Last week in inauspicious AI news …
Apple and Xbox announced price hikes because of AI-caused supply shortages. 400 newspapers filed a lawsuit against Microsoft and OpenAI for widescale copyright infringement. And, rubbing salt in the wound, it was also OpenAI’s turn to get throttled by the U.S. government.
But the award for Worst Week in AI goes to … REI.
Yes, the outdoor gear retailer, renowned for its environmental stewardship, got caught up in a good ol’ fashioned lambasting over an AI-generated photo in one of its Instagram ads.
The ad — which was live for about a week — features a woman and her bicycle, on an outing somewhere serene and wholesome in nature. No technology or environment-killing data centers in sight.
Except the bicycle has two sets of handlebars. One of which is attached to the bike seat. It’s got extra chains, too, plus there’s an AI-ification of the face of the real-life cyclist who posed for the original photo.
The slop is blatant enough that it’s hard to believe no one noticed for a week before taking it down. REI, in turn, has put the blame on Meta (which owns Instagram) for auto-enrolling it in an AI tool that generates personalized variations of images the advertiser supplies. As of this writing, Meta hasn’t provided an official response, and it’s murky whether REI had the opportunity to review the variation before it went live.
But the brand damage is done, and it could have tarnished a vendor, too. Van Rysel, the maker of the bike, had provided the original image and sponsored the entirely human photoshoot.
Autonomous agents didn’t do this
Who is or isn’t to blame for Bikegate is sorta beside the point. Nor is AI itself the bad guy. What happened is a cascading failure of human systems.
Human in the loop
No one’s volunteered a reason for why this didn’t happen with the ad image. It might have been a missed step in REI’s campaign processes. It might have been a tool design that didn’t give any option for a human review. Or … going to the dark side … it might have been a human with a grudge deliberately approving the slop. No matter the reasons, humans were responsible for the decisions that got to this point. No agent autonomously stole the photo from Van Rysel and launched a sloppy campaign with it.
Over-aggressive AI push
It’s easy to believe REI’s explanation that they didn’t know Meta had enrolled them in the tool. Mandatory AI creep is everywhere you look. Not to mention the dark patterns that bury the announcement deep in legal language or make you go through ten different screens to opt out. Seriously, did anyone ask for their advertising images — for which they often pay a lot of money to be exactly on-brand — to regenerate for every viewer? Again, a human decided the aggressive rollout was the right business decision for the platform.
Brand management
REI’s explanation conveniently doesn’t mention why the ad was up for so long, when it was drawing clearly negative sentiment. Was no one monitoring social? Campaign metrics? Did dashboards only register the spike in engagement and assume it was all hunky dory? Yep, humans either weren’t paying attention or didn’t have the right systems in place to catch it.
You can’t control everything
Van Rysel got caught in the middle of someone else’s mess. Plain and simple, when you give your IP to someone for any reason, legitimate or otherwise, you don’t have 100% control over what happens to it. Once again, humans are ultimately making those decisions, even if the decision is to cede it to a robot.
Same playbook, new tools
The knee-jerk reaction for an embarrassing situation like this is to start adding human gatekeepers with micromanagement precision, which will then slow all of your systems to an unacceptable crawl. That was true before AI, too, and AI has magnified the problem as usual by throwing even more volume and speed into the mix.
So, first, we all have to take a deep breath and understand mistakes will happen. Unflattering things will get released. Even the best gates will never be able to account for all the edge cases. Plus, even if you and your brand are staunchly anti-AI for creative work, you can’t control derivatives you might not like, and fair-use vs. copyright-infringement delineations remain murky.
The playbook for mitigating the risk is exactly the same pre- and post- AI. What’s different now is that we can use AI to our advantage for rote monitoring, and catch issues more quickly. (Notice I did not say using AI for creative development.)
Here are some ideas that could have averted or at least minimized Bikegate:
- Set up an agentic scan for announcements from tools and vendors you work with, looking for information abut new features, privacy policy updates, and similar news. Have the scan alert you when an action might be required to adjust a setting or seek deeper detail about impact. In some cases, an agent might be able to alert you when they notice new or changed options in tool settings.
- Measure sentiment in addition to engagement, if you’re not already doing this. There are off-the-shelf SaaS tools, or you can take the DIY route by setting up an automated scan of your customer channels, doing sentiment analysis via AI, and getting an alert when the numbers go off. For example, tanking sentiment plus high engagement likely indicates a problem’s gone viral and needs to addressed quickly. Conversely, a spike in positive sentiment plus stable engagement indicates the message really hit home with a key segment of your existing audience. That’s food for future campaigns.
- Human in the loop doesn’t work without the human part. Someone needs to follow up on the alerts the agents flag. That includes the custom alerts you’ve set up, as well as the pre-programmed ones coming in from the tools you use. Once you have a better idea of noise vs. signal, you can plug those guidelines into filters or agent instructions.
- Get your crisis communications toolkit ready ahead of time so you can respond quickly when needed. It’s not fair to make your social media team (or army of one) manage this on the fly, and ignoring critics — the “I won’t dignify that with a remark” strategy — might not be the right choice if it looks like you’re trying to hide something. Probably even more so if authenticity is core to your brand.
- Document and publicize exactly where you do and don’t use AI if responsible use is also core to your brand. For example, you might have a rule that AI is never used for creative writing, but helps automate content distribution.
AI is a tool, and tools are only as good as the processes behind them. Your human systems need to be in order, and they need to be flexible enough to properly steer your agents and account for the unique situations AI enables, including the ones we don’t yet know about.
Start with the basics, and go from there.
All opinions here are my own. All text is my own, too, including the em dashes. I welcome constructive comments and discussion on LinkedIn and Bluesky.
Leave a Reply