Toxic Data in Atlanta

The Successful AI+IM Global Summit in Atlanta

Last week, I was in Atlanta, Georgia, to speak at and attend the newly reimagined AIIM event - the AI+IM Global Summit. Whether by luck or design, it was a resounding success. It was a timely and important conference. What made it different was the heavy focus on education and conversation, particularly around the topic of Agentic AI and, indeed, AI in general.

Now, you likely think every tech event is focused on Agentic AI, and you would be right. However, the critical differentiator in Atlanta was that the attendees were all information managers in some form or fashion. In other words, the custodians and guardians of unstructured data, and without their involvement, enterprise AI ambitions are dead in the water. Yet, as of today, they are seldom involved in AI projects; hence, it's little wonder that the air is fast going out of the AI bubble.

The Challenge of Toxic Data in AI Development

I don't know if it's an old term that has been revived, but I encountered 'Toxic Data' for the first time at the event. In essence, think of Toxic Data like this. If you train a specialized AI model on 10,000 good and relevant documents, the entire training process can be poisoned if there is a cluster of poor-quality papers in the mix. And there is a lot of that going on right now.

At a meta (no pun intended) level, we see it in LLMs trained on the detritus of the internet, spewing out misleading and often fictional answers to essential questions. At a micro level, we know the struggle to develop meaningful and valuable AI applications that go beyond tightly ring-fenced data sources, for example, invoices or tightly controlled customer data.

AI projects fail alarmingly once we veer beyond these essential but minimal areas. Why? Because, quite simply, the data is not good enough. So, it should be crystal clear that information managers, along with their knowledge, and skills must be involved in AI projects. And what the conference told me, at least, was that they very much want to be part of the conversation. So, that leads to the question of why they are not part of the conversation?

The Disconnect Between AI Innovators and Information Management Practitioners

Primarily, I believe its because the people driving the AI hype engine are unaware of the existence and value of information managers. Earlier this year, I spoke at some length with a senior executive at one of the world's largest enterprise software firms. Who told me that the most successful AI projects involve a designated knowledge manager. He wondered if I had ever heard of knowledge managers as it was a new term for him.

I resisted the urge to scream and instead patiently explained what knowledge and information managers do, and their long history in the industry. But rather than marvel at the executive's ignorance, it got me to thinking why we have this divide, and why such roles were a revelation to him, and indeed to most of his peers. Maybe more importantly, what can be done to bridge the divide?

Breaking Down Tech Industry Silos

The fact is, the world of tech is and always has been incredibly tribal. Those of us in the information and document management world tend to stay in it and never stray elsewhere. The same is true if you are in the CRM, storage, networking, ERP (the list goes on) worlds. We just don't tend to cross pollinate. But AI demands that we do, and if we don't, AI simply doesn't work.

To date the world of AI has been driven by those in the structured data, GPU and data center tribes. But the time has come for that to change, in fact there is a deep irony that the experts in unstructured data, the very stuff that Agentic & Generative AI depends on have been left out in the cold. On a more positive note, there is a slowly dawning awareness that things need to change. But the information management tribe needs to also do a much better job of shouting ever louder for recognition.

Moving Forward: Collaboration Is Key to AI Success

The AI+IM Global Summit in Atlanta was an important baby step in organizing, educating, and changing the situation. I say baby steps because so much more needs to be done, but a baby's first steps are the hardest; with practice, they get easier and easier to the point of walking without effort.

If you are an enterprise, AI-focused vendor just dipping its toes in the world of unstructured data, then reach out to us here at Deep Analysis or AIIM and get involved. The same goes if you are an information manager; contact us. And be sure to download the (free) infographic that Deep Analysis and AIIM jointly produced, which provides five simple questions to answer at the start of every AI project. It's a great tool to level the playing field and bridge the divides.

Lastly, hats off to the AIIM team for organizing such an engaging, high-energy, and valuable event. Here's to many more in the future!

This blog post was originally published on Deep Analysis's site and republished with permission.