I agree with the overall diagnosis that data lock-in are the default trajectory of AI app/agent platforms from BigTech.
However, I don't think policy proposals 1 and 2, "Open API access for major platforms" and "Memory as a portable service" are realistic.
I don't think open-banking regulation is a good analogy because banking is a very stable domain with a very stable set of basic abstractions that have existed for decades and are not going to change anytime soon: account, balance, transation, etc.
Whereas in AI apps/agents, the technological solutions and the very product abstractions are way too much in flux. Open API access to "chats" in a chatbot UI surface is probably manageable, but what about live conversations? Mobile screen share with AI? Agentic browser experience that takes actions on user's behalf? Desktop client/agent that records desktop screen and takes actions on user's behalf on the desktop? A completely integrated wearable device, like smart watch or glasses? And the combinations of all the above, creating flows of information between the user and the AI platform across devices and modes?
For memory/personalization, "memory MCP servers" are based on a paradigm that might be too limiting for the specific applications. Tomorrow's AI model architectures may favour data-efficient fine-tuning or even continual learning over RAG with distinct "memory pieces" as excerpts of text (providers may still support memory erasure, but only in a form of rewinding to a certain checkpoint in the past or forgetting everything that have happened in a single interaction, rather than browsing the list of memory excerpts and hand-picking them).
And even if RAG-based memory is still viable, it's just not realistic to me that big providers will allow users plugging third-party "memory APIs" inside their infrastructure. Managing such plug-ins is giant hassle from the networking perspective. If, in theory, AI platforms are faced with such regulation that would oblige them to enable users to record all memories in an external service, I bet the platforms would rather turn off memories/personalization in these countries altogether than hassle with these plug-ins.
I agree that data portability regulations and norms do keep enshittification in check. Perhaps this have happened with browsers, where the ease of porting one's browsing history, bookmarks, and extensions between Chromium-based browsers probably have constrain Chrome's enshittification somewhat (although it still continues: Chrome disabled uBlock Origin earlier this year).
However, data portability regulations don't prevent effective oligopoly with 3-4 dominant players. Competition will keep them in check and will slow down enshittification, but this will still be a *monoculture* to a large extent (with cookie-cutter, "one size fits none" apps in the language of Alex Komoroske), and the relative *power* on the internet will still be further concentrated in those handful mega-providers. These are distinct outcomes from enshittification per se, and they are not great.
The meta-concern with such regulations is that huge AI platforms may bite the bullet and implement them, but they may add a lot of burden for new players. Startups will choose not to enter these markets, defaulting to the US alone. Local innovation is also stifled.
I agree with the overall diagnosis that data lock-in are the default trajectory of AI app/agent platforms from BigTech.
However, I don't think policy proposals 1 and 2, "Open API access for major platforms" and "Memory as a portable service" are realistic.
I don't think open-banking regulation is a good analogy because banking is a very stable domain with a very stable set of basic abstractions that have existed for decades and are not going to change anytime soon: account, balance, transation, etc.
Whereas in AI apps/agents, the technological solutions and the very product abstractions are way too much in flux. Open API access to "chats" in a chatbot UI surface is probably manageable, but what about live conversations? Mobile screen share with AI? Agentic browser experience that takes actions on user's behalf? Desktop client/agent that records desktop screen and takes actions on user's behalf on the desktop? A completely integrated wearable device, like smart watch or glasses? And the combinations of all the above, creating flows of information between the user and the AI platform across devices and modes?
For memory/personalization, "memory MCP servers" are based on a paradigm that might be too limiting for the specific applications. Tomorrow's AI model architectures may favour data-efficient fine-tuning or even continual learning over RAG with distinct "memory pieces" as excerpts of text (providers may still support memory erasure, but only in a form of rewinding to a certain checkpoint in the past or forgetting everything that have happened in a single interaction, rather than browsing the list of memory excerpts and hand-picking them).
And even if RAG-based memory is still viable, it's just not realistic to me that big providers will allow users plugging third-party "memory APIs" inside their infrastructure. Managing such plug-ins is giant hassle from the networking perspective. If, in theory, AI platforms are faced with such regulation that would oblige them to enable users to record all memories in an external service, I bet the platforms would rather turn off memories/personalization in these countries altogether than hassle with these plug-ins.
I agree that data portability regulations and norms do keep enshittification in check. Perhaps this have happened with browsers, where the ease of porting one's browsing history, bookmarks, and extensions between Chromium-based browsers probably have constrain Chrome's enshittification somewhat (although it still continues: Chrome disabled uBlock Origin earlier this year).
However, data portability regulations don't prevent effective oligopoly with 3-4 dominant players. Competition will keep them in check and will slow down enshittification, but this will still be a *monoculture* to a large extent (with cookie-cutter, "one size fits none" apps in the language of Alex Komoroske), and the relative *power* on the internet will still be further concentrated in those handful mega-providers. These are distinct outcomes from enshittification per se, and they are not great.
The meta-concern with such regulations is that huge AI platforms may bite the bullet and implement them, but they may add a lot of burden for new players. Startups will choose not to enter these markets, defaulting to the US alone. Local innovation is also stifled.
https://psychedelicsblog.online/so-lovely-to-see-the-dmt-rescue-support-group-at-our-dogs-day-out-event-in-largs-scotland-today/