Understanding the Impact of Complex AI in the Enterprise
It is hard to miss the buzz surrounding ChatGPT. It’s all over mainstream media and has caught everyone’s attention. It might appear to have come out of nowhere, but the underlying generative AI capabilities have been around for some time, and we’ve seen ‘snippets’ of the AI’s capabilities in blog writing, note taking and creative applications. We must give credit to ChatGPT for wrapping it up nicely into a chatbot application that can deliver cool results. But we must also be careful of how this is leveraged, especially in enterprises that depend on AI to deliver fact-based results.
What does ChatGPT and generative AI mean for enterprises and the IT
organizations that are responsible for architecting and implementing the technology?
Generative AI is a class of AI models that generate valuable output based on their training data and the input provided by users. In the case of ChatGPT and other similar types of systems, these large language models are trained to produce content such as blogs, articles, and other written works. They are large language models that are trained on their understanding of a language, its constructs, and some context around sequencing and meaning; and can brilliantly produce content that is indistinguishable from that produced by humans. Blogs and articles produced by these systems are well written, but there is an underlying danger and a very important caveat that needs to be understood by those reliant on these types of models, as pointed out by Dr. Elham Alipour in the article “Future of Generative AI for Enterprises: Are Large Language Models Viable Options?”
For those not familiar with how AI models work, it is critically important to understand that they are trained on a very specific ‘skill set’ – and a fairly narrow one at that. They are also trained using particular methods and data to support that very specific skill set.These AI models are far from the Artificial General Intelligence systems that can reason, inherently update themselves with new skills, and learn as they go on a broad scale. They are not the Hollywood version of AI.
Similarly to how humans develop and train their specific skills, and focus on mastering their skills, models do the same with data input. When it comes to the world of business, we typically rely on teams of experts with individual skills who work seamlessly to complete tasks, projects and initiatives. Even with all our day-to-day activities, either at work or at home, we typically rely on teams of skilled experts trained on anything we need. Take for example a major home renovation project where you rely on several different people with specific skills such as framing, electrical, HVAC, plumbing and interior design – even these people rely on other experts for supplies and equipment.
The world of AI is no different. Models have specific skills. And if you are in IT or another technology field, you will have to plan the models, build the models, train the models, install the models, coordinate the models and stitch them together so that they all work nicely in unison. It’s a headache inducing thought considering everything that has to materialize. However, to be successful in delivering solutions to business, it must be done.
Building and maintaining a library of AI models is a critical element of any IT strategy. Generative AI is only one, or one set, of AI models in that library of skills. There are many others, including purpose-built models for contract evaluation, credit reporting, privacy compliance, equity research, algorithmic trading, process automation, route calculation, image analysis, data analysis, and any number of requirements for prediction and recommendation.
This is where Adaptive AI comes into play. It is a nascent category of capabilities that are required in enterprise and IT in order to manage a library of models and put them to good use throughout the entire organization. Moreover, it’s about delivering on an adaptive AI framework that can easily adjust to changing business conditions, including the support of ‘upskilling’ models used by the business. The ‘upskilling’ component is essential – and it’s hard. This is where IT and industry experts must look at techniques such as transfer learning, reinforcement learning, and continuous learning to keep the models updated. Consideration must also be paid to new methods that orchestrate all the data that is needed to feed and support the system. It can be a heavy lift to do this with a large and knowledgeable team in siloed use cases, but across the entire business it is hard … extremely hard!
Adaptive AI is becoming a hot topic in 2023 as signaled by Gartner. It is something that companies have been looking at and considering for some time as they feel the weight and pressure of ‘care-and-feeding’ and ‘jury-rigging’ their myriad of models scattered around the business. Some CIOs that I have spoken to recently have pointed to their frustration around the duct-tape and baling wire necessary to utilize ‘one-and-done’ AI models.
At Charli AI we’ve focused predominantly on an adaptive AI framework and platform. Our Ancaeus platform is helping enterprise customers effectively and dynamically manage a complex system of intelligence, with tens, if not hundreds, of AI models in operation – including generative AI. The platform is seamlessly integrating and orchestrating the models; including incorporating, consuming, and feeding the purpose-built models developed by our customers. The platform is also intelligently applying and upskilling the models, as needed, by our customers, with an extensive focus on feedback-loops, transparency, and explainability – essential features in adaptive AI.
Many elements must come together for an Adaptive AI framework to be successful, including the careful architecture and design around active metadata management – the data needed by the models and the business. As Adaptive AI matures, IT organizations must consider a new breed of technology, tools, and solutions to manage their foray into the world of AI and the complex system of intelligence. This is no different, and likely far more important, than the digital transformation that IT had to go through to adopt cloud technology.