Reflections on Generative AI

Our Chief Scientist Committee’s mission is to serve the success of Elder Research by keeping us at the forefront of data science technology and methodology. This helps us to efficiently provide value to our customers and to serve as a trusted advisor in the broader field of data science.

Tom Shafer

Date Published:
May 10, 2023


Generative AI, encompassing Large Language Models (LLMs) and image-generation models, has captured the public’s attention with easy-to-use interfaces like DALL-E and ChatGPT. As the AI landscape evolves and public acceptance grows, organizations must weigh the benefits and risks these innovative applications bring to a business setting. As a group of data science, machine learning, and AI practitioners, Elder Research keeps a keen eye on breakthroughs and developments in the field. We also collaborate with our clients in these areas, actively investigating how these technologies can be applied. Here we will outline a few potential benefits and risks of generative AI in business.

Potential Benefits of Generative AI

It is impossible to list all of the potential benefits of a new technology, so we focus here on a few applications that are important to different elements of a business—especially, those applications that enhance or speed up existing workflows. Here are four areas where generative AI can clearly provide value already: (1) coding assistants, (2) general workflow augmentation, (3) knowledge retrieval, and (4) image generation.

1. Coding Assistants:

LLMs can generate large blocks of code for multiple programming tasks. This can significantly increase efficiency and enable programmers to focus on design and program details.

2. General Workflow Augmentation:

Generative AI is being experimented with to improve work processes. For example, combining language models (ChatGPT) with image models (Stable Diffusion) can create marketing emails, social media campaigns, and websites.

3. Knowledge Retrieval:

LLMs can serve as interfaces for searching and interacting with databases and document collections. Fine-tuning these models with company information allows businesses to access and connect internal knowledge in new ways.

4. Image Generation:

Image-focused applications can create convincing images and videos based on text prompts. This capability opens up new avenues of creativity and can, for instance, generate competitive business headshots.

It’s difficult to comprehensively list the potential business applications of generative AI; however, these examples demonstrate how these applications are already providing novel capabilities.

Risks of Generative AI and Mitigation

Generative AI also comes with risks, as do all ML applications. We focus on four key business risks: legal, data privacy, factuality or correctness, and risks around control and ownership of the models.

Legal Risk:

The legality of generative AI is uncertain, with ongoing lawsuits related to coding assistants and image-generation applications. These lawsuits target the data used to train these models and the lack of consent from data creators/owners.
step image

Corporate Data Risk:

Using generative AI tools may expose proprietary or sensitive information to the AI tool's owner. A mitigation strategy is to fine-tune an in-house, proprietary language model for secure usage.
step image

Factuality and Correctness Risk:

LLMs can provide incorrect information or "hallucinate" purported facts. One countermeasure is to critically evaluate LLM responses and implement middleware to help control and guide model outputs.
step image

Model Ownership Risk:

Consolidated ownership of generative models leaves businesses dependent on a few model owners. This can lead to opaque models and biases, and even the potential for malicious shaping of the model's worldview. Users and businesses should consider diversity and potential lock-in when incorporating generative AI into their workflows.
step image


Generative AI presents a substantial opportunity for organizations, but it also carries with it a number of risks that need to be carefully considered. When organizations explore these new technologies, it is imperative that they be diligent in safeguarding their personnel and their data. Organizations can better prepare themselves for the next wave of generative AI technologies by remaining informed about the current benefits and drawbacks of the technology. Investigating “convex” opportunities, where the potential for loss or harm is limited while the potential for gain remains unhindered, will provide organizations with actionable insights into these technologies and lay the groundwork for embracing the future of generative AI, whatever it may bring.