OpenAI thinks it might be a good idea for AI to text your friends for you.
Wait, what?
ChatGPT can now connect with Apple Messages on a Mac, allowing it to search conversations, analyze messages, draft replies, and send messages for you. It can work with iMessage, SMS, and RCS messages. Sending a message requires your approval by default.
But let’s back up for a second.
Do we want AI texting our friends in the first place?
And now we have to ask an even stranger question:
When you’re texting your friend, how do you know you’re actually talking to your friend?
What If Your Friend’s AI Is Doing the Talking?
Imagine you send your friend a text.
You know your friend. You’ve been talking for years. You know how they write. You recognize their sense of humor. You know their little expressions and the way they respond.
You send:
“Hey, are you still coming Saturday?”
A few seconds later, you get a perfectly normal response.
“Of course! Wouldn’t miss it.”
You continue the conversation.
Except your friend didn’t write that.
Their AI did.
Now flip it around.
What if your AI is doing the same thing?
Suddenly, the two of you could be thinking you’re having a conversation with each other while your AIs are actually generating the responses.
Your AI could be talking to their AI.
Yikes.
And the really strange part is that the conversation might be perfectly good. The AIs could understand the context, remember what was discussed previously, and respond naturally.
But it wouldn’t really be your conversation anymore.
OpenAI Is Giving AI Access to Your Messages
The new ChatGPT integration goes beyond simply helping you come up with a sentence.
You can ask ChatGPT to search your messages, find information from previous conversations, suggest follow-ups, analyze conversations, draft responses, and send messages. OpenAI says the Messages integration only accesses messages when you specifically ask it to.
The approval requirement is an important safeguard.
But the bigger issue isn’t whether ChatGPT asks you before pressing Send.
It’s that we are starting to normalize the idea that AI should participate in our personal conversations at all.
“Help Me Write This” Is Different
There’s an important distinction here.
If you tell ChatGPT:
“I’m trying to tell my friend I’m sorry I missed her birthday. Can you help me word this?”
That’s AI assisting you.
You’re still deciding what you want to say.
You’re still thinking about your friend.
You’re still having the conversation.
But:
“Text my friend and tell her I’m sorry I missed her birthday.”
is something different.
Now the AI is taking over part of the interaction itself.
And if you eventually say:
“Handle my messages while I’m working.”
the human involvement gets even smaller.
At some point, you aren’t using AI to communicate better.
You’re using AI to communicate for you.
This Is Exactly Where the Human Question Comes In
This follows what Bill Gates recently said about AI: we should choose which activities we want to remain human, even when AI is capable of doing them.
Gates has proposed the idea of “Human Reserved” jobs and activities – deliberately leaving certain work in human hands even when AI or robots could perform it. He compares the idea to a nature reserve: society could develop certain areas, but chooses not to because preserving them is more important.
Texting your friends may be a pretty good example.
When you text a friend, the point isn’t simply to transfer information from your phone to theirs.
The human interaction is the point.
Some are pushing the idea that if technology can save us time, we should use it.
And in many cases, that’s absolutely true.
Let AI summarize a giant document.
Let it help debug code.
Let it organize information.
Let it handle repetitive administrative work.
But there is a difference between eliminating tedious work and eliminating human interaction.
If AI writes the email to your insurance company, that’s one thing.
If AI writes your text to your best friend, that’s another.
The insurance company probably doesn’t care whether you personally chose the words.
Your friend might.
What Happens When Nobody Knows?
What if people don’t tell their friends they’re using AI to communicate?
Your friend may assume the message came directly from you.
You may assume their message came directly from them.
Neither of you knows what happened behind the scenes.
And because today’s AI is increasingly good at producing natural-sounding language, you might not be able to tell.
The old assumption was:
When I’m texting my friend, I’m talking to my friend.
The new reality could become:
When I’m texting my friend, I’m talking to whatever system my friend has instructed to communicate with me.
That’s a pretty big change.
And then there’s the really strange possibility.
Your AI knows how you communicate.
Your friend’s AI knows how they communicate.
Your AI has your conversation history.
Their AI has theirs.
Your AI is handling your routine messages.
Their AI is doing the same.
Eventually, the two systems could become extremely good at communicating with each other.
Maybe they make plans.
Maybe they decide where you should meet for dinner.
Maybe they even have a more efficient conversation than the two humans would have had.
Meanwhile, you’re both sitting there thinking:
“Wow, my friend really gets me.”
No. Your AIs get each other.
That’s funny when you say it that way.
It’s also a little horrifying.
Wait, Who Am I Talking To?
OpenAI’s new Messages capability may be useful. There are legitimate reasons someone might want help finding an old conversation, remembering information, or composing a difficult message.
But giving AI the ability to communicate directly with the people in our lives crosses into very different territory.
Because once AI can speak for us, we have to start thinking about what happens to the meaning of the communication itself.
If your friend sends you a message, do you want to know that your friend actually wrote it?
For many people, the answer will be yes.
And if the human part of the interaction is what gives it meaning, texting your friends is exactly the kind of activity we shouldn’t be so eager to automate away.
Because someday, when you get a text from your friend, you might find yourself wondering:
“Wait. Who am I talking to?”