We just finished a five week sprint with 200+ clients on Generative AI and the excitement is overwhelming. There’s no question companies are going to adopt these Copilots, Assistants, and Chatbots like crazy. But it’s still not clear how, where, are with what technology. Here’s what we found out.
First, about a third of the HR leaders and professionals we talked with are still figuring out what this is. They’re not sure how these things work, they’re confused about the proliferation of LLMs, and they haven’t come to grips with the complexities of “prompt engineering” yet.
One of the companies, a global pharma, has a mandatory training program for all employees, encouraging them to learn the concepts, terms, and ideas. As our Deep Dive on AI in HR research describes, there are a lot of interesting new computing concepts to understand, so companies want to get this basic knowledge into the organization.
For example, most people don’t yet understand that a Gen AI solution has at least three major components: the search element (finding, collecting, and analyzing data from documents, audio, video, etc), the language and generation element (answering questions, responding to prompts, understanding what prompts actually mean), and the orchestration element (what will the Chatbot or application actually DO after you find what you need? Will it produce a report, link you to a system, or something else).
As I talk about in this week’s podcast, vendors like SAP have partnered up with Microsoft and IBM to build in-depth copilots for their entire technology offering. IBM’s products Watson Assistant (the language element) and Watson Orchestrate (the process management layer) are being used to build a highly intelligent chatbot to help SAP customers interact with all the SAP applications without clicking all over the place to find the screen you need.
For corporate HR and IT professionals, this is all a bit daunting. One company told us their IT department is now afraid to come to meetings because everyone expects them to know “everything” about Generative AI. IT teams are learning as fast as everyone else, and they are only a few weeks ahead if even that. And most of the big vendors are still getting their act together, making IT departments’ decisions even harder.
One of our clients, a large CPG company, has created a “promptathon” – a series of teams (10-15 people in each) to test out up to 22 different use-cases for a GenAI platform. I think this is a marvelous idea, because every use-case for Generative AI is different and no single application or LLM will do everything you want.
Many of the HR and IT Professionals are worried about their users. Will they know how to use these things? Will the data be secure and protected? How do we make sure the Gen AI system knows this user’s identity, privileges, location, and job level so he or she only sees the information they’re supposed to see? As I talk about in the podcast, this is all moving into basic IT implementation stuff now: data integrity, security, identity management, and protection.
Another company has already built a “skills taxonomy” for new users who want to learn about AI. Working with their IT partners they built a list of 20+ “skills” people need to learn to use these systems, and this will form the basis of their change management strategy as they roll these new applications out. This is important, because so many of us will interact with these applications and we need to know how to use them and what to expect when they don’t perform as advertised.
Let me also add that Gen AI is not a “swiss army knife” that can do everything. While the underlying AI technology is quite powerful, the real use-cases will vary and we will want to build corporate solutions that excel at one thing at a time. A GenAI application for an ad agency that creates videos, audios, and images is nothing like a Gen AI tool that builds customized candidate emails or another one that serves as a teaching assistant to people going through technical training.
Our Gen AI solution, the JBC HR Copilot, has been trained to find, analyze, and answer hundreds of questions from our research corpus. But it isn’t very good at finding images, charts, and graphics yet. We’re training it to do these other things as we learn how people will use it.
We’ve also discovered that every Gen AI solution you build is going to create a “cavalcade” of user questions. Once I ask the Copilot about how to select an LMS or implement a capability academy, I then want to ask it about vendors, and from there I want it to give me use cases and implementation tips. When a user asks your chatbot about their vacation balance they may then ask when their sabbatical comes up and possibly the leave benefits for having a baby. So these are not just “retrieval and document generation” systems, they’re actually intelligent “agents” that can even predict a series of issues or questions employees may have.