Computer Vision is changing.
For years, AI systems learned to recognize objects. Today, they are beginning to understand situations, behaviour, and context.
That shift emerged repeatedly during PY4AI 2026 in Pavia, where researchers, engineers, startups, and AI practitioners gathered to discuss the latest advances in Artificial Intelligence.
Rather than summarizing every presentation, we’ve collected the five ideas that stayed with us—not simply because they represent technical progress, but because they raise new questions about how humans and intelligent systems will increasingly interact.
⸻
1. Machine Perception Is Becoming Temporal
One of the strongest messages we took away from the conference was that computer vision is moving beyond single images.
Modern AI systems increasingly combine information across entire video sequences, allowing them to interpret events rather than isolated frames.
One presentation illustrated this through sports analytics.
Imagine trying to identify a football player when their jersey number is blurred, partially hidden, or visible for only a fraction of a second. Looking at one frame isn’t enough. Instead, AI combines information over time, gradually building confidence until it can correctly identify the player.
What struck us wasn’t the sports application itself—it was the broader implication.
Machine perception is becoming temporal. AI is starting to “remember” instead of simply “looking.”
This opens new possibilities not only for recognition, but also for prediction, behavior analysis, and continuous understanding of dynamic environments.
Why this matters for design
If AI increasingly interprets movement instead of isolated images, designing interactions with machine perception may require thinking beyond single visual appearances. Time itself becomes part of the design space.
⸻
2. AI Is Learning to Understand Context
Another recurring theme was that recognizing objects is no longer enough.
Understanding what is happening requires context.
One presentation explored egocentric vision—AI systems that perceive the world from a first-person perspective.
The speaker shared a simple but powerful example: two people may both be eating, yet one uses a fork while another uses chopsticks.
Humans immediately recognize both as the same activity. For AI, however, cultural differences, prior knowledge, and training data can lead to entirely different interpretations.
The same applies to actions like cutting. Is someone preparing dinner? Building furniture? Working in a factory?
Without context, intention is difficult to infer. For us, this highlighted something fundamental.
Perception has never been only about vision. Humans naturally combine objects, actions, culture, and context. AI is only beginning to do the same.
Why this matters for design
As AI becomes more context-aware, designing for privacy can no longer focus only on what is visible. It must also consider how environments, behaviors, and interactions contribute to machine interpretation.
⸻
3. Segmentation Is Changing How Machines See
Bounding boxes have shaped computer vision for years.
Today, that approach is rapidly evolving.
Several presentations explored segmentation models and visual foundation models such as SAM, capable of identifying complete objects and surfaces rather than simply enclosing them inside rectangles.
This represents more than a technical improvement.
Machines are becoming increasingly capable of understanding how different elements relate to one another within a scene.
Rather than asking “Where is the object?”, AI is beginning to ask “What belongs together?”
Why this matters for design
For Cap_able, this evolution is particularly interesting. As machine perception becomes more precise, materials, textures, garments, and surfaces themselves become increasingly meaningful elements in how AI interprets the physical world.
⸻
4. Specialized AI Isn’t Going Away
Foundation models understandably receive much of today’s attention.
Yet another message consistently emerged throughout the conference:
Bigger isn’t always better.
Many real-world applications—from manufacturing and robotics to medical imaging and industrial inspection—continue to rely on specialized models optimized for specific environments.
These systems are often faster, more efficient, and more reliable than large general-purpose models.
Rather than replacing specialized AI, foundation models are likely to expand the ecosystem.
Different tools will solve different problems.
Why this matters for design
Designing for AI means recognizing that there is no single “machine observer.” Different AI systems perceive the world differently, and physical artifacts may interact with each of them in unique ways.
⸻
5. AI Is Moving Into the Physical World
Perhaps our biggest takeaway from PY4AI wasn’t about a particular model.
It was about where AI is going.
Across multiple talks—from Edge AI to robotics—it became clear that intelligence is increasingly embedded directly into physical environments.
Rather than operating exclusively in cloud services, AI is becoming part of cameras, industrial equipment, vehicles, wearable technologies, and everyday spaces.
Machine perception is no longer something happening behind a screen.
It is becoming part of the environments we move through every day.
Why this matters for design
As AI becomes embedded in physical spaces, the design of those spaces also becomes part of the conversation. Objects, garments, materials, and environments increasingly shape how intelligent systems perceive and interact with people.
⸻
The Bigger Picture
Leaving PY4AI, one idea stayed with us more than any individual presentation.
The future of Computer Vision isn’t simply about improving object detection.
It’s about building systems capable of interpreting time, context, relationships, and the physical world itself.
For Cap_able, this reinforces something we’ve believed from the beginning:
Privacy is not only a software challenge. It’s also a design challenge.
As AI systems become more capable of observing and interpreting our surroundings, the physical world—our clothing, materials, products, and environments—becomes part of the dialogue between humans and intelligent systems.
Designers therefore have an increasingly important role to play, not only in creating objects, but in shaping how those objects are perceived by AI.
⸻
Three Questions We’re Leaving PY4AI With
Rather than leaving the conference with answers, we left with new questions:
- What does privacy look like when AI understands behavior instead of simply recognizing objects?
- How can design intentionally influence the way intelligent systems perceive the physical world?
- As AI becomes part of everyday environments, what responsibility do designers have in shaping that relationship?
These are the questions we’ll continue exploring through our research at Cap_able.
Because the future of AI won’t be shaped by algorithms alone.
It will also be shaped by the way we design the world those algorithms increasingly learn to perceive.
