• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI
    Reaction

    The privacy trade-off in giving AI assistants more access

    A writer argues that withholding email, files and messages could limit an AI assistant’s usefulness.

    JE
    JC
    3 Sources, ,

    TLDR

    A writer argues that access to calendars, email, files and messages could help AI assistants connect information and carry out tasks. They call the potential loss of capability from withholding access a “privacy tax.” They also warn that granting access may expose other people’s information and doubt that permission warnings alone can resolve the trade-off.

    Combined views

    1.2K

    3 Sources, first seen 5h ago

    Combined views

    1.2K

    3 Sources, first seen 5h ago

    9 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    5h ago
    first seen 5h ago
    9 likes
    3 comments
    4 saves
    3 comments
    4 saves

    3 Sources

    @joshelmanApple understands this privacy vs utility balance extremely well. Privacy can be a reason to choose an iPhone because the iPhone is already a great phone. If the private phone were dramatically worse, the decision would get harder. AI pushes that problem much further because the value of context can compound. The payoff from sharing more is also immediate. You grant access and the answer gets better. Something that took fifteen minutes takes two. The agent finds something you forgot. Eventually it handles a task before you remember to do it. The cost is much harder to see. You click Allow and nothing bad happens. The product simply becomes more useful. So the next permission gets easier. After enough of those decisions, taking access away can feel less like protecting your privacy and more like turning features off. That creates what I think of as a privacy tax. You can protect more of your information, but you may pay for it in capability. And there’s another problem hiding inside this. A lot of the information you’re granting access to isn’t really yours alone. Your inbox contains things other people told you. Messages contain private conversations. Contacts contain their phone numbers. Photos contain their faces and locations. Your files may contain information about coworkers, customers, friends and family. You chose the AI. They didn’t. One person can make their AI substantially smarter by exposing information about hundreds or thousands of other people who never made the same trade. That makes the old idea of a permission dialog start to feel inadequate. The dialog assumes the important question is whether an app can access something. But an agent can potentially read that information, reason over it, connect it to other information and take actions based on what it finds. The permission can stay around long after the moment when you clicked Allow. Apple’s changes to Full Disk Access this week feel like an early sign that platforms see where this is heading. Full Disk Access used to be something most people might grant to a backup app. Now people are giving agents access to huge portions of their computers because that access is exactly what makes the agents useful. Making the warning more explicit is a reasonable response. I’m just not convinced warnings can solve the underlying problem. If the thing behind the warning gets meaningfully better when you press Allow, the warning is fighting the product. People learn which button makes the useful thing happen. We’ve trained ourselves to do that for decades. With warnings you can get immune to the blithely clicking through on any app you want to try. And then miss protecting yourself against the rogue app where you give it all way.5h
    @JoshConstine@joshelman We need an AI loyalty benchmark to measure what else we’re trading off for “better” AI. Is my agent doing its best for me? Or trying to minimize token spend, maximize retention, and drive me to its rev share partners?1h
    Featured Source

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    3 Sources

    @joshelmanApple understands this privacy vs utility balance extremely well. Privacy can be a reason to choose an iPhone because the iPhone is already a great phone. If the private phone were dramatically worse, the decision would get harder. AI pushes that problem much further because the value of context can compound. The payoff from sharing more is also immediate. You grant access and the answer gets better. Something that took fifteen minutes takes two. The agent finds something you forgot. Eventually it handles a task before you remember to do it. The cost is much harder to see. You click Allow and nothing bad happens. The product simply becomes more useful. So the next permission gets easier. After enough of those decisions, taking access away can feel less like protecting your privacy and more like turning features off. That creates what I think of as a privacy tax. You can protect more of your information, but you may pay for it in capability. And there’s another problem hiding inside this. A lot of the information you’re granting access to isn’t really yours alone. Your inbox contains things other people told you. Messages contain private conversations. Contacts contain their phone numbers. Photos contain their faces and locations. Your files may contain information about coworkers, customers, friends and family. You chose the AI. They didn’t. One person can make their AI substantially smarter by exposing information about hundreds or thousands of other people who never made the same trade. That makes the old idea of a permission dialog start to feel inadequate. The dialog assumes the important question is whether an app can access something. But an agent can potentially read that information, reason over it, connect it to other information and take actions based on what it finds. The permission can stay around long after the moment when you clicked Allow. Apple’s changes to Full Disk Access this week feel like an early sign that platforms see where this is heading. Full Disk Access used to be something most people might grant to a backup app. Now people are giving agents access to huge portions of their computers because that access is exactly what makes the agents useful. Making the warning more explicit is a reasonable response. I’m just not convinced warnings can solve the underlying problem. If the thing behind the warning gets meaningfully better when you press Allow, the warning is fighting the product. People learn which button makes the useful thing happen. We’ve trained ourselves to do that for decades. With warnings you can get immune to the blithely clicking through on any app you want to try. And then miss protecting yourself against the rogue app where you give it all way.5h
    @JoshConstine@joshelman We need an AI loyalty benchmark to measure what else we’re trading off for “better” AI. Is my agent doing its best for me? Or trying to minimize token spend, maximize retention, and drive me to its rev share partners?1h