As I reflect upon my time at AIIM, it strikes me that most of my tenure was, in fact, during a previous period of revolutionary change. And reflecting on that revolution might give us some insights into the new revolution that is now upon us.

Radical changes in the price/performance of computing, storage and communications technologies marked the period from 1995 to 2015. Plot any of these — even on a logarithmic scale — and the curves shoot wildly off the top of the page.

1995 = DEC Alpha 21164 - 9.3M transistors

2005 = Intel Pentium D - 115M transistors

2015 = IBM z13 - 4B transistors

1995 = $26.25 million/TB

2005 = $76,172/TB

2015 = $3,661/TB

1995 = $213,846/TB

2005 = $406/TB

2015 = $28/TB

1995 = 24 kbps

2005 = 1 Mbps (1000 kbps)

2015 = 39 Mbps (39000 kbps)

This revolution wasn't limited in its implications for the document and content space or even the technology sector. True revolutions have societal implications by their nature, which is why they are rare.

Of course, true revolutions aren’t always recognized early on, even by those smack in the middle of them. Figure 1 is an excerpt from the lead editorial in AIIM's Inform magazine in April 1996, the month before I arrived at AIIM.

It took a while, but revolutionary computing, storage and communications forces eventually swept through the document and content management space. The winds of process change initially drove these forces. In Crossing the Chasm, Geoffrey Moore documented how Documentum applied document technologies to mission-critical, vertically-focused processes like new drug applications and then used this beachhead to “cross the chasm” into broader enterprise applications. Many other vendors followed suit.

Revolutionary winds also swept through ordinary knowledge work involving documents. In Clayton Christensen fashion, Microsoft SharePoint disrupted the market with a simplified content solution (at least compared to the initial process-driven content solutions). Box then disrupted SharePoint with an even simpler freemium cloud-based model. “Traditional” process-centric content management vendors initially dismissed both as not "real" ECM.

Ultimately, process-centric and knowledge worker-centric content management intersected, converged and overlapped in a wildly chaotic fashion multiple times. We in the AIIM community ultimately tried to rationalize all of this in a model called "Systems of Record" and "Systems of Engagement," (Figure 2) but that's a story for another day.

Fast-forward to the present day and the market tumult around AI and LLMs. Regarding document and content management, are we entering a revolutionary new period? I believe the answer is “YES.” The indicators are all around us.

Almost overnight, ChatGPT pushed the revolutionary AI and LLM forces, brewing for some time, into sudden popular consciousness. Figure 3 from Google Insights shows the relative search interest in "document management" and "content management" over the past two years. It is about what you might expect.

It's humbling for those in the content space who sometimes have exaggerated impressions of our importance, but it's about what you might expect. Now add ChatGPT (Figure 5).

Revolutionary indeed.

Why did such a rapid change occur? More than any other single reason, it's because ChatGPT crossed over into the consumer realm. Once technologies are consumerized, change accelerates.

Add a few more indicators of pending revolution to the mix.

So, let's stipulate that society is on the cusp of an AI and LLM-driven revolution just as fundamental as the one driven in the 1990s by the collapse of computing, storage and communications costs. At this very early stage, what should user organizations keep in mind as they begin to incorporate these revolutionary forces into their planning?

The first thing to remember about AI is that emerging technologies are imperfect. Given the conversational elegance of ChatGPT, it was natural that many confused conversational elegance with information accuracy. Just about anyone who has dabbled with consumerized versions of these technologies since ChatGPT went mainstream in early 2023 has a story about AI factual mistakes. These range from trivial and fun errors — it's incredible how factually deficient early ChatGPT was at tasks like writing a bio or solving a Wordle puzzle — to serious mistakes — like making up phantom cases to support a legal brief.

There is even a cottage industry around mistakes made by Google AI Overview searches. These include responses to queries like:

The good news is that AI learns from these mistakes; if you try to recreate them now, they will be gone. The bad news is that risk-averse enterprises need to figure out how far and how soon they want to push the AI risk/innovation curve.

According to The New York Times (How Tech Giants Cut Corners to Harvest Data for AI), “In late 2021, OpenAI faced a supply problem. The artificial intelligence lab had exhausted every reservoir of reputable English-language text on the internet as it developed its latest AI system. It needed more data to train the next version of its technology — lots more.”

There are three implications of this for enterprises looking to apply LLM models to practical content and process problems:

3. AI particularly impacts two long-standing content management focus areas:

1) document processing and

2) document structure.

Dan Lucarini from Deep Analysis believes we are entering the fourth wave of Intelligent Document Processing (IDP).

Wave 1 solutions used Optical Character Recognition (OCR) to convert characters in an image to text characters a computer could understand.

Wave 2 solutions used forms and templates to give a computer hints of what information lay where on a particular form so that organizations could more easily extract data from that location.

Wave 3 solutions use machine learning to analyze sample sets of documents with known structure to help computers learn new document types.

Wave 4 document processing will use LLMs to bypass the need for sample data and learn how to curate, categorize and extract information from documents over time.

Machine processing of documents also carries fundamental implications for the actual structure of documents. Alan Pelz-Sharpe, also from Deep Analysis, notes, “The business information contained in files accounts for a small proportion of the overall file size but typically represents most of the file’s value…Traditional ECM systems were designed for human interaction, but automation, driven by AI, is shifting the balance toward machine processing.”

Process-driven AI adoption: Most jobs and departments in an organization have responsibility for: 1) processes or parts of processes, 2) goals related to those processes and 3) specific tasks that somebody must complete. Using AI to automate specific document-intensive processes will follow a similar pattern to those used during the previous computing, storage and communications revolution. Deployments will begin with specific niches and then try to move more broadly.

We have already started down the path of using technology to automate some of these tasks, particularly those that are predictable and routine. Many organizations use robotic process automation (RPA) to speed up processes like data entry, extraction, invoice processing, customer service and order processing and reduce error rates. AI technologies will become an increasing part of the offerings of RPA companies and expand the kinds of tasks that organizations can automate. Successful vendors will focus on specific "horizontal" processes or processes unique to particular verticals.

Knowledge worker AI adoption: The large platform vendors will increasingly push AI tools at knowledge workers, often even when their employers are reluctant. Microsoft users now see Copilot everywhere. Knowledge works will eagerly embrace AI to streamline mundane tasks like creating a first draft or marketing copy, or summarizing a document or creating images for a website. In the same way that Microsoft and Box hooked individual knowledge workers once they realized they could use SharePoint and Box to collaborate, knowledge workers will be reluctant to relinquish these tools. And that’s just the beginning.

The way that process-specific innovation and knowledge-worker innovation chaotically intersected, converged and overlapped multiple times during the first revolution will repeat itself during the AI revolution. But with even more far-reaching implications.

A final thought.

Sangeet Paul Choudary's work provides some of the best examples of how processes and knowledge workers will collide in the AI era.

He notes that previous automation may have reduced headcount for specific functions, but humans ultimately remained in charge of processes and workflows. However, as AI agents begin to unbundle workflows, they will not only automate specific tasks but also find additional automation opportunities — on their own. Ultimately, AI agents will be able to redefine the workflows themselves and the roles and goals associated with those workflows.

"We frequently make the mistake of thinking of AI as 'just another technology'. We arrive at the vague conclusion that AI will automate some jobs and augment others. But there's a larger, less understood, nuance to understanding the potential of AI. AI – particularly autonomous AI agents – are goal-seeking. Goal-seeking technologies are unique. They take over planning and resource allocation capabilities, and in doing so, restructure how work is organized and executed.” (Sangeet Paul Choudary, “How AI Agents Rewire the Organization”)

Fasten your seat belts. It's going to be quite a ride.