AI Policy

The word ‘policy’ suggests something that’s set in stone, quasi-legal. Nothing about our current era is that solid. Models, tooling, and norms are shifting daily, and with each of these leaps we see new and unintended consequences pooling around our ankles.

So this isn’t a policy, yet, so much as a collection of hunches and postures that will evolve with time. Maybe it will crystallize into something harder and more directive, or maybe we’ll keep continue to tend and prune this document as it reciprocally tends and prunes our practice.

A General Posture

In general, we are trying to seek a posture of ‘productive tension’ when it comes to our relationship to AI, both in our practice and in the services we design for, through, and with. Here are some statements that sum up our point of view (and gesture towards those tensions!):

  • AI is already changing large swaths of the knowledge and service economies (and probably will change the ‘knowledge’ part of non-knowledge sectors, too).
  • As a firm that sells expertise about services and the way they are brought to life, it is our responsibility to engage deeply with this technology. This engagement is not an endorsement.
  • AI is capable of outputs that verge on magical and is also fallible in ways predictable and unpredictable and with potentially life-changing consequences, especially for already-vulnerable people
  • The widespread usage of AI will democratize expert knowledge and skills, but at the same time risks deskilling an entire cohort when it comes to basic critical skills like reasoning and effective communication
  • AI wielded well may supercharge services, but those same super powers are going to be in the hands of both antisocial end-users (who are willing to eat all of the proverbial cookies in the cookie jar) and outright malicious actors. Things are about to get weird!
  • Monopoly power and financialization are a direct threat to service quality and access, and the decline of service quality and access are by extension threats to rich democratic life. The current political-economic structure of AI seems to be an accelerant to monopolization and financialization, and we should be cautious about embracing for the sake of convenience platforms which will foreclose better futures for services and the people who use them

Specific Applications

  • We’re using these tools both to support software prototyping, and as a medium for prototyping agential flows
    • We’re also finding the back and forth with Claude helpful in driving us to think through our specifications more deeply
  • We’re using these tools to accelerate the early parts of desk research (sketching out broad knowledge frameworks, identifying possible sources, crunching through datasets, etc.) but as we get deeper into a project it’s important that we be able to stand behind 100% of what we’re delivering. That means reading sources ourselves, double checking everything, and not relying on a direct agent->deliverable pipeline We don’t use AI for anything that we write.
    • We feel pretty strongly that writing is a tool for thought as much as for communication, and outsourcing our own thinking seems like a bad idea
    • We also would like folks to read what we write, and if we don’t care enough to do the writing, why should you read it?
  • We aren’t putting sensitive personal information into these tools – including recordings or transcripts from research
    • There’s great AI-enabled tooling to support qualitative research, but there are simply too many examples of agents accidentally exfiltrating data or otherwise acting unexpectedly. It’s just not worth the risk at our scale… but we’re keeping our eyes on this space, and are open to working within a client’s frameworks and tools

Are you striking a similar balancing act around AI? Read about our offering or give us a shout.