Post

As someone who's given many talks on practical ethics for software engineers, I've been struggling for the past few months to know what to do. 🧵

I've made it no secret that I don't want to use gen AI tools for my work or personal life, for many of the reasons that @charity.wtf has shared in her thread. But every day, I'm told that if I don't use them at work, I will lose my job.

When I gave my ethics talks in the weird but halcyon days of 2022 and 2023, my goal was to empower engineers to identify when they're being asked to do unethical work, and push back on it -- because *we held the power*.

And now? I feel completely powerless. Pushing back on any AI usage is a fast track to a layoff. I've never before been forced to use a tool that so fundamentally changes the way I do my job. I've been fortunate that usually my work speaks for itself.

But I have no good answers for anyone who wants to avoid these tools for ethical reasons. I can't afford to leave my job, and all tech companies are forcing these tools upon their engineers.

how is this any different from if your company was moving from bare metal to the cloud, and you refused to use the cloud due to many of the same ethical reasons (data centers, clean water, etc)?

This is a really good question. (I have also given a talk on cloud sustainability so this is a topic near and dear to my heart!) I think there are three differences. First, unlike this AI push, I can see much more clearly the benefits of using the cloud. (I work at Heroku so the cloud is my life.)

The tradeoff seems more reasonable. I don't love it, but I see the benefits clearly for both engineers and customers. Second, maybe I came to the industry too late but the AI hype feels like it's on a whole new level. I truly can't stand hype.

AI hype is like cloud hype on steroids. It has all of the cloud issues and more. Third, and this might be a bad reason, but it doesn't affect me emotionally in the same way. Other than working on the cloud platform itself, nobody told me what text editor to use or told me how to write my code.

we issued an AI mandate at honeycomb. i mean to write about this, but i'll give you a preview of the reasons why. -- we are a devtools company. we have a responsibility to stay ahead of our customers -- we see exciting product opportunities with nondeterminism, as well as exciting workflow changes

-- we suggested and encouraged people to make time to learn the tools, and it just was not happening a mandate can be stupid and punitive, but it doesn't have to be. a mandate is a way of putting organizational muscle behind a decision. it's a way of funding a change, getting crisp on goals, and

acknowledging that hey, we're all going to get a little bit slower and worse in the near term. we accept that deadlines may be missed, we accept a higher rate of failures and incidents. we're all going to go through this together. it forces the question, what does success look like?

without a mandate, you're effectively telling people they should learn these skills in their spare time, if they can and if they feel like it. which is like telling them, this is NOT a priority. which is actually a betrayal, if you are only looking to hire people with these skills moving forward.

i agree, i hate being told what to do. i agree, lots of mandates are done poorly and without a clear sense of outcomes. i agree, this is all happening very fast. but i think any employer that isn't supporting people in skilling up right now is a shit employer that does not serve employees well.

I wasn't as opposed to it before the metrics became totally unhinged - rewarding people on longer PRs, more PRs, more tickets completed - all measures that we've agreed, as an industry, are useless.

We've also not really seen much in the way of relaxed deadlines, or valuable comparison between AI and non-AI driven work. Instead, we're told to "estimate" how long work would take without AI and compare it to how long it took with AI.

Working at an enterprise level, where we're rarely greenfielding and our primary job is stability and reliability, it feels absolutely insane to me to push these tools at this speed without doing it any sort of circumspect way and without allowing for any pushback.

I know we've always had an issue calculating "developer productivity," and I wish we focused on better global metrics, like uptime, incident rates, actual revenue... and the moving goalposts of our particular mandate are a mix of impossible and unrealistic.

A technical choice like cloud-vs-on-prem doesn't implicate cognitive biases and psychological dependence and it's possible to do fairly direct ROI calculations. We're not far from corporate mandates for nicotine use (with leaderboards), and even that would have a narrower impact.

i seriously disagree that it's anywhere nearly that clear cut. you might be able to make a case for AI vs cloud computing, but the better comparison is AI vs the internet with its ensuing moral panics over internet addiction, porn addiction, decay of literacy and numeracy, basic research and so on.

Useful, I think, to remember that while we can dispute back and forth e.g. *do these data centers really use too much water?* the fact remains that this technology is misanthropic in its design, intending overtly to replace humans + human cognition—this is its purported ROI, how it’s sold to boards.

Tell me I’m wrong, you won’t be the first. But tech company boards are still out there telling their executives to be ready to replace 40-50% of their workforce in a year or they’ll be considered falling behind. Whether or not that comes to pass, it is what the target customers buying AI believe.

So 💯x yes, @jedbrown.org is right to focus on power consolidation. This is the purpose of the technology; it is THE biggest play to date by the technocrats who have engineered our current state of wealth and power consolidation. It must evaluated considering alignment with their posthumanist agenda.

Consider the COO of Uber stating they have been unable to find ROI on "AI" use, and the broader context of information distortion in which nobody can acknowledge widespread failures. We may see companies fire 40-50% as a last-ditch stock-juicing tactic, but that doesn't mean the work is being done.

Beyond cognitive biases and distortion of judgment (e.g., the internet did not lead to mass sanction of lawyers for Rule 11 violations), this context of power is very different from the internet, in which panics were related to decentralization and threats to consolidated power.

Wikipedia is an instructive example. Schools banned its use and traditional media depicted it as inherently untrustworthy. Its track record was good, but it distributed power away from entrenched and hierarchical control. Those institutions now have "AI" partnerships and Wikipedia has banned its use

I think we're talking past each other. I see the point you are making, and it's a good one, but it is not addressing the conversation I am trying to have, which is that new technologies come with flaws and down sides and inequities that must be addressed. AI is not unique -- or uniquely evil.