By Paul Huang, VP of Engineering and Research at Premise
Recent advancements in generative AI have taken the world by storm, and AI is now a term heard at the kitchen table and in the conference room. This wave of technological innovation is poised to revolutionize work as we know it and expand the horizon of what is possible when it comes to business and product capabilities.
However, how does an organization successfully integrate and adopt these technologies, without taking on additional risk or disruption to their organization? There are a few key ingredients to success:
Organizational and Product Alignment (The Kitchen, The Chef and The Menu)
To maximize return on investment (ROI) out of AI investments, the organization and products need to be in sync and support the initiatives. While R&D or machine learning teams might be able to validate the idea or show its feasibility, many organizations will run into hurdles taking that prototype and seamlessly integrating it with the organization’s products and processes. In short: turning an AI idea into action.
To do that, it’s critical to incorporate AI research and development efforts as a core part of the product development cycle. This careful alignment ensures not only that there is support across the organization, but also that AI is the appropriate tool for the need at hand (which may or may not be the case). It takes deliberate organizational and product development setups involving people, processes and technology across the entire company to establish a strong AI culture.
Model Architecture (The Recipe)
Selecting the right model to solve the problem is oftentimes a daunting task. There are dozens of model architectures out there for any particular problem space. There are a couple of things to keep in mind that will help ease the process. First, is to have a clearly defined problem and objective in mind. These will help guide the teams to establish a concrete way to evaluate their approaches.
Next is to evaluate which approaches fit the constraints of the organization, product and problem. For example, if the problem is highly dynamic, then an online approach may be more effective. Meanwhile, if the product feature needs a fast synchronous result, then a lighter weight low latency approach may be best. This should narrow down the candidate set of models to select from.
After that, the teams should test and validate by gathering data and testing the various candidates against the problem in both a limited and real-world experimental setup. And as one might already notice, when it comes to models, it should not be all about the latest flavor-of-the-month techniques but rather selecting what fits the problem and needs.
Data (The Ingredients)
The classic saying that data is the new oil is more applicable than ever in the world of AI. Data volume, specificity, quality and relevancy are some of the key qualities to making the AI product and solution successful.
Where organizations often run into challenges is in access to high volume and task-specific data to effectively solve challenging business problems. For popular technologies like ChatGPT, its AI power lies in the volume and types of data being utilized. Since the underlying data is very diverse (with images even coming in GPT-4), they can create strong results. However, to apply the model to solve a more nuanced use case, one would need to provide more nuanced data, either through refining the context window, finetuning the pre-trained model to your specific requirements, or other methods.
Premise is a great example of an AI-based company that has put AI refinement into action. The company was able to build a robust product identification system that automates the detection of various consumer products. By gathering and labeling many real-world images taken by users, the product was able to learn the noisiness and complexity of the real world and apply it in its predictions more effectively.
Feedback (The Taste Test)
Continuous monitoring and feedback on the production AI system is the key to sustained success and competitive edge, like how humans learn continuously via experiences and interactions with the world around us. Not only does this help the model advance in an ever-dynamic world, but continuous monitoring also ensures that if there is a problem with inputs or outputs, it is flagged quickly and corrected. Improving the system based on these in a continuous and regular cadence would ensure the value and capability of the system.
A Bright Future for AI Possibilities
The power and capabilities of AI aren’t new, but recent advancements have put its potential into sharp perspective for companies around the world. As this technology becomes more and more accessible and commonplace, businesses will want to ensure they are using best practices to ensure they are realizing the full value of their investment. In doing so, they can truly ride the wave of our new Age of AI into great success.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.
Other Topics
Technology
NightDragon
NightDragon is an investment and advisory firm focused on growth and late-stage investments within the cybersecurity, safety, security and privacy industries.
Read NightDragon's Bio