a possibility that we are probably going to see some substitution? Absolutely.”
But he noted that once “substitution” starts happening — once AI starts taking over particular skill sets “lock, stock and barrel,” as he put it — it will need to be certified and regulated to protect consumers.
Regardless, substitution is not what Deloitte’s customers are asking about right now. The current focus for Mittal’s new generative AI practice is purely on augmentation. He told me there’s been a sudden demand this year for GenAI help from Deloitte’s clients, and the new practice which he leads was created (in just two months) to meet that demand.
Deloitte isn’t the only one of the “big four” business consulting firms to jump on the generative AI trend. Last week, the Wall St Journal reported that PWC “plans to invest $1 billion in generative artificial intelligence technology in its U.S. operations over the next three years.”
The Impact on IT Departments
One of the immediate impacts of generative AI in the enterprise is through the IT department. Not only are developers now commonly using AI-assisted coding tools like GitHub Copilot, but IT is facing a potentially seismic disruption to the tech stack.
Mittal believes that generative AI is going to result in a fundamental restructuring of IT. He explained that IT departments will need to gear up for the new field of LLMOps — the operational side of large language models (LLMs). This, he said, will involve continuously refreshing data sets, intuitively training models, and generating the intended modality.
He also emphasized the importance of putting guardrails in place to mitigate the risks associated with LLMs.
In a broader sense, he thinks the IT department will have to de-focus from its traditional role of implementing systems and building a workforce around those systems. Instead, IT will need to focus on “building models at a rapid pace, deploying these models, training these models, and consequently getting into the business of LLMOps to maintain these models throughout the organization — because they’re inevitably going to proliferate across every business function, workflow, process and customer interaction associated with that company.”
The Middle Layer
I noted that some business intelligence and data intelligence companies argue that the data that LLMs ingest will need to be cleaned and prepared — that it’s “garbage in, garbage out,” as Aaron Kalb, a co-founder of Alation, told me in a previous interview.
Mittal doesn’t accept that. He thinks there is still a debate about whether a middle layer, such as business intelligence or data intelligence, will be needed. Perhaps, he suggested, increasingly sophisticated models will be able to handle raw data streams. One method of doing this is to essentially predict what data is missing, which is what generative AI does anyway. There’s a growing trend for “synthetic data” solutions that do just that.
Use Cases
So what are some of the early use cases for generative AI in the enterprise that Deloitte is seeing?
Mittal mentioned one in the healthcare field where they are working on a PoC [proof-of-concept] to generate accurate billing codes from electronic medical records and physician notes. Another example is with banks, which are looking to take advantage of the speed and lower cost of generative AI to handle customer interactions — such as chat-to-email conversion, or client correspondence via chatbots.
Mittal argues that these examples show how generative AI can be used to augment various professions, including healthcare and banking, to improve productivity and efficiency.
What Skills Do IT Departments Need?
One profession that will definitely be in demand for the foreseeable future is prompt engineering, a field that is still being defined. I asked Mittal what backgrounds and skill sets does his own GenAI team at Deloitte have?
He replied that his team has had a year-long head start in the field of generative AI, but they had to train their existing workforce to become prompt engineers — since it was not easy to find and hire them. He explained that prompt engineering requires the ability to ask the right questions, be curious, learn quickly, and analyze problems creatively.
He added that when it comes to building and refining large language models, a deep STEM background and software development experience are still necessary.
