For anyone that’s been watching this space, you might have noticed that I’ve been rather quiet for a while. Some of my last posts were exploring the impacts and uses of AI in consulting, collaboration and public engagement.

My early experiments quickly convinced me that AI tools offered significant value in collaborative work but I didn’t see a way to keep it as a core tool in my practice with the options that were available. What I did see, though, was that with the right tools, we could finally overcome some long-standing challenges in collaborative work to enable new forms of collective action and complex decision-making.

Goal17 is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Through several decades of work in designing large-scale, complex, multi-stakeholder decision-making processes, there were several persistent problems we’ve been working to correct for that finally seemed within reach, and, to me, offered the possibility for a real leap forward in how we organize ourselves around complex decisions.

Whether in large organizations, in projects or across coalitions, the quality of coordination and decision-making has always been a function of how quickly information gets where it needs to be, how widely distributed it is, and whether those in authority are aligned in their understanding.

What became clear very quickly was that we could improve the speed of information passing through teams and networks, ensure that “edge insights” become more distributed and bring groups to shared understanding more efficiently. As I tested recording in breakouts and across sessions, I could see that assumptions we had held about the quality of what groups would report about their work paled in comparison to the richness of complete capture.

I felt that if you could synthesize disjointed, distributed or parallel work quickly enough, at scale, that this would no longer be a question of documentation after the fact, it would completely change the speed and comprehensiveness of collaborative work.

Consultations would be able to not only capture in minute detail the perspectives of constituents, but could adapt their focus - at scale - as unexpected themes emerged.

Just off the top of my head:

* Strategic conversations could merge the views across all stakeholders and document inputs.

* Workshops could iterate quickly on emerging ideas and save time on lengthy reports between teams.

* Conferences could become knowledge-producing hubs.

* “Systemic” conversations and policy processes could assemble first-hand knowledge from more people than was ever practical before.

I was convinced of the potential, and I remain convinced. Anyone in professional services that is involved with engaging and assembling knowledge, perspectives and intent across large groups will, I believe, fall behind if they don’t account for this step-change in capability. However, I am not a blind AI fanboy. Ethics and modelling outcomes have been central to my practice for years, and I have been a long-time critic of how the unprincipled development and application of technology has had predictably dire outcomes for society. Having previously delivered AI ethics and governance projects, I also knew that rolling out something like this would require making decisions with very real consequences.

I am not going to go into detail here on how, exactly, I addressed each of these issues. Many are worthy of future posts. This is not even an exhaustive list. Suffice it to say, however, that in my view, AI platforms and any prospective users of those platforms, need answers to these questions and many more. During development, I’ve noticed a deep chasm between those with technical expertise and those who are shaping policy and engaging in discussions on ethics. Part of my motivation in developing this platform has been to create a kind of “thesis” project that is fully production-ready, that is a technical reflection of many of the more philosophical, ethical questions relating to this technology. It’s why my writing has lapsed in this space for a while: I was writing something else.

Now on to some of the issues and questions I addressed in my design.

Privacy

Listening to conversations at scale creates a host of privacy concerns. If “collective intelligence” is the goal, then trust and the confidence of individuals to speak freely is a precondition. If they are suspicious of the recorder, they won’t share.

* What happens with the recordings, and who can listen to them?

* How will personal side conversations be managed?

* Where are the recordings being processed, and what are the policies of the processor?

* Are statements being personally attributed, and if so, who has access to those attributed statements?

* Are my conversations being used to train models, build someone else’s product, or serve someone else’s secondary business model?

Security

If these conversations are confidential, proprietary, or otherwise sensitive, there are a host of questions that emerge end-to-end in terms of how data is gathered, transmitted, processed and stored.

* Can the radio frequencies from any recording devices be intercepted in order to eavesdrop on in-person conversations?

* Do recordings stay on any recording media between or after meetings?

* If recordings are being transmitted, can the data be intercepted in transit?

* Can AI models learn, retain or leak any of the materials being processed beyond the scope of a given group?

* Is all original, processed or derived data stored securely?

Cognition

Artificial Intelligence, when applied to cognitive work, can either augment human performance or systematically degrade it. Especially when applied to difficult, strategic conversations about the unknown, AI can adversely impact the ability of humans to tackle thought-work.

* Is AI being used to replace creativity, decisions or tasks requiring judgement?

* Who is generating recommendations or options - humans or AI?

* Is the “work” being done by the AI assistive, or authoritative?

* Does the application of AI generate additional cognitive load through over-generation?

Human Agency

AI systems can be extractive and erode the importance of human roles, or can be assistive and extend human capability.

* Is the role of AI generating new value in the work being done by humans, or replacing existing roles previously done by humans?

* Does AI draw from outside or unverifiable sources that compete with qualifying knowledge and expertise of human operators?

* Is AI being used to make judgements that carry liability or normally demand formal accountability?

* Is the role of AI creating dependence that might diminish the critical faculties of human users of the system over time?

* Who decides what is important, and are those decisions clearly and transparently recorded?

Sovereignty

The development of AI is taking place on the backdrop of renewed geopolitical conflict, and control of the technology, infrastructure and data involved is now subject to state surveillance, restriction and control.

* Is any part of the AI system - from collection, to transmission, processing and storage - transiting a potentially adversarial jurisdiction?

* Are any critical components of the system vulnerable to import or access controls that might render the system inoperable?

* Does the system rely on the goodwill of foreign or adversarial governments?

* Is any part of the system in a jurisdiction that can compel involuntary access or disclosure?

* Is the system in a jurisdiction that will credibly enforce stated privacy and content protections?

Trust

Many of the leading AI labs have systematically violated existing copyright laws and are backed by governments that prioritize rapid growth and technological supremacy over human, labour and privacy protections. Further, frontier labs have opaque cost and pricing structures that do not appear to align with publicly available estimations on delivery cost.

* Does the AI provider have a verifiable set of protections in place for secondary uses of customer data?

* Has the AI provider ever been found to infringe on the use of data they do not have rights to?

* Do the AI provider’s prices accurately reflect the cost of delivering services, such that current prices are a reliable predictor of future prices?

* Has the provider ever been seen to misuse, mishandle or misappropriate private information?

* What assurances do users have on the accuracy or provenance of information provided by the service?

* Are the stated aims and objectives of the provider or its leadership aligned with the interests of its customers?

And So…

Where did this leave me? There was, somehow, nothing I could find on the market that could record 20+ conversations in a room simultaneously, and no provider for processing all of that information in a way that satisfied all of the conditions above (and many more).

So, that is why I built Knowledge Commons and the Raven Recorders. I could sit on the sidelines and complain about how the AI industry was taking the wrong path, or, I could build something that demonstrated what I thought the right path should look like.

Goal17 is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Get full access to Goal17 at goal17.substack.com/subscribe