By Naveen Kamat, VP & CTO, Data and AI Services, Kyndryl
In recent years, Artificial Intelligence (AI) has been gaining traction across businesses and industries. Investments in India’s AI capabilities are growing at a compounded annual growth rate (CAGR) of 30.8% and are expected to touch $881 million in 2023, according to Invest India. Stanford University’s Artificial Intelligence Index Report 2023 reported that India saw the fifth highest amount of private investments in AI at $3.24 billion. India also has the third-largest talent pool in the world for AI.
The way that enterprises use AI has also been evolving from traditional algorithms (Artificial Narrow Intelligence) that focused on specific, repetitive tasks, moving towards new-age generative AI models that are pre-trained to perform a broad set of pre-defined tasks and to generate new content. The evolution to Artificial General Intelligence (AGI), where there will be intelligent agents who will potentially be capable of defining tasks and be comparable to human cognition is still very distant. However, the current innovations in generative AI and the capabilities that they represent mark a key milestone in this evolution towards AGI.
The inevitable ubiquity of generative AI
A NASSCOM report found that funding in the generative AI sector grew by a whopping 12 times year-on-year in 2022, making it the most funded year in the sector. Unlike traditional AI, generative AI algorithms elevate knowledge work and have a much greater impact, especially on content creation, content discovery, augmenting conversational AI, and AI simulations. Therefore, generative AI presents several opportunities to augment customer or employee experience and rapidly accelerate the innovation cycle within any enterprise. A popular use case, for instance, can be contact center technology deployed in call centers.
So far, conversational AI interfaces such as chatbots have been used in basic customer engagement. Now, with generative AI, a virtual service assistant can bring far more sophisticated insights and a broader range of interactions from the customer calls that can help with summarisation, translation, sentiment extraction, call feedback for agent self-improvement, customer specific insights during the interaction.
Pitfalls amid potential
As with any revolutionary trend, generative AI not only promises immense potential for enterprises in broadening their sphere of services but also brings with it certain pitfalls. Enterprises must recognise certain specific challenges such as ethical concerns, hallucinations, threat of data breach, and inadequate supervision, among others. While these challenges are present with traditional AI, they are even more nuanced for generative AI use cases.
Establishing the Responsible Use of AI
To capitalise on the benefits of this technology optimally, certain considerations are essential. The following are important to ensure that generative AI technologies are used most optimally and responsibly:
Building the right data foundation
Since AI technology is only as strong as the data it’s fed, it’s critical to have a strong data
foundation built on quality, accuracy, accessibility, and interoperability of data. To avoid the pitfalls of generative AI, enterprises will need to ensure seamless integration of data from various sources to form the fundamental elements and structures to inform business decision-making.
While using generative AI, organisations are bound to find the promises and benefits of more nuanced interactions or hyper-personalisation. However, in the process, they also must ensure that customer data and personally identifiable information (PII) are not compromised.
Extensive input & output validation
A key aspect of getting the best outcomes from generative AI is prompt engineering – which
essentially implies having the right prompt sequence and design, prompt templates that can
help simplify and optimise the way the technology is used. But an inherent risk here is to ensure prompts are not hacked by prompt injections, prompt leaking or jailbreaks. Prompt injection implies injecting harmful text to manipulate the output, leaking extracts of critical information (company or personal), and jailbreaking bypassing in-built safety measures.
With some of the popular large language models (LLMs) are either unsupervised or semi-
supervised, there are risks around hallucinations and inappropriate output. As enterprises move beyond Gen AI exploration to deployment at scale, they must set up guidelines including prompt filtering, human-in-the-loop process for feedback and output validation, and hallucination monitoring.
Judicious use of LLMs
Cost-effectiveness and sustainability will continue to hold pre-eminence over the use of technology. Just because LLMs are available for use does not warrant their utility in all cases. Organisations must ensure judicious use of LLMs and toggle between generative and traditional AI so that significant processing can be done using the latter when required.
Depending on the choice of large language model deployed, caching strategies, type of data
structure or configuration parameters while querying, the cost of execution, and wasteful
consumption of resources can sharply vary to achieve similar outcomes.
Despite the pitfalls, the growth of generative AI is inevitable, as its benefits like accelerated
innovation, increased productivity and efficiency, etc outweigh the concerns. The experience economy will greatly benefit from it and the most common impact for enterprises in the next decade will be greater productivity and accelerated innovation.
However, the right guardrails are crucial to ensure that the ethical concerns of all stakeholders are met while ensuring judicious and sustainable use of resources.
