AI alignment is the hot topic, but are we ignoring the elephant in the room: societal misalignment? This urgent conversation around AI alignment for human safety might actually be a powerful opportunity to confront a deeper, more fundamental challenge...
There's a lot of buzz and anxiety about AI alignment, asking whether these increasingly intelligent systems will share our values and pursue goals that benefit humanity, rather than lead us down unintended, possibly dangerous paths. Simply put, alignment is about different people, organisations, or systems sharing common goals, values, and expectations, working harmoniously rather than at cross purposes. Misalignment happens when these shared purposes break down, leading to friction, inefficiency, and outcomes no one intended. (BTW - for a good overview of the alignment tests and efforts to catch rogue AI take a look at Matthew Berman - his most recent YT overview is a good primer as at 20 March 2025: here).
My open question for this newsletter is: What if our concerns about AI alignment actually reflect deeper, more fundamental societal alignment challenges? And could we use any of the current AI alignment approaches to help us constructively re-align our societies?
Let me start by asking a fundamental question about whether we are values aligned as societies, institutions, or even as individuals? Well of course we are not - it's easy to see there is widespread misalignment in our own fractured world. The values on display by powerful institutions, ultra-wealthy elites or those with entrenched political interests, or special factions (religious or other) are clearly not aligned to many of us more "ordinary folk". This "human misalignment" problem is of course not new - we have always had this fracture in our societies.
But my contention is that the more recent that social media amplification and deliberate spread of falsehoods through rapid media cycles (mainstream and via "influencers") now exposes these differing values more starkly, more rapidly and more dangerously than previous times.
But so what? you might ask. If it were just different people having different opinions, then that would be fine, but when we have powerful groups optimising for narrow self-interest or self- preservation, (even when it conflicts with broader societal good) then there is potential for wider harm. Consider environmental damage driven by short-term profits, or current cuts to social or research programs in pursuit of fiscal austerity. These outcomes, whether intentional or not, highlight a serious misalignment between powerful entities and the wider community.
What I find myself asking frequently is whether I feel truly aligned with decisions shaped predominantly by special factions, powerful individuals or political parties. I confess that I don't often feel aligned to many of their espoused values. Especially when rules often appear written largely to benefit those already with power or leverage.
If you accept my premise, then the alignment problem isn't purely a technical AI problem. It's rooted deeply in societal structures, power dynamics, and the values underpinning our trust in one another. Are there solutions?
Can we rethink alignment in human societies generally, in ways informed by approaches in aligning AI models? Here are some ideas:
- Transparency and interpretability: In AI, we push for systems to clearly justify their decisions. Let's continue to push that standard to corporations and governments. Greater transparency about decision-making processes is the minimum to maintain trust and accountability. We must continue to shine light in dark crevices!
- Value specification: AI forces us to explicitly define the values we embed in technology. Could we similarly articulate clear, shared values in our institutions—like social equity, environmental sustainability, and democratic participation? Let's be clear on our values - so at least we know up front when we are mis-aligned.
- Robustness testing: Just as AI systems undergo rigorous testing to withstand adversarial scenarios, we could stress-test societal systems against challenges like inequality, corruption, or polarisation. Building resilience in our institutions could ensure they serve us more reliably. Perhaps we could even use AI to help us - by running AI "twins" of our societies we could test alignment issues and see if there are pathways to resolution.
Even the concept of "alignment metrics" can translate to societal use. Indicators such as wealth inequality, social mobility, public trust, or democratic engagement could measure whether we're aligning better or drifting further apart.
Naturally human systems aren't neatly programmable algorithms—our reality is messy, complicated, and riddled with conflicting interests. Achieving perfect alignment might indeed be a utopian ideal. Yet striving towards better alignment is essential and entirely possible. The urgent conversation around AI alignment provides us with a valuable framework to reflect critically on our human structures.
Ultimately, this isn't just about preventing a dystopian AI future—it's about improving our present. By applying principles of transparency, clear value definition, and robust testing to our human institutions, we can better tackle misalignment and move toward a fairer, more just, and more genuinely aligned society.
I'd love to hear your thoughts. Do you see alignment issues in your world? How might we address them together?
Acknowledgement: Drafted by Paul, edited with assistance from ChatGPT 4.5 and Google Gemini 2.0 Flash Thinking Experimental. #AIAlignment #SocietalAlignment #Leadership #Transparency #Accountability #FutureofWork #Ethics
Leave a Comment