Garbage In, Average Out
If AI can reproduce a game quickly, players will judge whether it's worth playing. Other developers might dismiss it as a clone, but that judgment can overlook what the game does well. Every game released has to pass the same test: is it interesting? If a game made entirely with AI is better than what came before it, putting the genie back in the bottle will be impossible. We'll need to embrace that future.But I struggle to see AI generating anything creatively unique. After more than a year of working with AI-produced content, I still see imitation and recognizable slop, even with the best models from OpenAI and Anthropic. I haven't seen them invent a new game mechanic that excites me. AI can speed up production, but I think the creative judgment behind something that delights players will come from humans for a long time.
[...]
You can produce amazing-looking games with AI, and the tools are making that easier. But can AI help you know whether a game is fun? I've tried getting AI to play mobile games through simulators and score them. In those experiments, it really didn't have a clue how to measure fun.
That's why I don't see AI as a silver bullet for creating new hit games. It can help you build faster, but becoming a better designer still requires human judgment. In my experience, AI can assemble the pieces like an engineer without understanding what the game needs to become fun.
People, ideas, machines - in that order!
I am in broad agreement with Joakim, AI is just a sharper tool that introduces a generational shift similar to when Unity catalysed the era of #indiegames.
The only thing I'd add is that nobody knows how to measure fun, it's such a nebulous concept and subjective score card. It's why we used The Denki Difference, it helped us think, make, edit and analyse like a player (feel, drama, life, convenience and twist) and a maker (immediacy, preparation, toys, magic, moments).
Ben Thompson, in conversation with Ari Emanuel about entertainment and AI:
Ben Thompson: The thing is, the AI will give you what you put in. And so if you talk to the AI about an agent-related question, the quality of analysis that you would get out of the AI would probably blow you away, this is really, really good. If I ask the AI an agent question, it's going to be the most Wikipedia-esque, basic thing. I've talked about this, there was a friend — there was some court case that happened, I think it was something App Store related, it was relatively early on. I'm like, "It was unbelievable how good the AI analysis was", because I was going back and forth with it, and he's like, "Oh really? I wasn't impressed at all", and I looked — he had asked the AI, "What do I think about this case?", and it was the most generic answer, it's such a general question. Whereas I already had an angle on it, because I knew a lot about the area. But I also asked a very specific legal question about it, and then it was like very in-depth.Ari Emanuel: That's why I think there's going to be an incredible creative unlock in our business. When Baz Luhrmann is inputting what he wants as a scene, as a shot, etc., and he's giving it exactly what you're saying — what he needs — or Ben Affleck, or Pete Berg, or whatever — the conversation is completely different from you doing it.
Ben Thompson: Right — I have no idea what to ask for.
Ari Emanuel: Yeah, and I just think there is that 5% that I do not think the AI, because it's the greatest average, is going to be able to kind of take the next step. I don't believe it gets past that point. I still think the human condition is needed, and that's where the greatest value comes from.
Ben Thompson: Right. Also, for stuff that is sort of net new, and taste — taste is not a verifiable function.
Ari Emanuel: It's the black box.
You get above average if your judgement can shape the inputs and iterate the outputs.
The machine is the greatest average. The great stuff still comes from people and ideas, in that order.