Generative AI: Why boardrooms must embrace the next frontier of innovation

Samsung subsequently banned employees from using such AI tools. An internal memo reportedly reasoned that data transmitted to AI platforms such as ChatGPT and Google Bard would be stored on external servers. The data would, therefore, be difficult to retrieve or delete and sensitive company information could be compromised.

The incidents at Samsung offer a glimpse into the Pandora’ box of concerns global enterprises could run into if it is opened, as they grapple with the dizzying pace of generative AI’s rise. Since its launch five months ago, ChatGPT has racked up more than 100 million users. Individuals (and even smaller companies) across the world are using it to do everything from writing blogs, reviews and resumes, to making short films and realistic images, to generating software code and analysing broad economic trends. All of it without any human intervention.

While the interest has made generative AI impossible to ignore, the bigger companies are probably right to proceed with caution.

Chatbots like ChatGPT are trained on billions of words from sources like the internet, books, and multiple online sources including Common Crawl and Wikipedia, which makes them more knowledgeable but not necessarily more intelligent than most humans. The bots may be able to connect the dots but not necessarily understand what they spew out.

There are other concerns. The US Federal Trade Commission (FTC) has cautioned that a scammer could use these AI tools to clone the voice of a relative with just a short audio clip and create havoc. On 29 April, for instance, CNN reported that Jennifer DeStefano, a woman from Arizona, believes she was a victim of a virtual kidnapping scam, with someone cloning her daughter’s voice and asking for ransom. Likewise, a scammer can use these tools to impersonate a sibling and ask us for an OTP (one-time password) and siphon out money from a bank account.

Many prominent tech leaders, including Elon Musk, Yoshua Bengio and Stuart Russel, have called for a six-month moratorium on training systems that are “more powerful than GPT-4″, arguing that they should be developed only when the world believes it can contain the risks. The risks are high, and it’s not just industry leaders who are sounding the alarm. No less than Geoffrey Hinton, one of the godfathers of AI, known for his work on the deep learning that powers today’s generative AI tools, recently quit Google to “freely speak out about the risks of AI.”

Given this clear and present danger, mid-sized and large companies, especially those in the banking, financial services and insurance (BFSI) space and healthcare, are proceeding with abundant caution. Major financial institutions, including CitiGroup, Bank of America, Deutsche Bank, Goldman Sachs, Wells Fargo, and JPMorgan Chase, have already placed restrictions on their employees’ use of ChatGPT amid concerns over sensitive information being leaked while using the technology.

The chief information officer (CIO) of a leading multinational bank in India, who asked for anonymity, told Mint: “ChatGPT clearly has potential but it’s a bit early for the highly regulated BFSI sector. We deal with very sensitive customer information and cannot afford to play around with that. I would rather wait for the technology to mature and have some guardrails around it.”

Use cases emerge

Nonetheless, corporate boardrooms are brimming with conversations around generative AI, which was discussed by 17% of CEOs in the January-March quarter of this calendar year, spurred by the release of ChatGPT and discussions around its potential use cases, according to the latest ‘What CEOs talked about’ report by IoT Analytics, a Germany-based market insight and strategic business intelligence provider.

In fintech company Paytm’s earnings call, on 11 May, AI and AGI (artificial general intelligence) was mentioned eight times. Experts believe that in the not-too-distant future, an AGI machine will be able to understand the world as well as any human, and in many cases, even surpass human intellect.

Microsoft, Google, International Business Machines (IBM) and Nvidia are enhancing their generative AI platforms so companies can use them with fewer data and security concerns. Microsoft, for instance, has already begun providing enterprise users “with the tools necessary to build ChatGPT-powered applications”.

And OpenAI, too, is working on ‘ChatGPT Business’, which, it claims, “will not use data of end users to train its models by default”. Nvidia offers a cloud service (NeMo) to integrate generative AI capabilities into enterprise applications. Amazon Inc. has its own generative AI platform called Bedrock, while IBM offers WatsonX and has partnered with open source generative AI company Hugging Face, whose HuggingChat competes with ChatGPT.

Scepticism around data security hasn’t stopped enterprise use cases from sprouting. Companies such as travel and holiday fare aggregator Expedia have already begun using ChatGPT to provide the best flight tickets and also help travellers plan their trips and vacations.

Shopify Magic, an AI product from e-commerce platform Shopify, is generating product descriptions from a list of keywords or product descriptors in tones that merchants choose, while retail giant Carrefour is experimenting with ChatGPT to make videos answering customer questions such as ‘how to eat healthier for less’.

Generative AI has use cases in the world of human resources (HR), too. Tasks such as onboarding, training, performance management, and employee queries and complaints can be automated using ChatGPT. In the financial sector, AI can help with compliance, credit risk management, investment research, and legal document processing.

India’s play

In India, the Mahindra Group is exploring some use cases for its business units. “Generative AI is evolving at a fast pace and exploring the right use cases for our business is something we are very excited about. There’s no FOMO (fear of missing out) and no pressure (from the top management) — we are focused on finding the best use of this technology for our businesses,” said Rucha Nanavati, chief information officer of the Mahindra Group.

Walmart-owned Flipkart, too, believes generative AI has a lot of potential to address one of the core problems that any e-commerce platform is trying to solve — connecting consumers to products they may be interested in buying.

“Gen AI allows us to build more conversation, human-like agents or assistants to handhold the user through this entire discovery, purchase, and post-purchase customer service journey,” explained Jeyandran Venugopal, chief product and technology officer at Flipkart. “Gen AI can help us build high-quality content (pictures and descriptions) for our catalogue of products, and our merchandising and advertising campaigns. It can help us summarize product descriptions and user reviews to reduce the cognitive load for our customers,” he added.

Sanjay Mohan, group CTO at MakeMyTrip, is currently using generative AI for proof-of-concept (PoC) work. According to Mohan, large language models (LLM) are very good at summarizing things very smartly and crisply. “In the case of reviews, if someone is saying ‘outstanding’ and someone else is saying ‘excellent’, it knows that both are the same. So, that summarization is something that we can use”.

Audio platform Pocket FM uses generative AI to automate creation for long-tail content (which uses specific keywords such as ‘size 7 hiking boots for men’ rather than just ‘hiking boots’), trailers and promos, and to provide personalized recommendations by analysing user data, according to Prateek Dixit, its co-founder and CTO. He added that the adoption of this technology had helped Pocket FM reduce translation time by more than 40%.

Haptik, a Mumbai-based startup, is using generative AI to make bot conversations less robotic and more free-flowing, according to Akrit Vaish, its co-founder and CEO. With ChatGPT, Haptik also hopes to “create content and add countless variations of bot responses”. Homegrown Zoho Corp., too, has launched 13 generative AI Zoho application extensions and integrations powered by ChatGPT.

The limitations

But integrating the application programming interfaces (APIs allow applications to talk to each other) with the business workflows of other units has its own set of challenges for companies. As Sumanta Kar, technology partner at consulting firm EY India, points out, once you adopt a tool like ChatGPT, you have to continuously monitor, re-train, and fine-tune to ensure that the models continue to produce accurate output and stay up-to-date.

According to Sanjeev Menon, co-founder and head of tech and product at E42.ai, a natural language processing-based AI platform, while ChatGPT excels at generating text and answering questions, it may not be as capable of automating complex workflows in an enterprise environment, which explains why such models have to be fine-tuned before being put to use.

The reason is that enterprise data includes structured and unstructured data such as videos, audio files, social media posts, emails and more. Organizations, therefore, rely on specialized tools that can interact with third-party and internal systems. Also execute specific actions based on the data gathered.

“Today, there is no customer meeting without a discussion on generative AI,” said Jaya Kishore Reddy, co-founder and CTO at conversational AI startup Yellow.ai. But he, too, highlighted the need for an “orchestration layer” to connect generative AI models like ChatGPT to business systems, which requires significant customization. “Even the plugins (tools that help ChatGPT access up-to-date information, run computations, or use third-party services) need to be able to connect different systems and multiple workflows in enterprises,” he added.

Bharath Shankhar, vice president of engineering at conversational AI company gnani.ai, emphasized the importance of defining a boundary or scope for GPT systems in specific domains to make them efficient. Shankar, too, underscored that there is a lot of effort required to make GPT “integrate with enterprise backend systems such as ticketing tools, CRM (customer relationship management), etc.” He said companies will have to ensure that there are no regulatory violations in accessing or sharing patient data, for instance, that would result in a HIPAA (Health Insurance Portability and Accountability Act) violation in the healthcare sector. Further, he pointed out that the response time of a ChatGPT bot may not work in scenarios where users need a real-time response without a lag.

But then, generative AI models are quickly growing in strength. IBM predicts that the so-called ‘foundation models’—models that are trained on a broad set of data and can be used for different tasks—will soon use self-supervised learning. They can apply the information they have learnt to a specific task, dramatically accelerating AI adoption in business.

The frenetic pace at which these models are training themselves, and the impending introduction of ChatGPT Business, do not augur well for business executives sitting on the sidelines.

Prasid Banerjee & Abhijit Ahaskar contributed to the story