Buzz around Generative AI
In an era of Artificial Intelligence, Generative AI (GenAI) has gained massive popularity in the industry. The below image (image1) shows exposure of GenAI in various domains.
Image 1 – Exposure of GenAI per McKinsey^^
But, even after having so much popularity, McKinsey study^^ shows only 24% respondents use GenAI tools for their work in fintech industry. As per general belief, incorporating GenAI is often considered straightforward, so what obstacles do companies face in using it effectively?
Why organizations are struggling to make use of Generative AI
To comprehend why companies are struggling to harness GenAI’s potential effectively, let’s delve into a simple example – Q&A Chatbot. Our objective is to develop an intelligent Q&A chatbot for the needs of a financial company –
- Answer customer queries on their products and policies
- Provide suggestions for suitable financial products
- Prefill web application (online form) for account opening
- Detect fraudulent activities
Now, the question is – can we simply use the LLM (Large Language Model) as-is for querying and retrieving answers? The answer would be NO. These Large Language Models are trained with huge publicly accessible data, potentially leaving them devoid of organization-specific i.e., local context. The Below example (image 2) demonstrates how GenAI model lacks local knowledge.
Image 2 – ChatGPT snapshot
Can we effectively engineer the prompt? Answer – Effective prompt engineering might solve about 20% of the problem, still major part of the problem will be left unsolved
Can we finetune with company specific data? Answer- Yes but acquiring thousands of quality data points is challenging for most companies. Even if they obtain such data, converting it into a usable format and fine-tuning it requires significant effort and resources, with potential improvements being marginal.
A more meaningful approach possibly involves generating embeddings and establishing an index of business-related documents. While this may address a portion of the issue, it’s still not perfect as it lags in keeping up with the most recent information (image 3).
image 3 – Experimentation with GenAI model to access latest information
Using none of the above approaches, we can offer relevant product suggestions, detect fraud etc. So, what is the underlying issue that companies seem to be overlooking or misunderstanding?
GenAI as a tool rather than a comprehensive solution
People have unrealistic expectations from GenAI. It’s important to understand that GenAI serves as a component to enrich a dish, rather than being the dish itself. LLMs exhibit impressive skills in emulating human-like text production but lack current information, organization specific (local) context, or even tasks where basic mathematics is required etc.
How to leverage GenAI effectively
Having understood the challenges associated with GenAI, let’s now explore effective strategies for leveraging it to address our specific concern related to chatbot –
- External agent for latest info – External aid can be extended to LLM models in areas where their knowledge is lacking. In chatbot scenario, external agent can be utilized (image 5) to facilitate LLM models’ access to the most current internet data.
image 5 – Usage of agent to retrieve latest information
- Machine Learning solution for specific tasks – By incorporating a machine learning model, we can offer product recommendations, which in turn helps the LLM model deliver results tailored to a specific domain rather than generic ones. ML based fraud detection engine can monitor the chat and detect fraud. Similarly, Computer vision algorithms can help in prefilling the online form for better user experience.
- Human in loop – Engage in continuous monitoring and evaluation of these models. Incorporation of human in the loop, can identify any biased or erroneous content, verify contextual appropriateness, spot possible issues, and make informed judgments when the system’s result demand extra attention.
GenAI, with its inherent limitations, can’t handle every task on its own; therefore, supplementing it with external support can lead to remarkable outcomes. It should serve as a productivity-enhancing tool rather than being considered a standalone solution.
To sum up, Generative AI stands as a remarkable catalyst for both present and future technological advancements, but it’s true efficacy lies in its appropriate application. If used in the right way GenAI has the potential to enhance business decision-making processes, leading to increased efficiency and cost reduction. To become unstoppable, business must uniquely incorporate GenAI and automation into their business strategy. The way Business incorporates GenAI into their business strategy will ultimately dictate their survival or success.