Day: November 13, 2023

Up to this point in our discussion of transparency, we have been talking about things that are very real technologies. Today, I want to take a small step into near term reality based on current tools.

So far, we proposed opening the doors of information locked away in proprietary boxes, putting that information into perspective through concept mapping, and constructing linked canvases of ideas through layered concept mapping. This week, I want to fuse these ideas with the technology of the moment, Generative AI.

Information processing is a labor-intensive process. Whether we’re creating documents for large companies, or just trying to keep a handle on our own thought processes. I have filing cabinets of semi-sorted paper, hard disk drives of data, and thousands of pages of my writing.

Since the days of HyperCard, one of my chief struggles in life has been making logical sense out of all the information I collect. I have found very few shortcuts and have massive data forests to contend with.

Even more frustrating is when my brain tells me I need a particular piece of data and I can no longer find it. I have tried various tools over the years to organize that data going all the way back to HyperCard, but maintaining those tools was a labor-intensive process. Even after I put the data into the tool, it quickly became dated and sometimes inaccessible.

When I write, I know I am getting knowledge from a vast range of sources. Sometimes those are explicit, such as my current recollection of Vannevar Bush’s Memex,. Often, however, years of research shape my writing in subtle ways. Sometimes I can figure out those connections with a great deal of work, but sometimes it’s just too hard to backtrack my thinking that far back. Even if I can, it’s a major effort to locate the original source.

Generative AI is making a vast array of connections as it creates documents. It does this by scraping data from a range of sources. Like the bread crumbs on the web, the connections to these sources are hidden. This practice has caused considerable controversy among content creators. I asked ChatGPT “How can we combine Generative AI with concept mapping to create connective maps of the world?” This was its response in text form:

ChatGPT has produced some interesting responses, but I want to explore where it’s getting its information from and how we might connect those ideas into a more comprehensive approach to mapping ideas.

This should be possible. Generative AI is an augmented connection device. However, like the web, ChatGPT hides those connections.

If we made Generative AI transparent, it could also be an augmented perspective device. I took those ChatGPT results and plotted them on a Miro concept map.

This map represents an iterative step. There is nothing new being added to the map compared to the ChatGPT response. I Generative AI could generate this kind of map. A further iteration would add sources and applications as well as suggest connections between items on the list. Right now, humans (like me) must do this part.

Coming at this from complex to simple patterns, one of the more interesting tools I found in recent years is the Open Syllabus Project. It plots a heat map of all the readings from open college syllabi. It forms a knowledge heatmap of what we teach in higher education through the materials on our collective syllabi.

There’s a lot about this diagram that is automated, but it still requires a fair amount of labor to set up and maintain. However, I could see this kind of tool being fused with AI to create automatic heatmaps from any collection of data.

I can imagine a host of uses for this kind of tool. As you may recall from the previous blog, I created a connected set of ideas and projects for my Miro resume.

This was a tedious and labor-intensive process, even though I had all the data I needed from IdeaSpaces. Computation excels at replacing tedious labor.

These are just three examples of what is possible with AI tools that automatically aggregate datasets and propose linkages. Consider for a moment just how much of today’s professional work is spent doing just that.

I used to work at an architecture firm. Most new architects spend huge amounts of time creating what are called construction documents. These connect the design to the materials, furniture, fittings, and equipment necessary to build the building.

This is tedious work. It is also a task that consists almost entirely of making connections between various bits of data. Generative AI is made for these are the kinds of tasks.

This would disrupt the architectural practice as it exists today. Creating construction documents is a central part of the process for breaking in new architects. Demand for junior architects would decrease.

However, this is also an opportunity for firms and schools to up their game and bring architects into practice at a higher level than is currently the case. Also, someone is going to have to manage this AI process.

On a personal level, I would love a Large Language Model AI that would look at my writing and suggest where I’m getting my ideas from or to provide suggestions for additional exploration of adjacent ideas.

This would automate that frustration I described earlier. Probably 90% of the resources I have accumulated over the years have been digitized, either by me or by some other entity. Connecting them is the real challenge.

This kind of tool would also tell me if someone else has explored the same territory. I can’t tell you how many times I’ve been writing and thinking “surely, someone has thought of this before.”

Concept maps are an ideal tool for surfacing the connections that the AI finds. Instead of providing a list of sources or materials, an AI coupled with a concept mapping tool like Miro could give us a connective heat map that is both live and evolving. Humans could then focus on creativity and novelty.

This brings us full circle back to the blog from the beginning of this series. Imagine an AI that creates a concept map or a heat map of all the connections, while constantly pulling from live data. We could use this to fine-tune the AIs. Visualization also creates opportunities for oversight to mitigate bias or other kinds of intellectual corruption from occurring.

We could use these kinds of maps for regulatory oversight to spot areas of mistakes or deliberate gaming of the system. A generative system, coupled with a visual output, could create these kinds of documents automatically, making compliance with government oversight much less onerous and labor-intensive than today.

If we can get past the alarmist rhetoric and look at the possibilities for human augmentation that these tools provide, we can unlock stores of overlooked or hidden information to improve our societies, to advance science and knowledge, and to help us overcome the very real challenges we face as a species today.

Information is at the core of everything. Learning has always been the killer app for the human species. It’s what gave us key advantages in our revolutionary struggle. The explosion of information over the last 40 years or more has made comprehending what we’re seeing more difficult, not easier. This is because we optimized technology to distribute information, not connections.

It is time to take the next step and connect that store of information to our human abilities to make connections and find patterns. For that, we need to think differently about the tools at hand, and the tools we should prioritize building. Perspective creates wisdom. Wisdom is in short supply these days.

The real issue in software design is the design of ideas. But most people are looking at the wrong levels, fixating on particulars, and not seeing the immensity of option – or the imperative of cleanly condensed structure. – Ted Nelson Dream Machines, 2nd ed., p. 70

Ted Nelson is interested in modeling ideas. Building on the work of Vannevar Bush, he saw the potential of computing technology to transform how we see and manipulate information. As I argued in a previous blog, seeing information in context is at least as important as having access to it. Information is not transparency. Perspective is.

Nelson understood this key facet of transparency. His lifetime of work has reimagined connections between different parts of our atomized information network. In his vision of the web, which he calls Xanadu, he sees a floating network of ideas connected bidirectionally through a series of hyperlinks, a term which he coined.

However, when Tim Berners-Lee invented the World Wide Web in the late 80s, these connections became buried. The web that emerged after Berners-Lee lost the critical trail of breadcrumbs which were central to Bush’s and Nelson’s vision of a network of ideas.

On the World Wide Web, the information itself took precedence over its connections to other information. The paper metaphor also persisted, and with it, the limitations of text. Berners-Lee recognized this shortcoming (see his book, Weaving the Web). The technology of the time limited the web he created.

Xanadu represents a practical implementation to realize Nelson’s vision of free-floating information bits and the mapping of the hidden connections between them. This is both conceptually and technically difficult. Nelson was ahead of his time, but I can imagine that it could be possible with AI and XR augmentation to approach his ideal of information transparency (more on this next week).

Concept mapping, however, excels in highlighting connections between disparate pieces of information and ideas. By putting things into perspective, and then allowing us to shift that perspective on a canvas, concept mapping can provide powerful insights into the workings of our own brains and the collective workings of brains in a distributed group.

Perhaps concept mapping can form a bridge between how we now manage our information stew and Nelson’s network of connected ideas. Note this illustration from Nelson’s Geeks Bearing Gifts (Mindful Press, 2008, 2009). Nelson seems to envision a concept map at the top of the graphic, but the connections extend vertically instead of just horizontally.

The Generalization of Documents from Nelson, Geeks Bearing Gifts

The limitations of concept mapping is that it represents a two-dimensional canvas, instead of the three-dimensional space that Nelson envisioned. And unless you engaged in the exercise with a group in a room, the technology limited concept mapping to mapping your own ideas.

I discovered a new trick with Miro that allows you to make links out of objects. We can use this technique to link to outside references like documents on the web but, more interestingly, we can also link to additional canvases within Miro. Using this trick, you can create interconnected canvases of connections. I have done this with my resume.

We can use this third dimension of interlinked canvases to explore and flesh out different subsets or rabbit holes of ideas. We can also link these to the outside web (knowing that’s a one-way street).

As in Nelson’s diagram, we can extend a two-dimensional idea space downward or upward into infinite interlinked two-dimensional idea spaces. This resembles the web as we understand it today.

However, there are two critical differences. First, it makes creating “pages” of connections almost effortless and accessible to non-technological users.

Platforms such as Miro are accessible with minimal instruction. I’ve worked with groups that created complex maps in an hour with no prior knowledge of how to use the software or even the idea of concept mapping.

Second, if you create a root level that requires the illustration of connections between these subsets of ideas, you have begun to create what Vannevar Bush referred to as breadcrumbs.

Concept mapping brings the pathways and connections to the forefront. A root guide or group of guides can create the top layer of the nested idea diagram. Other group members can then expand that core map horizontally or into linked canvases. You can even label these connections so you know who created them.

This highlights another unique aspect of a tool, like Miro. It is a collaborative and asynchronous activity. Groups of people can work together on root-level documents, subgroups of people can work on connected subsidiary documents, and all groups have the choice of working synchronously or asynchronously on a persistent canvas or sets of canvases.

I realize this only goes halfway to the environment Ted Nelson envisioned for Xanadu (again, for that we need AI and XR). However, it uses a tool that already exists, is accessible, and enhances the connections and perspective essential for achieving the transparency I wrote about in my last blog.

There is also a fourth dimension possible here. By creating persistent, adaptive idea spaces, we can see how ideas and systems evolve. I’m pretty sure that Miro does not have a history feature, but that might be something the company should design.

One of the key elements identified by Vannevar Bush in “As We May Think” is understanding how we got to where we are. Perspective over time is at least as important as perspective in space when making connections. Both aspects are critical to transparency.

As an example of how this is done, let me return to the blog map of my writing that I created earlier this year. If you go to the map now, you see that I have turned each blog title into a link linking back to the original blog.

I have also turned the reference list in the center into live links as well. What I have done here is to trace pathways anyone can follow between my work and the ideas of others.

It is up to the reader which pathway to follow. The visualization makes the connections obvious. This example doesn’t illustrate the potential for groups to create similar documents, but that is a question of scaling.

Transparency requires seeing. Seeing requires perspective. We now have tools that help us weed through that complexity, but we need to be creative about how we apply them to achieve our goals.

There are a lot of ways to assemble these interconnected idea maps. We could start with a central goal, and then have groups or individuals work outward from that goal. We could create a node of shared understanding and then work backwards from that understanding to its roots. Or we could start with a complex system map like this one and then use it to explore how that system came to be and where it lost connection its connections to human learning.

The Disconnect in Higher Education

These explorations can become very meta quickly. Groups could explore how they explore, for instance.

Before we can find our way out of the information thicket, we need to make maps that help us navigate the forest. Layered concept mapping could turn into a very useful tool for mapmaking.

New tools are on the horizon, which could augment this effort even further. Generative AI is all about making connections. I could envision mapmaking AIs that could form basic maps which groups of humans could enhance further in a symbiotic partnership. But that’s the subject for the final (at least for now) transparency blog.