How AI Will Change Collaboration

How AI Will Change Collaboration

Cover Photo by Google DeepMind on Pexels.com

[Note: How AI Will Change Collaboration is an early version of a report that will be published following interviews with collaboration vendors and AI companies.]

Collaboration remains an essentially human task. Generative AI challenges notions of human-centricity. As I re-examined my 2024 Collaboration Portfolio: Features diagram, I found that it holds up to the injection of AI into collaborative workstreams. 

The diagram already included AI assistants, analytics, and discovery, all of which had some level of presumed intelligence underlying them. I did add a new core service called co-creation/analysis, which more precisely aligns with generated content, summarization, and co-creation, such as content feedback and re-writing.

How those services manifest themselves will change the nature of each branch of the collaboration portfolio. This post attempts to cover all the ways collaboration features will likely employ generative AI in the future. 

I did not include general areas, such as formula writing in Excel, as it fits into an even broader future of work domain around AIs’ impact on individual competencies and capabilities. Regardless of the tool or the domain, AI will play a role in helping people better use their tools. Collaboration tools will not be an exception.

This post, however, will focus on direct collaborative activities that will benefit from AI as part of a collaborative engagement, which, at this point, I believe will remain an engagement that takes place between two or more people (more on that later). I do not see the current generation of AI being capable of managing entire workstreams, but I do see it shortening the workstream execution durations.

The text is organized by the major branches on the 2024 Collaboration Portfolio: Features diagram with references in the analysis to the individual features.

Meetings

Even before generative AI became widely available, earlier implementations of intelligent features included transcription, simultaneous translation, real-time closed captions, capturing to-do items, and voice recognition. Computer vision offered security with capabilities like sensing a speaker, monitoring for lurkers, and adjusting the camera based on the participant’s environment. Intelligent audio contributed noise reduction and voice-optimization features.

Many features are often touted as “AI,” especially since the release of ChatGPT, but they are neural adaptive technologies that I refer to as autonomic intelligence, meaning that in animals, these types of brain activities happen autonomically, with little or no conscious cognitive processing.

Here are some of the ways generative AI may change meetings:

• Automated agenda development for project meetings. Based on the previous agenda, items including project plans, project output (with status reports), and the state of assigned tasks from previous meetings, should provide the basis for generative AI to create good agendas for meetings that focus on what is important to a project based on its description and intent, and what is important to attendees.
• Automated agenda development for non-project meetings. Non-project meetings will demonstrate generative AI’s strengths and weaknesses. From the strength perspective, generative AI will likely offer a suggested agenda for a wide range of meetings, from morale events to product kickoffs to large conferences. However, unlike project meetings that include a regular cadence of new content as input, a one-off meeting agenda may prove less aligned with an organizer’s vision without significant input to generative AI to describe intent and desire. It may remain easier for an organizer to create an agenda rather than describe the parameters for the meeting.
• Adaptive agendas. If a discussion veers a meeting away from the agenda, generative AI can adjust the agenda, perhaps by removing lower-priority items or moving topics to future meetings.
• Brainstorming and ideation. While this topic has its own feature under content creation, it is often an item on a meeting agenda. With virtual whiteboards and creative spaces, generative AI will offer brainstorming prompts, leverage existing ideas to suggest associated ideas that don’t appear, and will likely help organize virtual whiteboards, offering “clean-up” creative spaces on demand.
• Automated meeting minutes that include summaries of reports, links to referenced content, and attendee information. AI will likely also be able to generate outputs in different formats, for instance, producing a mind map of a meeting rather than a simple linear report out.
• Automated task assignments. Not only will generative AI identify action items, but it will also translate them into actions assigned to individuals in their task system of record.
• Inviting people to collaborate. While privacy and other restrictions may curtail this feature, ideally, AI will be able to identify subjects, topics and concepts across a variety of collaborative spaces, suggest synergies, and make introductions. This could help reduce duplication while also increasing serendipitous encounters. Group chats, team spaces, and persistent chat would benefit from this application of AI.
• Avatars are already available as replacements for video. AI will likely offer the ability for an AI surrogate to substitute for a person, within limits, answering questions of expertise and status. Personal AI models will also allow former employees, including deceased employees, to continue to contribute. 
  AI-based avatars will likely raise legal and ethical issues, but those will be worked through over the next several years, eventually leading to permission not only for the living employed to create models but for organizations to model employees and retain those models after their departure, much as they would any other data associated with a former employee.

Current AI features, such as transcription and translation, will continue to improve, forming the foundation for continuous learning as the outputs become not just historical artifacts or catch-up tools for those who missed a meeting; they will become part of the knowledge graph that improves the AI’s view of the organization, its activities and its people. 

The translation of transcriptions and their incorporation of audio, and perhaps other output like Braille or sign language, will make meeting activities more accessible.

Cooperative Content Development

Much of generative AI’s appeal starts and ends with content. It starts with content as the underpinning of its foundational models, and its output is often aimed at the creation of new documents. The following areas will adopt generative AI features, in most cases, completely transforming the experience from one of human creation through an application to co-creation with applications.

• Dynamic Content Generation. Generative AI empowers content creators by autonomously generating diverse forms of content, including text, images, and code. This capability will accelerate content creation processes, enabling creators to explore suggested ideas and incorporate them into various forms of content. Content creation will require appropriate prompts in order to generate contextually accurate content despite generative AIs’ semantic understanding. Without proper prompts, generative AI will be unable to ascertain the intent of the creators. User interfaces will likely emerge that manage content generation prompt parameters and controls.
• Context-Aware Writing Assistance. Generative AI analyzes context and provides real-time writing assistance, suggesting improvements, enhancing clarity, and ensuring coherence. This feature will need to mature and become more aligned with personal writing styles—it can be a more creative partner, rather than a “compliance-based” one. Current models, such as Grammarly, often attempt to eliminate nuance, style and subtly from writing. Domain knowledge will also need to improve, so that writing assistance more easily handles overloaded terms, like “strategy” which it often fails to recognize as an object with no need for an article.
• Automated Summarization. Generative AI excels at summarizing extensive content, condensing information while retaining key insights and making it suitable for diverse audiences. Most long-form content will be automatically summarized in the future, with summaries acting as a gateway to content. Summarization may also include automatically interpreted structure, allowing readers to more easily access the parts of a long document they want to read, rather than working through it after its discovery.
• Multimedia Content Enrichment. Generative AI will be able to enrich content with multimedia elements, suggesting relevant images, infographics, and visual aids. This feature will reduce the time required to make content more visually attractive. Organizations will need to monitor for copyright issues to ensure that retrieved or generated content comes with proper licensing to support the target distribution.
• Adaptive Language Styling. Generative AI understands different language styles and adapts its output to match the desired tone. This will allow content creators to maintain consistency with audience-tailored communication. Organizations will also be able to leverage this feature for managing brand identity.
• Real-time Translation and Localization. Generative AI will facilitate real-time translation and transform content localization. As generative AI matures, the roles for content localization will likely be impacted, with AI tools doing most of the work for static content and on-the-fly needs.
• Customizable Content Templates. Generative AI offers pre-designed content templates that creators can customize, providing a foundation for various types of materials. This will accelerate content creation workflows by reducing the need to generate content prompts from scratch.
• Document Assembly. Regardless of the document type, generative AI should be able to assemble documents, not just generate them. The legal world is already pushing in that direction with contracts and other legal documents. AI would draw from repositories, and rather than looking for the right word order, it would look for larger chunks like paragraphs, perhaps entire pages, of pre-approved, current content that it stitches together into a single document.
  Document assembly may use a combination of pre-defined content components and generative features to create multi-part documents, like proposals, customized pitch decks, and client-tailored assessments. Document assembly could also be a transformative act, such as turning a persona-focused messaging architecture into sales enablement content. It is highly likely that AI-generated layouts will also be part of the document assembly workflow.
• Intelligent Content Expansion. Generative AI will assist in expanding content by suggesting additional sections, relevant details, or complementary information to enhance completeness. Content creators can ensure comprehensive coverage of topics, adding depth and value to their materials with AI-generated insights that go beyond grammar word counts and tone to offer conceptual, reasoning, and argument support.
• Contextual Feedback. @mentions within Word solved a collaborative issue for people working with a document at a granular level (words, even letters). If generative AI bots become part of the directory, they could be called upon via @mentions to provide specific contextual feedback within the body of textual documents, spreadsheets, and presentations. They will likely also be able to do the same for video, images, and audio, though the collaborative environment for multi-media files is not as sophisticated as it is for general office documents.
• Brainstorming and Ideation. AI will often be the starting point for brainstorming and ideation, eliminating the dreaded “blank” board or sheet paper. Current implementations in tools like Miro and TheBrain use generative AI to populate topics for ideation, brainstorming, analysis and tasks. Generative AI can create bulleted lists for many areas, and integrating creativity tools transforms those bullets into jumping-off points for further ideation.

Content Sharing

Content sharing essentially manages the post of content to a file repository. Much content remains in the realm of personal file stores and simple blogs or posts. The feature does not imply content management, which is the next feature branch. It could be argued that it should, but because many collaborative content projects don’t end up in managed repositories, I leave this feature as a standalone for now.

As for AI, it would be ideal to not only incorporate this feature into content management, but for AI to manage transformation and management.

Automated workflows. Imagine that this blog has been completed as a text document, however that happened. WordPress is its intended target. I should be able to authorize an AI to migrate the content to WordPress, apply appropriate design styles, incorporate either royalty-free images or licensed images from an authorized repository, optimize for SEO, add tags and categories, and publish the blog. I may want to review a pre-publication draft, but it should be 99 percent ready to go.

The same process would hold true for any target platform, from Instagram to LinkedIn, from Threads to X. And yes, Jetpack and other tools offer social sharing upon publication, but they don’t do anything to optimize those shares. 

With AI, a piece of content could act as the impetus for a mini-campaign, where the AI would generate multiple social posts of the initial content, much as an author would manually craft intriguing posts to drive traffic back to a longer blog. The same techniques could be used within enterprises to drive relevant content to message boards and promote it on team pages.

Content stores vs. content posting bots. This is an area where a content arms race may develop within enterprises between social content targets and the tools used to post content.

Imagine this scenario: Personal or team AIs attempt to post content more frequently, or less on target, than desired. Repositories using AI to manage the acceptance of posts based on their appropriateness and politeness (not over-sharing, perhaps) reject overly aggressive or untargeted posts. Posts may be evaluated for currency, relevancy, or other attributes to determine if they are worthy of being posted. Inside enterprises, aggressive postbots versus AI-based content guards will not likely spur an arms race. On the Internet, they will.

Marketing teams are likely to encounter post management issues in campaigns. Social media will prove more amenable to posts as their metrics are more about engagement than relevancy or accuracy, but every target platform will have its own policies about how to deal with assertive bots.

Content Management

Generative AI will likely revolutionize content management in several ways, including automated classifications, new approaches to retrieval, and challenges to what data is managed where and what versions of data get managed.

•  Automatic Document Classification will automatically classify documents, reducing the need for user-developed taxonomies and ontologies. This will range across all document types.
• Innovative Content Retrieval Methods. AI will also offer many new ways to retrieve content that leverages its models and parameters rather than those assigned via tags and taxonomies. This will be an evolving space over several years as techniques emerge.
• Dissonance in Repositories. Perhaps most importantly, generative AI will introduce a dissonance between application development repositories and content management repositories as organizations decide how best to manage models and other data related to generative AI.
• Metadata Challenges for Chunked Content. Other questions will likely arise, such as metadata related to content that has been chunked and prepared for Retrieval-augmented generation (RAG). How will the original documents be updated to reflect that they exist in internal generative AI extensions, and will their “chunks” be registered as a version of the document?

Messaging

Messaging will act as an interface to chatbots. AI will initially serve as an augmentation to message creation. It may ultimately eliminate human-to-human generated messages as it delivers value to collaborative spaces that offer more context than e-mail or other forms of messaging.

• Writing assistance. Like various synchronous and asynchronous meeting tools, messaging, like e-mail, chat, text, and social direct messages (DMs), will include writing tools, though hopefully, people aren’t using these tools for long-form content that require generative AI intervention to make it comprehensible.
• Consumer bots. It is more likely that consumer versions of these tools will host specialized generative AI bots that answer questions about banking, ride queues at Disneyland, and what stocks to buy, along with the go-ahead to execute buys.
• Enterprise bots. Within enterprises, similar technology may include campus navigation, meeting requests, content summaries and links for requested content (search via a message), and advisory bots with specialized or generalized knowledge of enterprise business domains, policies, practices, and procedures. 
• The end of e-mail? It isn’t clear if generative AI will finally kill e-mail and move enterprise collaboration to collaborative workspaces and their internal messaging functions, which is much more efficient when used well, but it should minimize messaging within the enterprise as it takes over the reason people write e-mails with more proactive mechanisms.

Work Management

Current generative AI is not very good at planning, nor is it particularly adept at execution. The content publishing scenario described above requires intimately connected systems, from Word to WordPress, from LinkedIn to Instagram. The AI would need to understand how to create content that matches the format expectations of those systems, as well as their parameters. Those all exist as APIs for the most part. For AIs to perform automated actions, they will need to master those APIs.

The rise of LAMs. While the nascent Large Action Models (LAM), such as those displayed in Rabbit, suggest generative AI can navigate data interfaces between apps, the security models are primitive and restrictive. LAMs would need to secure extensive interaction among business applications in order to automate processes at a meaningful level.

As an example, a command could be as simple as hitting a button within Microsoft Word that says, “Publish to WordPress.” No need for prompts or prompt engineering. The AI would already know the author and the target, and it could derive the metadata from the content.

This kind of process automation eschews rules for the most part, though setting up the AI would, as is done today with the likes of ChatGPT, require some level of custom instructions. And again, depending on privacy and other constraints, those should be derived assertions. I should be able to, for instance, point a generative AI tool at my LinkedIn profile and my website, and it should build a better model of me than one that I incrementally build manually.

I think much of the “how should ChatGPT respond” parameters will be contextual, set via a user interface rather than as text, much as Microsoft is starting to do with the Copilot in Edge.

Adaptive UIs. We are getting ever closer to a vision I proposed over twenty years ago: adaptive user interfaces. If you think of applications as data, they can also have rich metadata. An intelligent interface could sense a need (or have it explicitly stated) and assemble a set of features that would solve a particular problem. 

Although PowerPoint, for instance, includes “Smart Art” in today’s AI world, that art isn’t so smart. The need might be for some well-designed graphics, beyond Smart Art, and for which the current AI layouts prove insufficient. A program like Wondershare’s EdrawMax has the right features, but it is currently only peripherally associated with PowerPoint (able to explore its drawings in PPTX or a compatible drawing format for import).

EdrawMax, however, already has some AI. What if that AI was able to take a handoff from PowerPoint to create a graphic from a table in PowerPoint? There is no reason that the AI could not manage the Windows-level memory or clipboard exchanges to transform the table into a graphic that is built in EdrawMax and transferred to a PowerPoint slide. Yes, for those with a long memory, some of this is reminiscent of AppleScript, which offered the mechanisms without the intelligence.

Imagine the following prompt: “Turn this table into a circle graphic with related icons above the next. Each segment should be a different pastel color, and the icons and text should be white.”

If the AI was unsure of the assignment, it could ask questions like, “Column 3 seems repetitive and would complicate the graphic; what would you like me to do with the data in column 3?” This is a question that a graphics designer might ask of a line-of-business owner trying to turn data into meaningful graphics but not really thinking through all the implications of the data they have at hand. Graphics design could be collaborative with textual dialog initially, followed by visual feedback, such as, “Regenerate the chart using the updated visual in segment one as a model.”

Currently text has become the UI to tools like ChatGPT. That is because they have been implemented as chatbots. That need not be the case. Other inputs could be used to trigger generative AI actions. We already know that images, video, audio, text and code can be generated from generative AI tools. 

Beyond the Task at Hand

Work management will need to eventually include plan generation and project management (tasks management, reschedules). This will require generative AI to understand human time, interface with calendars and generate Robotic Process Automation (RPA) instructions.

This level of integration requires deep connections between applications and the free flow of data between them.

Collaborative decision-making. Collaboration isn’t just about creating or sharing content, designs, or experiences. It is also about decision-making. Generative AI will support decision-making across the organization, including in the C-suite. Wherever there is data, asking an AI what it means and what it suggests the organization learn or do based on the data will be irresistible at first and, depending on the results, inevitable or abandoned.

Unfortunately, some early work like the Harvard Business School study documented in the paper, Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality, suggests that as much as AI offers productivity improvements, it can equally create false senses of security, even behaviors such as defending poor data.

While AI may offer insights, even novel solutions, its suggestions must undergo the same scrutiny, or more so, than suggestions from humans on the same topics.

AI should be used to encourage and elevate human decision-making, not supplant it.

People/Directory

Discovery. The first AI topic for people/directory that usually comes up is discovery. Find people related to the work I’m doing or the content I’m creating. A directory as a source for learning and sharing.

Maintaining profiles. Most directories suffer from poor maintenance. People don’t keep their profiles up-to-date, and therefore any discovery available runs off old data. With generative AI, profiles could be constantly rewritten to reflect current work (with permissions in place for those working on projects that other people aren’t supposed to know about; even in these cases, abstractions of skills and capabilities could be incorporated into profiles without giving away secrets).

AI as a directory member. The aforementioned generative AI bots would also have directory entries so they could be mentioned, and people could access them through other directory-driven features like messaging.

Collaboration Reconsidered

Collaboration remains an activity of two or more people. The future may hold that we think about two or more entities. Ultimately, the two or more entities may both be AIs, as it is likely that the cost of models will become more expensive, and therefore, rather than ever more inclusive large models, more specialized, smaller models will evolve, and those models will leverage their interaction skills to collaborate with one another.

In the near term, however, AI will act as a third wheel in a collaborative setting—a scribe at times, a contributor, a critic, a curator, and perhaps for some, a counselor or confidant. 

Every collaborative act will include an AI component just waiting to be engaged. It will react quickly to queries, creating and critiquing concepts and content, designs and desires. As ubiquitous as it may be, it will remain an option—it will require a human act to engage it. In some cases, that human act will be thrust upon other humans, such as when a content management system adopts AI-generated summaries for all queries. In others, it will sit aside on the user interface, patiently waiting to be called upon.

And we will need to call upon it. We understand the power and the promise. The only way to understand how to benefit is to engage. To test and prod, challenge and reflect.

Generative AI will likely take unique use cases to a new level. Each person will have their own versions of AI that they collaborate with, that get to know them, and that best reflect their quirks and biases. We will continue to anthropomorphize this invention, making it intimate and personal.

Our PCs and our smartphones weren’t smart enough to let us lead them there. We personalized them to an extent, often to the point others have difficulty navigating our taskbars or application folders. But once in an application, save for a custom menu, most applications work the same regardless of their user.

That will not be true for generative AI; that it generalizes and adapts to context will mean that personalized versions will exist. Rather than two or more people in a collaboration or two or more entities, we will always start with at least four for humans, as each will bring their favorite AI for comfort and completeness, perhaps even for competitiveness.

The future of collaboration won’t just be about multiple people converging on a problem and seeking the best solution, but in most cases, the history of humankind descending on the problem. The best users of AI will understand that collaboration is as much about asking the right questions as converging on the right answers—and to ask good questions, one needs to know the capabilities of the source to which they pose those questions. Generative AI may well drive people to be smarter, not dumber, as the only way to collaborate with an intellectual bon vivant is to understand enough about the world to have a good conversation.

For more serious insights on AI, click here.