My entire family is obsessed with this letterpress print by Liana Finck for Jewish Currents.

If you spend any time in the same digital and IRL spaces as I do (slacks, conferences, instagram feed, LinkedIn, NYT, and so on), you might find discussions similar to the cartoon above, but instead of “Jews Jews Jews” (hehe) it would look like “AI, AI, AI!” I haven’t said anything here on the topic because I wanted to take time to learn. As it turns out, learning about AI has proven to be harder and more circuitous – but also more empowering and energizing – than I expected. A lot of people have a lot to say… but most of it boils down to fluff and conjecture. I wanted to get down to the guts. In this blog post, I’m going to take you through some of my learning journey and share my DIY syllabus. I’m not done learning (I hope you’re not, too) and I welcome comments about where your learning has taken you!

Phase 0 – avoidance

I was overwhelmed, stuck, a bit paralyzed even. I decided I was okay with being “a late adopter” of this AI stuff (if at all). I was really persuaded by the labor, environmental, ethical, privacy, artistic, creative concerns about AI. I did the bare minimum that I needed to keep up my professional responsibilities. I dabbled in Microsoft Copilot (because that’s what we had at work) but it didn’t work very well and I didn’t have the energy or enthusiasm to do more.

Phase 1 – getting activated

In March, 2026, Anthropic (the company that makes Claude) published a research paper about the amount of a job that AI can do. This sparked my interest because I’ve always paid attention to the impact of automation in any project (spoiler alert, my experience shows that automation frequently ends up leading to more work, not less, but the work is better and more impactful). When this paper came out, something in me clicked. I could learn ABOUT AI without necessarily becoming a “super user.” I could put my research skills to the task. I started to feel the itch of curiosity. Also, if the claims in this research paper were persuasive, then I was going to approach future AI projects differently.

I conducted a literature review, facilitated by access to journal articles primarily through Google Scholar and arXiv.org. Here are some of my top learnings:

• This paper from Harvard Business School asks whether AI complements or displaces workers by querying ~77k job listings. “These results suggest that while generative AI may decrease demand for highly automatable jobs, it simultaneously increases demand for jobs that can be augmented by this technology.” 
• This paper from PNAS surveyed 100,000 workers in Denmark about their attitude toward and use of ChatGPT and analyzed 18,000 respondents. “Workers in the exposed occupations see a substantial productivity potential in ChatGPT, estimating it can halve working times in about a third of their job tasks.” “Finally, we document that workers’ perceived time savings from ChatGPT are only a weak predictor of their use of the tool. For example, among workers who believe ChatGPT can halve the time to complete a task, only about 23% plan to use it within the next 2 wk. Workers report employer restrictions on use and needing training as the primary barriers to adoption.” 
• This is the paper from Anthropic that made a big splash by comparing “theoretical AI coverage” and “observed” AI coverage based on the company’s proprietary metric. They found a big gap between the theoretical ability to use AI for certain jobs and the actual current use of AI for certain jobs. The extent of this difference varies by industry. But what does “theoretical ability” even mean? “Our findings reveal that around 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of LLMs, while approximately 19% of workers may see at least 50% of their tasks impacted.” IMO, we should be somewhat skeptical when considering this paper; it is reasonable to assume that the researchers are incentivized to show outcomes that are in the best interest of the company. Therefore, it is not surprising that the company’s research is extremely optimistic about the capacity for AI to replace or significantly change/impact jobs. Also, their definition of “jobs” and “job tasks” was from a very clunky federal database (I looked at it with my own eyes) and I was like, OMG this does not seem like a very accurate characterization of many jobs in the working world!
• This paper analyzes 4M tasks in Claude. “AI usage primarily concentrates in software development and writing tasks, which together account for nearly half of all total usage. However, usage of AI extends more broadly across the economy, with ∼ 36% of occupations using AI for at least a quarter of their associated tasks. We also analyze how AI is being used for tasks, finding 57% of usage suggests augmentation of human capabilities (e.g., learning or iterating on an output) while 43% suggests automation (e.g., fulfilling a request with minimal human involvement).” 

I used these papers to make predictions about how much I thought AI could to transform the work tasks that we do at my job. Overall, I thought that AI could help some people work differently or faster, but that the core things that people need to do on a daily basis were going to stay relatively similar. If anything, AI could potentially enable people to do things they couldn’t do before, which paradoxically would require more work and not less work. (And all that is only if we could get past the ethical and technological barriers to actually leveraging this stuff).

Phase 2 – hitting the books

While I was reading scholarly papers about macroeconomics and labor and all that, I also wanted to read about AI in the popular press, particularly with respect to the serious ethical drawbacks of AI. This trail led me to the following books, each of which prepared me to have conversations about how AI could or should or shouldn’t be used.

• Co-Intelligence: Living and Working with AI by Dr. Ethan Mollick
• Unmasking AI: My Mission to Protect What is Human in a World of Machines by Dr. Joy Buolamwini
• Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI by Karen Hao
• Race After Technology: Abolitionist Tools for the New Jim Code by Dr. Ruha Benjamin

I haven’t found one book yet that I feel encompasses all of the themes that I want to braid together, so I can’t recommend just one! However, perhaps a good place to start in longer form media is this excellent interview where Naomi Klein interviews Karen Hao (it was so good, I watched it twice – which is a theme you’ll see a few times in this post).

Phase 3 – technical learning intensifies

The more conversations I was having about AI, particularly the ethics of AI, the more my technical brain felt dissatisfied with the zeitgeist. On Instagram, I was seeing wildly different figures about the environmental impact of a single AI “chat.” On LinkedIn, I was seeing so many “get rich quick” type of posts about how to use AI, but no thoughtful commentary about how AI actually works. When I thought about introducing AI tools at work, I felt that I needed to really understand what is happening “under the hood.” The question that guided my next phase of learning was,

What is “AI” “doing” when it “produces” an output?

Trying to answer this question has taken me on a bit of a wild goose chase. For anyone who wants to learn along with me, here’s where I’ve found materials and explanations that “stuck” with me.

This playlist by Grant Sanderson was so good, I watched all of the videos twice! Buckle up for in depth explanations of linear algebra, calculus, computational linguistics, high-dimension word embeddings, and so on! Most of these concepts were new to me, and definitely not in my comfort zone.

These videos were deeply compelling and challenging and in many ways beautiful. In some places, the narrator would say things like, “well there’s a lot of different ways this could work, we’ll just discuss the most popular/most simple.” These breadcrumbs made me realize… even though I didn’t understand 100% of the topics, I was still hungry for more!

That inquiry led me to study a Deep Learning course offered at MIT and made available on OpenCourseWare from Fall 2024. This is a verrrrrrry abstract/theoretical course and some lectures went above my head. At the same time, I felt that learning my way through this material helped me fill in some gaps from the 3Blue1Brown videos (which I look forward to watching a third time). As of writing this, I’m most of the way through 20 lecture series.

With all of this abstract math and theoretical underpinings, I felt myself longing for more practical examples that still had a technical orientation. The best option that I found so far was this video, which really helped me put some of the building blocks together. It was long/complicated, filmed in a command line interface (code, not “point and click”) and I had to watch it in multiple installments over a period of about 2 weeks. BTW, thanks to this creator for coining the word “vibe coding!” (I don’t plan on ever coding my own GPT, but seeing someone else do it was illuminating).

Learning the math behind AI, even in my extremely non-expert way, has helped me slowly and deliberately come to my own conclusions about the potential and the pitfalls of AI.

Phase 4 – professional applications

To be frank, being able to sort of explain what “back propagation” is in my own words, or what the difference is between GPT3 and GPT 4, or the history of Anthropic splitting off from OpenAI… doesn’t equip me with a lot of day-to-day guidance about what to do with AI or how to do it. I’ve needed to lean on my professional networks and my own trial and error to make headway.

My job supported me to complete the AI for Foundations Professional Certificate from NTEN and TAG. The coursework and final project for this oriented toward real world use cases and covered topics like change management and incorporating AI into an existing technology roadmap. These themes didn’t come up in my other learning and I’m glad I had the chance to explore them.

big takeaway(s)?

It’s hard to come up with one big conclusion from all of this learning (if I had to guess, hundreds of hours of learning!!), but I’m going to try. It’s clear to me that the big AI and tech companies have a vested interest in driving home a few very sneaky ideas:

• AI automating aspects of daily life is inevitable
• AI chat-style tools working the way they do is predestined
• AI *poof* started with ChatGPT and has no antecedents or history
• AI is “magic” and normal people can’t possibly understand how it works

Here’s my reframe:

• We get to decide how and if we use AI
• The way chat-style tools work is the result of very intentional engineering. While consumers like us don’t CURRENTLY have much ability to shape the technology and influence what a “good” response looks like, there’s no reason why we couldn’t in the future, and lots of research labs already do this
• The history of AI comes from the conditions that led to the Turing Test, and machine learning, and deep learning, and linguistics, and engineering. Advances in coding and in infrastructure directly led to the capabilities of GPT3… it didn’t come from “nowhere”
• The math behind AI can be accessible to a nontechnical audience, in fact, it’s fun to learn about!

Moreover (on that last point) avoiding the math is making us more susceptible to all of the groupthink and clickbait and doomy/gloomy predictions. Through my experience with this DIY curriculum, I learned about:

• “Supervised fine tuning” and “reinforcement learning,” which require meticulously labeled data inputs. We should be just as concerned about human rights violations targeting Data Labelers and Content Moderators as we are about environmental impacts emanating from society’s insatiable demand for Data Centers. (Read this report)
• AI as a category of technology with a lot of different applications. This exposed how diabolical and limiting the agenda from the big tech companies is… and also showed me that there can be less harmful types of AI that are nowhere near as extractive as the popular “chat tools” – so much worth exploring here! Sidenote – did you know that when I was naming The Data Are Alright, I almost chose the name “Little Data in the City?”
• Studying “computer vision” and image classification is leading to advances in how to stop or confuse the surveillance state (adversarial pattern stickers! and here’s the math!)
• iNaturalist is an app that crowdsources plant and animal classification and has one of the most comprehensive repositories of data to use for ecological research (link to some cool research) – this also makes me want to try iNaturalist with our kids!
• Using AI models to remove the distortion of water from underwater photography (link) which is both incredibly cool and also important for being able to monitor change to endangered aquatic and ecological environments

I guess another way of saying this is, what I’ve learned about AI is both a lot more horrifying AND more encouraging than what I see in the conversations happening around me. There’s so much more to it than delegating email writing or automating your grocery list.

Moving from a posture of avoidance to active engagement has helped me feed my natural curiosity and exit the vicious hype/outrage cycle.

Perhaps most of all, it’s taught me that the only people who are “experts” in this are people who will never say they are. There’s so much to learn, it’s truly humbling, dizzying, overwhelming. Given all of that, I’d so much rather be in it together.

Learning new technology is daunting; good thing we’ve had a lot of practice:

• i’m learning CiviCRM!
• I’m learning text banking!
• i’m learning Mastodon!
• I’m learning FlossCross!