The Impact of Generative AI on Data Analytics

Perhaps no other technology in recent times has garnered such interest and generated so much hype as AI has; even Bill Gates has reportedly said that “AI is the most revolutionary technology in decades.” Generative AI tools like ChatGPT have caused a great deal of ruckus as they can create original content, unlike traditional AI which used pre-existing data to classify information or make predictions. People have used generative AI to do all kinds of things: write rap, take the bar exam for lawyers and identify bugs in computer code.

The next question to pop up is, how will AI impact the job market? It has the potential to take over certain roles traditionally held by humans, such as writing reports and creating documents. As AI continues to improve, will more and more current jobs be threatened by automation? What will this change look like in the world of data analytics? There is no definitive answer but it is clear that major adjustments will be required.

An interesting study was carried out on the impact of generative AI at a customer contact centre by researchers from Stanford and MIT. The employees at the centre were supported by an AI assistant that instantaneously suggested responses or solutions to a customer’s problems. The researchers analysed six months’ worth of data and discovered that the number of problems the workers were able to resolve in an hour had increased by 14%. They also found that customer satisfaction increased, and employee turnover decreased.

The researchers then looked at who benefitted most from the AI assistant. They found that the improvements mostly came from less experienced workers – people who had been on the job for less than a couple of months, while the most experienced workers showed almost no improvement. This suggests that because AI learns from previously human-generated experiences, the experienced workers, already well-versed in the process, found little use for the assistant’s suggestions.

Another experiment done at MIT on college-educated advertising copywriters yielded similar findings. Unsurprisingly, using AI made everyone more productive and saved a lot of time. But the copywriters who were good improved only a little, while the relatively poorer copywriters improved sufficiently to become ‘average’. Essentially, generative AI helps reduce productivity inequality by raising the floor considerably.

This is the point where conjecture on the impact of AI on jobs and analytics begins. The advent of the computer age made elite workers better at their jobs (and much better off), while simultaneously making a lot of middle level jobs (like typewriting) redundant. The hope is that AI can help reverse the trend by making certain kinds of expertise more accessible and cheaper. However, there is no empirical evidence to support this claim yet.

Generative AI could have a similar impact on analysts, as it can reduce data processing efforts and free them to focus on generating insights. An immediate use case could be identifying the relevant features in a data set by understanding the different relationships among the variables. Another exciting use case is synthetic data – artificially created data that resembles real-world data, but doesn’t contain private information. It allows the analysis of sensitive data, like medical records by healthcare researchers, without breaching data-sharing agreements.

Generative AI could also streamline the process of storing and managing data. It can help entities improve the quality of their information by augmenting customer level databases with industry or macro-level data. This, in turn, can help increase the velocity of processing ‘big data’, by speeding up exploratory analysis and real time prediction capabilities with generative AI assisting in the code-writing process. Perhaps it will help blur the competencies of data engineering, data sciences and data analysis.

Despite the excitement, all these capabilities serve little purpose without a key ingredient that AI does not have yet – domain knowledge. Without it, you have no idea what questions to ask AI for a given data set, or the appropriate context to interpret results. The greatest value of human analysis lies in our ability to answer ad hoc questions that are complex and nonlinear. Generative AI is ‘trained’ on existing data to generate a probability-driven answer. How would you train a model on multi-layered, situational questions that have never been asked before?

There are other limitations. Entities will have to adapt significantly in order to use generative AI. Samsung recently banned its employees from using AI when its coders accidentally shared sensitive IP on ChatGPT. Therefore, organisations will have to build bespoke generative AI tools for internal use, and not everyone has the resources to do this. Moreover, generative AI models can be trained on data sets that have biases – companies like Visa are acutely aware of this and are wary of using generative AI for use cases like credit scoring.

The question is not whether AI will be good enough to eventually take on more cognitive tasks, but rather, how do we adapt to it? The status quo between technology and labour has been disrupted several times and humans have always adapted (without realising it). In the early 1900s call operators were one of the most common jobs for young women in the US. Two decades later mechanical connectors had taken away up to 80% of those jobs, and a lot of women moved into secretarial labour. A more recent example of adaptation is ride hailing. Although traditional taxis leveraged navigation systems and online maps to optimise their operations, companies like Uber and Careem changed the landscape by leveraging the same technology to enable ride hailing.

The reality is that nothing stops technology from marching forward. I take solace in the findings from Dell’s recent report: 85% of the jobs that will exist in 2030 have not been invented yet. Or it just may be that the rise of AI will concentrate power in the hands of the people who design it, leaving the rest of us scrambling for work. We will soon find out.

Ans Khurram is an analytics and insights professional.
anskhurram@gmail.com