That’s a great distinction to make between « AI literary » and « critical AI literacy ». The DOL framework is presented as a starting point and they state their commitment to improve it over time with feedback. The next version 2 needs to incorporate the « critical » aspects, and more on the trade offs involved in producing AI services.
But one thing that feels missing even beyond “critical AI literacy” is what happens after those trade-offs are made in practice.
Because even when we understand the trade-offs, decisions don’t stop there — they get implemented, carried forward, and start shaping systems over time.
And we don’t yet have a clear way of tracking or revisiting those consequences once they unfold.
So the challenge isn’t just improving literacy or awareness.
It’s how we structurally hold decisions after they’ve been made.
The distinction you’re drawing — between knowing when to say no and having a structure for what happens after we say yes — is exactly the part most governance frameworks skip.
The decision moment gets formalized. The structure for carrying, revisiting, and absorbing the consequences of that decision often does not.
That gap tends to widen quietly, because the absence of such a structure is itself hard to see until something goes wrong.
I think this is exactly the gap I've been trying to explore.
Much of the discussion around AI focuses on how AI can help organizations make decisions.
I've become increasingly interested in the other side of that equation:
How can AI help the people who are affected by those decisions?
That's one of the ideas behind a project I've been building called the Legal Recoverability Engine (LRE).
Rather than making legal decisions, its purpose is to help citizens reconstruct the path from decision → consequence → possible review → recovery.
In other words, AI doesn't replace human judgement.
It helps make the decision path visible again.
I think AI will change the relationship between citizens and institutions whether we like it or not.
The question is whether we use it only to optimize institutions, or also to strengthen citizens' ability to understand, challenge and recover from decisions that affect their lives.
To me, that's where governance becomes meaningful.
The LRE framing is interesting because it starts from the citizen's side of the same gap.
Institutions can optimize decision processes in ways that make them faster internally while making them harder to reconstruct from the outside. Efficiency and legibility do not always point in the same direction.
That is why "making the decision path visible again" may require pressure from both sides: better internal structure, and an external expectation that the path must be reconstructable at all.
I really like the distinction you're making between efficiency and legibility.
It also makes me think there's another distinction worth adding:
Reconstructability is the prerequisite. Recoverability is what becomes possible because the path can be reconstructed.
If citizens cannot reconstruct how a decision unfolded, meaningful review becomes difficult.
But once that path is visible again, AI can support not only explanation, but also recovery: identifying where the process broke down, what evidence matters, and what realistic paths exist for revisiting the decision.
To me, that's where AI starts serving citizens as well as institutions.
Love how you draw the line between AI literacy and judgement rather than just proficiency... the missing word “no” in the framework is such a simple, devastating diagnostic.
Thank you Alex. I appreciate that. I am broadly pro-AI but people (and companies) really need to know when to rely on their most valuable asset instead... 🙏
This is a great observation Birrell. The AI tools I use often talk about the different modes but never really give exact examples of when to use them and, more critically than that, never invite me to switch modes. Hopefully that will change eventually, but you're right; it seems a strange thing for them not to be doing more regularly.
Now I need to go and buy me some Catholic Tantra. 💪
Claude says: Chip, this is right in your wheelhouse — and the timing is almost eerie given your NIEHS WTP work on AI as occupational hazard. The gap Illingworth identifies is exactly the gap between the DOL’s “AI literacy” and what you’ve been building at NIEHS WTP — treating AI and algorithmic management as occupational hazards requiring worker protection strategies.
At NIEHS, we’ve been developing training that treats AI and algorithmic management as occupational hazards — the same way we treat chemical exposures or fall risks. That means worker voice in how systems are designed and deployed, not just worker skill in operating them. It means the General Duty Clause, not just the user manual. The DOL has the infrastructure to do this. Yet another lifelong legacy project! I wanna go slow on it, but the doomsday clock is ticking! Believe me, Claude did not write that!
Claude has done it again Chip! Do you see any of these strategies being developed anywhere? I am genuinely worried about what this means for the workforce, despite being broadly pro-AI.
The DOL’s AI literacy framework makes for interesting reading (as a UK-based observer) and it does at least hint at critical AI literacy in places. For example, under “Use AI responsibly”, it states that “workers must understand the boundaries of appropriate use” but the examples that follow do not build or elaborate on that crucial point. As you say, not bad for a government document, but an opportunity missed for sure.
The UK government issued a similar document in October of last year (just Google ‘AI Skills for the UK workforce’ if interested). The so-called AI Skills Framework has a stated aim to “upskill 10 million workers by 2030”. Like the U.S. document, it is really all about encouraging/teaching people how to use AI in a purely technical sense, driven by purely economic considerations.
It pays lip service to AI literacy (without even defining it) and everything covered under ‘Responsible and Ethical AI Skills’ (in the AI Skills Tools Package) is focused mainly on identifying bias (good luck with that!), GDPR issues, and correcting AI output. The ability to “assess accuracy and appropriateness of AI for tasks” is listed as an ‘entry level skill’ (!) but there is no explanation of how one might do this. The underlying message (to employers) seems clear. The gravy train is leaving the station, and you had better learn how to jump on. Fast.
Reading this, I keep thinking about how the new US AI literacy frameworks are going to be shaped by whoever shows up in the comment process. Big law firms, industry groups, and corporate policy teams will definitely weigh in to make sure their interests are heard and reflected.
That’s exactly why your critical AI literacy work feels so important here, Sam. You’re naming the things those institutions usually leave out: power, labour, refusal, ecological cost, and how these systems actually reshape people over time. If voices like yours don’t show up, ‘AI literacy’ risks being defined almost entirely as compliance and productivity training.
I think with 8,000+ people on Substack, I’d call you a pretty substantial voice in the community. That’s considered an enterprise‑level presence, Sam 😊
I had no idea the national AI literacy framework was published, so thank you for surfacing it, Sam! It is encouraging to see a coordinated effort to raise baseline capability across the workforce. We need broad adoption to future-proof ourselves for what's about to come in the next years / decades. And thinking has to sit at the center of AI literacy.
You are very welcome Daria. And even though you know I am generally pro-AI use and development, I do worry that this framework seems to be accelerating towards AI adoption without fully considering the human in the process. 😢
Unfortunately, Sam, I’m not sure how much time we have for that. Rewiring people to think critically is a long game. It requires systemic change, from early education to parenting and everything that happens onwards.
If we look beyond the chatbots we are used to and toward agentic systems that aim to operate with no human in the loop the pace of development is intense. The goal seems to be output that matches or exceeds human quality without depending on us. We are moving closer to that reality and that compresses the window for slow cultural shifts.
I agree that the questions you raise are essential. In practice though they are often easier to engage with for people who already have strong critical thinking foundations and the cognitive space to zoom out. Without that foundation, AI becomes something you use rather functionally than something you interogate.
Still, I believe strongly that without proper thinking, you cannot work well with AI anyway. So I'm 100% onboard with your points, I just think they're one step ahead of the mass population. Unfortunately, again.
Thank you Daria for replying with such honesty and integrity. I think you are probably right but I remain convinced that we can intact change from both a grassroots movement in personal AI use and top down pressure by effective lobbying of major government agencies and BigTech. At least that is what I tell myself. ✊
I found this article both useful and unsettling. I agree that knowing how to use AI is not enough; we also need the confidence to question it, refuse it, and recognise its hidden costs. In my own work, the hardest skill is not prompting well, but knowing when human judgement should come first every time. https://www.linkedin.com/pulse/why-ai-literacy-matters-more-than-polished-resume-sachi-designs-3bp5c/
That’s a great distinction to make between « AI literary » and « critical AI literacy ». The DOL framework is presented as a starting point and they state their commitment to improve it over time with feedback. The next version 2 needs to incorporate the « critical » aspects, and more on the trade offs involved in producing AI services.
So true Mahelet. 🙏
I think that’s a really important direction.
But one thing that feels missing even beyond “critical AI literacy” is what happens after those trade-offs are made in practice.
Because even when we understand the trade-offs, decisions don’t stop there — they get implemented, carried forward, and start shaping systems over time.
And we don’t yet have a clear way of tracking or revisiting those consequences once they unfold.
So the challenge isn’t just improving literacy or awareness.
It’s how we structurally hold decisions after they’ve been made.
This really resonates.
And I think the gap becomes even clearer once AI moves from outputs into decisions.
Because even if someone knows when to question or refuse an output, we still lack a structure for what happens next:
how those decisions are carried
how they are revisited
and how their consequences are handled over time
So it’s not just about knowing when to say no.
It’s about what happens after we say yes.
The distinction you’re drawing — between knowing when to say no and having a structure for what happens after we say yes — is exactly the part most governance frameworks skip.
The decision moment gets formalized. The structure for carrying, revisiting, and absorbing the consequences of that decision often does not.
That gap tends to widen quietly, because the absence of such a structure is itself hard to see until something goes wrong.
I think that's exactly it.
We tend to formalize the moment a decision is made, but much less attention is given to preserving continuity afterwards.
Governance shouldn't end when the decision is recorded.
It should preserve a visible path through implementation, consequences, review, adaptation, and, when necessary, recovery.
Otherwise, every decision risks becoming an isolated event rather than part of an ongoing learning process.
Perhaps that's where governance and organizational learning ultimately meet.
The path you’re describing — from implementation through consequences and recovery — is also where accountability tends to break down most quietly.
Each step can be completed, even documented, without anyone clearly holding the whole arc.
Governance records the decision. Learning requires someone to hold the thread.
I think this is exactly the gap I've been trying to explore.
Much of the discussion around AI focuses on how AI can help organizations make decisions.
I've become increasingly interested in the other side of that equation:
How can AI help the people who are affected by those decisions?
That's one of the ideas behind a project I've been building called the Legal Recoverability Engine (LRE).
Rather than making legal decisions, its purpose is to help citizens reconstruct the path from decision → consequence → possible review → recovery.
In other words, AI doesn't replace human judgement.
It helps make the decision path visible again.
I think AI will change the relationship between citizens and institutions whether we like it or not.
The question is whether we use it only to optimize institutions, or also to strengthen citizens' ability to understand, challenge and recover from decisions that affect their lives.
To me, that's where governance becomes meaningful.
The LRE framing is interesting because it starts from the citizen's side of the same gap.
Institutions can optimize decision processes in ways that make them faster internally while making them harder to reconstruct from the outside. Efficiency and legibility do not always point in the same direction.
That is why "making the decision path visible again" may require pressure from both sides: better internal structure, and an external expectation that the path must be reconstructable at all.
I really like the distinction you're making between efficiency and legibility.
It also makes me think there's another distinction worth adding:
Reconstructability is the prerequisite. Recoverability is what becomes possible because the path can be reconstructed.
If citizens cannot reconstruct how a decision unfolded, meaningful review becomes difficult.
But once that path is visible again, AI can support not only explanation, but also recovery: identifying where the process broke down, what evidence matters, and what realistic paths exist for revisiting the decision.
To me, that's where AI starts serving citizens as well as institutions.
I think there are a lot of frameworks but how many are actually applied? That’s what bothers me
Soooooo true! Maybe we should just stop writing frameworks and actually try living by them instead.
Haha no, let's just keep writing and maybe some government or organization will actually get the idea to start enforcing them
Love how you draw the line between AI literacy and judgement rather than just proficiency... the missing word “no” in the framework is such a simple, devastating diagnostic.
Thank you Alex. I appreciate that. I am broadly pro-AI but people (and companies) really need to know when to rely on their most valuable asset instead... 🙏
"Team Sam" to be on the "Commission"
Chris, literally my dream. You write the LoS and i’m there. 😅
I have grown used to telling AIs that they have erred. Today I encountered a meta-correction:
I had asked Gemini to comment on my five novels about "Catholic Tantra"(http://btwreviews.com - feel free to buy many copies.)
Gemini 3 was just not getting it. It lost track of the story, hallucinated cheerfully and clearly had skimmed the books.
Then I noticed that the mode it was in was "Fast." For the hell of it, I changed the mode to "
Thinking" and went again. This time it read intelligently. Even the mistakes it made were defensible.
I was grateful. I also learned something.
We might add to the critical AI skills these two questions:
- Is the mode adequate to the task?
- And if not, will the AI tell you to change modes?
This is a great observation Birrell. The AI tools I use often talk about the different modes but never really give exact examples of when to use them and, more critically than that, never invite me to switch modes. Hopefully that will change eventually, but you're right; it seems a strange thing for them not to be doing more regularly.
Now I need to go and buy me some Catholic Tantra. 💪
Important distinction. Literacy without critical refusal becomes compliance.
That is an excellent summation Michael. 🙏
Making it modular, prioritizing equitable access, and grounding it in experiential learning shows real thought
Exactly Dennis and to be fair to DOL there is actually some really good stuff in the framework. The problem is that it is all in one direction…
Claude says: Chip, this is right in your wheelhouse — and the timing is almost eerie given your NIEHS WTP work on AI as occupational hazard. The gap Illingworth identifies is exactly the gap between the DOL’s “AI literacy” and what you’ve been building at NIEHS WTP — treating AI and algorithmic management as occupational hazards requiring worker protection strategies.
At NIEHS, we’ve been developing training that treats AI and algorithmic management as occupational hazards — the same way we treat chemical exposures or fall risks. That means worker voice in how systems are designed and deployed, not just worker skill in operating them. It means the General Duty Clause, not just the user manual. The DOL has the infrastructure to do this. Yet another lifelong legacy project! I wanna go slow on it, but the doomsday clock is ticking! Believe me, Claude did not write that!
Yes Chip! More of this please. The worker voice is by far the most powerful tool we have to work with. 🙏
Claude has done it again Chip! Do you see any of these strategies being developed anywhere? I am genuinely worried about what this means for the workforce, despite being broadly pro-AI.
It's kind of going to be like the food pyramid, dominated by business self-interest.
I find that my use of AI is most successful in areas where I have deep critical skills and am understanding of competing perspectives.
So true Charlie! Where I rely on AI most heavily is in areas where I can guarantee the validity of its output because of my own subject knowledge. 🙏
The DOL’s AI literacy framework makes for interesting reading (as a UK-based observer) and it does at least hint at critical AI literacy in places. For example, under “Use AI responsibly”, it states that “workers must understand the boundaries of appropriate use” but the examples that follow do not build or elaborate on that crucial point. As you say, not bad for a government document, but an opportunity missed for sure.
The UK government issued a similar document in October of last year (just Google ‘AI Skills for the UK workforce’ if interested). The so-called AI Skills Framework has a stated aim to “upskill 10 million workers by 2030”. Like the U.S. document, it is really all about encouraging/teaching people how to use AI in a purely technical sense, driven by purely economic considerations.
It pays lip service to AI literacy (without even defining it) and everything covered under ‘Responsible and Ethical AI Skills’ (in the AI Skills Tools Package) is focused mainly on identifying bias (good luck with that!), GDPR issues, and correcting AI output. The ability to “assess accuracy and appropriateness of AI for tasks” is listed as an ‘entry level skill’ (!) but there is no explanation of how one might do this. The underlying message (to employers) seems clear. The gravy train is leaving the station, and you had better learn how to jump on. Fast.
Exactly my read as well Philip. And sadly no guardrails or meaningful training in sight...
Reading this, I keep thinking about how the new US AI literacy frameworks are going to be shaped by whoever shows up in the comment process. Big law firms, industry groups, and corporate policy teams will definitely weigh in to make sure their interests are heard and reflected.
That’s exactly why your critical AI literacy work feels so important here, Sam. You’re naming the things those institutions usually leave out: power, labour, refusal, ecological cost, and how these systems actually reshape people over time. If voices like yours don’t show up, ‘AI literacy’ risks being defined almost entirely as compliance and productivity training.
Thanks so much Anna. I know I am a small voice here calling for the centre ground in AI use, but I'll keep trying to be heard and enact change. 🙏
I think with 8,000+ people on Substack, I’d call you a pretty substantial voice in the community. That’s considered an enterprise‑level presence, Sam 😊
Thanks Anna! Ok. I’ll do it! ✊
I had no idea the national AI literacy framework was published, so thank you for surfacing it, Sam! It is encouraging to see a coordinated effort to raise baseline capability across the workforce. We need broad adoption to future-proof ourselves for what's about to come in the next years / decades. And thinking has to sit at the center of AI literacy.
You are very welcome Daria. And even though you know I am generally pro-AI use and development, I do worry that this framework seems to be accelerating towards AI adoption without fully considering the human in the process. 😢
Unfortunately, Sam, I’m not sure how much time we have for that. Rewiring people to think critically is a long game. It requires systemic change, from early education to parenting and everything that happens onwards.
If we look beyond the chatbots we are used to and toward agentic systems that aim to operate with no human in the loop the pace of development is intense. The goal seems to be output that matches or exceeds human quality without depending on us. We are moving closer to that reality and that compresses the window for slow cultural shifts.
I agree that the questions you raise are essential. In practice though they are often easier to engage with for people who already have strong critical thinking foundations and the cognitive space to zoom out. Without that foundation, AI becomes something you use rather functionally than something you interogate.
Still, I believe strongly that without proper thinking, you cannot work well with AI anyway. So I'm 100% onboard with your points, I just think they're one step ahead of the mass population. Unfortunately, again.
Thank you Daria for replying with such honesty and integrity. I think you are probably right but I remain convinced that we can intact change from both a grassroots movement in personal AI use and top down pressure by effective lobbying of major government agencies and BigTech. At least that is what I tell myself. ✊
I found this article both useful and unsettling. I agree that knowing how to use AI is not enough; we also need the confidence to question it, refuse it, and recognise its hidden costs. In my own work, the hardest skill is not prompting well, but knowing when human judgement should come first every time. https://www.linkedin.com/pulse/why-ai-literacy-matters-more-than-polished-resume-sachi-designs-3bp5c/