When Wikipedia first launched, teachers immediately lambasted the site, discounting its credibility as a source and sending students off to conduct “real” research in libraries full of outdated books. It was so long ago that you could still handwrite essays, double-spaced on notebook paper (ask me how I know). Fast-forward 25 years, and Wikipedia’s accuracy is still up for debate, though most of what I found puts its rate north of 80 percent. There’s some evidence that it’s as high as 98 percent, comparable to traditional encyclopedias. If I had to guess, I would put it somewhere in the middle of those two figures.
The Wikipedia model involves a strict citation policy and editing process that’s not to be taken lightly. I cannot tell you how many clients I’ve had who wanted their own pages, even hiring agencies that claimed to have magic abilities, only to be shot down in the end when the site deemed them not relevant enough for a dedicated entry. Ouch. Either way, the Wikipedia brand has lost more of its luster in recent years as shiny AI tools like Claude, Gemini, or, cringe, “Chatty,” have entered the market.
Of course, AI has plenty of use cases, but when it comes to conducting research, I always think back to my high school teachers and their knee-jerk reaction to Wikipedia. Most of the initial concern stemmed from the fact that anyone could edit the site. Even now, when a famous person dies, Wikipedia is forced to lock their page almost immediately before rumors start to spread, so I get it. Except the thing with Wikipedia is that the way it operates is pretty straightforward. The sources are readily available at the bottom of the page, clearly cited in an easily recognizable format, with LOTS of human involvement alongside technology.
AI, on the other hand, isn’t exactly known for its transparency. I would venture a guess that most of us don’t really understand how it finds the answers to our questions. Fair warning: I am going to simplify this way (way) down and leave a lot out, for the sake of focusing on two particular actions. The first is what’s known as “generation-time citations,” where the AI creates the answer and pulls citations in a single step. Meaning that the AI connects with a source, database, or search tool while writing. Think of this like doing your research from the library or drafting a paper with the book you need right in front of you.
The second is “post-hoc citations,” in which the AI creates its draft and then adds or checks citations separately. So, the AI writes the answer to your question, puts a period on the final sentence, and then finds sources to back it up. Which, in real life, is a completely bananas way of creating content, though there are people who do this, drafting up what they think and then looking for research to support their theories. Ultimately, studies show there are pros and cons to each approach, but that gets into factors of latency and questions about different datasets, and we don’t have time for all of that today.
Here’s where things get really sticky with AI and citations: reputable sites often block AI. We know for sure that this includes news outlets like The Wall Street Journal, the Associated Press, the BBC, and even the New York Post. These sites have embedded block code to prevent LLMs from training on, retrieving, and indexing their content. But it doesn’t stop there. Amazon has taken measures to block competitive AI crawlers. GoodReads. Yelp. eBay. Consumer Reports. You get the picture.
With fewer current sites to … cite, AI has to get creative. A lot of the time, it simply reuses the same sources over and over, which is why you see it regurgitating recycled material. Ever try to get a more interesting take on a question you’ve posed only to find the AI doesn’t really have a better response? In other instances, AI digs deeper into the recesses of the internet and points to outdated pages from like, 2014, which might as well be 1914 at this point. Most publications I deal with won’t accept research older than three years, let alone 10 years, making these results pretty much useless. I suspect educators wouldn’t be too pleased either.
The point is, for the average casual AI user - the same person who relied on search engines and Wikis for research just a few years ago - there’s a lot happening behind the scenes that they might not know or understand, and it’s impacting output and outcomes. And as much as I harp on the importance of media literacy and recognizing what’s underpinning the content we consume, that goes double for AI literacy, especially as more sites move to protect published materials. Without the ability to cast a wider net, seek higher quality sources, and demand better data, the information superhighway will meet a fiery end, encircling itself like an ouroboros, as we prompt, “Another, please.”
Image via Wikipedia (duh).
How do I?
Dear Laura,
Whatever happened to the idea of work-life balance? I thought AI was going to save us all this time, but it feels like I’m working harder than ever, and the expectations are even higher. I don’t get it. Has this concept totally disappeared from HR? What are you hearing?
Thanks for weighing in,
Never Enough Time
Dear Never Enough Time,
I’m hearing HR Pros are more exhausted than ever, so you’re spot-on and not alone. Firstly, doesn’t it feel like the conversations have completely stopped when it comes to work-life fill in the blank (balance, integration, etc.)? So, I’m glad you’re bringing this up.
For years, the HR Community has been working toward a better balance between work overload and a peaceful, joyful life outside of work. Now, it seems like everyone is plugged in 24/7 without coming up for air. This isn’t good. Great point you’re making about AI... if it takes off the administrative, tactical, transactional grunt work -shouldn’t the humans behind the computer be freed up to work more thoughtfully, strategically, proactively? And thereby maybe have some free time to take PTO without logging in, to take a walk without zooming, to go to a gym class without fear of missing a deadline.
Hmm. You’re stumping me here on the why this is happening/what can be done. My first thought: Let’s all look inward. Each professional in the HR community has to stop waiting for permission to take a break and take care of themselves. It’s an inside job. Then, collectively we can keep brainstorming about how to get the conversation started again (thanks for bringing it up here!). Let’s all do our part to practice what we preach and enjoy this one wild and precious life without work completely consuming us.
Your friend,
P.S. Laura recently took over the “Practitioner Corner” podcast on WRKdefined, and her first four episodes are already live! Hear directly from leading HR and TA pros about what’s keeping them up at night in this wacky market.
How work is working.
Not looking great for the vendor lawsuits … The Guardian is the latest to resurface the slate of class-action and other pending claims against companies ranging from HR tech providers to tech giants like Meta. The article doesn’t say anything we haven’t read before, but going back to my earlier point, it doesn’t exactly promote expansive thinking either.
Speaking of, the case against Otter.ai is moving ahead, with a federal judge allowing the plaintiffs to proceed with their theory that the assistant acted as a “third-party eavesdropper” because it allegedly retained meeting communications for commercial purposes, rather than simply serving as a transcription tool. Those bulky, old-school conference room phones aren’t looking so bad anymore.
Especially since you never know what your employer might be doing to surveil you, writes Cathy Bussewitz. One line really jumped out at me: “Some states, including New York, Connecticut, Delaware and Maine, require employers to notify workers if they’re being monitored.” According to Google, there’s only one other missing from that list. So five of 50 states require direct notification. Cool.
ICYMI, from around the space.
Pete Tiliakos dropped “The future of work is colliding with the future of money,” the first in a new series on why HR leaders need to be more mindful of payroll and fintech as part of their overall talent strategies.
Tami Nutt is “Bringing Sexy Back to HR Compliance,” and if anyone is up for that challenge, it’s Tami. She takes the compliance we’re accustomed to and reframes it for today’s workplace, connecting the pieces along the way.
Says Deloitte, via HR Dive, only 1 in 5 organizations are prepared to move toward autonomous AI agents because business processes will need to be redesigned or rebuilt. Insert stock language about digital transformation here.
HOW opportunities.
Fast Company Most Innovative Companies - early deadline September 4
Ragan AI Awards - deadline September 10
The Cloud Awards - deadline October 23
And finally, a break.
Reading: Ruth Reichl’s Substack, La Briffe, is giving me lots of inspiration for the forthcoming colder months, when I actually spend time in the kitchen cooking
Watching: “Lorne” on Peacock. An in-depth look at SNL’s enigmatic leader.
Listening: Paul Simon - Graceland. See above, and you will understand my reasoning for revisiting this album. It plays a role.
Exploring: I am cutting my screentime down with a Brick, so I’ll leave you with the following quote, as it feels timely:
“The ability to be wrong is one of the most important virtues, and it is extremely difficult. It should be emphasized in families and schools alongside standard virtues like honesty, sharing, and kindness, starting as young as possible.”
Until next time.
- Katie