When Advanced AI Makes Things Worse
Nvidia's DLSS 5 just dropped for its first game this week, and the reviews are... complicated. The technology uses generative AI to literally change characters' faces frame-by-frame during gameplay, creating what Nvidia calls "AI-enhanced" visuals. The problem? Players hate it. Frame rates tank, faces look uncanny, and the feature that's supposed to make games better is making them actively worse.
This isn't just a gaming problem. It's a warning sign for every company deploying AI to customer-facing operations.
The Technology That Nobody Asked For
DLSS 5's face-changing technology is technically impressive. It uses generative AI to analyze and "improve" character faces in real-time, supposedly making them more detailed and realistic. But here's what actually happens: characters' faces shift and morph between frames, creating an unsettling effect that players describe as "deeply weird" and "borderline creepy."
The frame-rate hit is just salt in the wound. Users with expensive 50-series graphics cards are seeing performance drops instead of the improvements Nvidia promised. The technology works exactly as designed from an engineering perspective. It just solves a problem that doesn't exist.
This disconnect between technical capability and actual value is everywhere in AI right now. Companies are asking "can we do this with AI?" instead of "should we do this with AI?"
Customer Service's Uncanny Valley
The same perception problem plaguing DLSS 5 shows up constantly in AI customer service deployments. Companies roll out chatbots that technically work but feel wrong to customers. The bot responds quickly, uses proper grammar, and even retrieves the right information. But something about the interaction feels off.
Maybe it's the overly cheerful tone that doesn't match the customer's frustration. Maybe it's the perfectly formatted responses that scream "I'm a machine." Maybe it's the slight delays at weird moments, or the way it completely misses emotional subtext. The technology functions, but the experience falls into an uncanny valley that makes customers want to spam "AGENT" until they reach a human.
This is why we approach AI deployment by asking the deeper question: what does the customer actually perceive and feel? Not just what does the system technically accomplish.
The Difference Between Working and Working Well
DLSS 5 technically works. The faces do change. The AI does process each frame. But "working" isn't the same as "working well."
The best AI implementations are often invisible. When we deploy an AI agent to handle customer conversations, the goal isn't to showcase impressive AI capabilities. It's to make the customer feel heard, understood, and helped. If they walk away thinking "wow, that was surprisingly smooth," we've succeeded. If they walk away thinking "that was some impressive AI technology," we've probably failed.
Consider how voice AI handles customer calls. A system that responds instantly every time feels robotic. A system that includes natural pauses and vocal variations feels human. The technically "worse" implementation creates a better perception.
The same applies to email and chat. An AI that mirrors customer tone, knows when to be brief versus thorough, and recognizes when to escalate isn't just executing tasks. It's managing perception.
What Customers Actually Want
Nvidia assumed gamers wanted more detailed, AI-enhanced faces. But gamers wanted better frame rates, smoother gameplay, and consistency. The enhancement they didn't ask for came at the cost of things they actually valued.
Customers calling support don't want to be impressed by your AI. They want their problem solved quickly and painlessly. They want to feel like whoever's helping them actually understands what they're asking for. They want consistency—the same quality of service at 2 PM and 2 AM.
These are the metrics that matter:
- Time to resolution, not response sophistication
- Customer satisfaction scores, not technical capabilities
- Escalation appropriateness, not containment rates at all costs
- Natural conversation flow, not keyword matching accuracy
When we build AI agents, we optimize for these outcomes. The underlying technology can be incredibly sophisticated, but from the customer's perspective, it should feel effortless.
Building AI That People Actually Want to Use
The lesson from DLSS 5 isn't "don't use advanced AI." It's "understand what problem you're actually solving." Generative AI for faces could be amazing for certain applications—maybe restoring old films or creating character customization. It's just wrong for real-time gameplay where consistency matters more than enhancement.
The same thinking applies to customer service automation. AI that handles routine inquiries, routes complex issues to specialists, and maintains conversation context across channels—that solves real problems. AI that tries to sound hyper-personal by using the customer's name twelve times per message? That's DLSS 5 for customer service.
Speed matters here. We ship, test with real customers, measure their actual responses, and iterate. Not what we think they should want. What they actually respond to. The AI landscape changes daily, and customer expectations shift just as fast. The only way to stay aligned is through continuous learning and rapid iteration.
The Path Forward
Nvidia will likely improve DLSS 5 over time. They'll tune the models, optimize performance, and maybe find use cases where face enhancement actually adds value. Or they'll learn that this particular feature should be optional at best.
That's the right approach. Build, deploy, measure, learn. Just make sure you're measuring the right things—customer perception and satisfaction, not just technical benchmarks.
For businesses considering AI workforce solutions, the question isn't "what can AI do?" It's "what will make our customers' experience better?" Sometimes that means deploying sophisticated multi-model AI systems. Sometimes it means knowing when the simple solution is actually the right one.
The companies that win with AI won't be the ones with the most advanced technology. They'll be the ones that make that technology invisible—where customers walk away satisfied without ever thinking about AI at all.