I agree that having a few walled gardens is a problem, but having more smaller walled gardens probably won't change much. Bad actors can still sign up, post, spam, harass, etc. on several small walled gardens. What we need is a way to better identify and handle bad actors.
The internet in the 90s seemed more fun and more open. Perhaps it's because the only people really participating were not interested in abusing sites or people. It was mostly nerds sharing nerdy things. Once money got involved, and free everything was available, it turned into this soup of bots, trolls, AI, fake this and that, big money, swaying public opinion, and gross content.
Smaller sites and discussion boards have been at a disadvantage recently trying to fend off spam, bots, and sock accounts and very often lose out to the big sites. Effectively controlling abuse and doing it a cost-effective way can be very hard.
RealPerson.io (https://realperson.io) was created as a way for websites to verify that a user is a real person, but without disclosing personal details about the user. Each person can create a single account on RealPerson.io and then can create separate, randomly-generated codes for each site they use. Websites register on RealPerson.io and then for each user signup on their website they simply asks the user for their RealPerson code for their website. A backend REST call to RealPerson.io with the user code for the site returns "yes" or "no." Sites don't share codes so you can't track across websites. No shared logins or authentication code. Bots would have to pay for a code for each account which would be cost prohibitive to run a bot farm.
The first implementation of this approach was done with RealPeople.io (https://realpeople.io) which uses RealPerson.io to verify that the user is real.