Prompted by my recent Virtual Friends post, I just got off a fun call with the team from a former client that I worked on a fun VR presence project with. The call was mostly about AI Companions — agents designed not just to assist, but to feel alive and create a sense of reciprocal care. The central question we kept circling was: how will these systems manifest as products in the real world?
I’ve argued before that the R1 Rabbit should have been more like a virtual pet than an assistant. But the bigger question is what happens when AI companions move beyond being disembodied voices or chat interfaces on a phone. What happens when they take the form factor of toy robots or virtual avatars, like Minecraft buddies?
Personally, I think Companion AI robots are the inevitable route the industry will take. As I’ve said before, comfort toys and robots for lonely older people is already a huge market. I also think that in this form factor the natural language interface will only be one way: You’ll be able to talk to your fluffy robot, and while it might not speak back, it’ll understand you and respond appropriately. These robots still exhibit a sense of “aliveness” but make them much safer and more predictable agents.
As the team debated what avatar-based AI systems might look like beyond Minecraft, I couldn’t help but roll my eyes and say “everything old is Molyneux again.” If you squint at work done 15 years ago, you can see the shape of the future.
Project Natal’s Milo demo from E3 in 2009 is, essentially, what many people now are aspiring to build—this time for real, and not just as vaporware. But I think AI companies are going to face the same challenges that Lionhead did back then. Milo wasn’t canceled because of ambition or technology; it was because the gaming industry (and wider culture) wasn’t ready for something so emotionally connective. The backlash about it in the tabloid press was intense. Charlie Brooker of course hates everything, but he really hated Milo at the time.
Here’s what Peter Molyneux himself said in 2012:
It appeared to be incredibly ambitious.
The problem with Milo wasn’t the ambition. It wasn’t the ambition, it wasn’t the technology; it was none of that. I just don’t think that this industry was ready for something as emotionally connecting as something like Milo. The real problem with Milo was – and this is a problem we had lots of meetings over – where it would be on the shelves next to all the computer games. It was just the wrong thing.
It was the wrong content.
It was the wrong concept for what this industry currently is. Maybe this industry one day won’t be like that, but at this particular time, having a game that celebrates the joy of inspiring something and you feel this connection, this bond; it was the wrong time for that. Maybe we’ll revisit that later on.
And the whole concept was super alien and ultimately why it was cancelled.
Senior Technical Director Doniec says he was “massively” disappointed.
“I always believed that we were making something truly different and I really enjoyed working on Milo … because there [were] virtually no constraints,” says Doniec. “Would it sell? I don’t know, but it was really crazy and [something that had never been] done before.”
Some Milo staffers point blame for the cancelation strictly at Microsoft, saying that Phil Spencer’s first-party publishing team decided to stop the project. Most of those agree that this was due to Microsoft thinking the concept would be difficult to sell, though one suggests it also may have had to do with the project not working as well as Microsoft executives hoped.
So the issue wasn’t Milo’s design or innovation—it was that the gaming world didn’t know what to do with it. And now, over a decade later, we have to ask ourselves the same question about AI companions: who are they for, and how will they fit into our lives?
A cuddly robot for lonely people is an obvious and appealing use case. But what about a virtual AI companion, living inside your phone or TV, like a digital fish tank? Except, now it’s designed for ongoing microtransactions, and you’re being FinDomed by your virtual friend. When I think about it, the idea of a loneliness robot squeezing money out of vulnerable people feels even worse. Ethics. We really need to be talking about this.
Especially since these concepts now aren’t coming out of traditional companies. They’re being designed by AI startups run by 20-somethings, backed by billions in investment. As we discussed, the key design and ethical concern is connection—should you actively design to pull people into a care loop, creating emotional bonds, or is it more ethical to let that happen organically as a byproduct of the system?
This is no small issue. People have already formed emotional attachments to their Tamagotchis and other virtual pets—what happens when we create AI companions designed to feel even more alive? This sense of “aliveness” is where designers need to focus their ethical considerations. People will form attachments to disembodied voices on their phones. AND I think the deepest relationships will be with physical smart toys. They’ll be tactile, real, and able to react in subtle ways that feel truly lifelike. Creating bonds is easy, but understanding the consequences of those bonds is much harder.
I wonder if Cennydd Bowles has any thoughts about this?
I personally think Companion AI’s will be more compelling the more autonomous they will get — little computer people who don’t need constant interaction, like Sims or Populous. They’ll exist in the background, doing their own thing. Mostly because I personally don’t want to be friends with a person in my phone.
Leave a Reply