← Back to The Atlantic

Jensen Huang’s Apology May Be the Only Correct Use of AI

The Atlanticby Alexandra PetriOpen original ↗

Look, I just want to say that if Nvidia CEO and reported spy-plot target Jensen Huang did get help from AI to write an apology email to his family and company about his affair (he’s said he uses AI to help write all of his first drafts), this might be the one use of AI that makes sense to me.

People (me) are always saying about AI, “This is a solution looking for a problem!” and “Who would possibly want to delegate things like knowing information about your children or creating works of art to a third-party series of zeroes and ones?” But there are some things in life that nobody wants to do, such as get a root canal, amputate a limb, or deal with your cable provider. Or, perhaps, apologize simultaneously to your employees and your wife of 40 years for jeopardizing your business and personal relationships and perhaps also national security? (To me, though, those feel like two separate emails.)

But I do see one problem for someone attempting to delegate the writing of this email: I don’t think this specific email has ever been written before, by anyone. “Business/Personal Email Apologizing for Affair/Possible National-Security Breach” is the email equivalent of those wedding dress codes—“Beach Formal”; “Back-of-Your-Closet Country Club”; “Cocktail Business Black Tie”—that are beyond what the human brain can understand.

I’m just wondering what training data set one would even consult in this scenario. What template was AI going to work with here? James Bond scripts? (“I owe everyone here at Goldfinger Enterprises an apology. My fixation on James Bond and the time I have wasted designing a laser specifically targeted at his groin have distracted us from our mission of blowing up Fort Knox to increase the value of our gold supplies. To my many colleagues, especially Odd Job, I value our working relationship. Your dedication to throwing a sharp hat has not gone unnoticed, and you deserve better from me going forward.”) How many emails were in this data set? Were there, in fact, multiple emails from people apologizing to their families and employees simultaneously for potentially derailing the world economy over their relationship with a foreign spy that it could study? Is this a widespread problem?

I understand that Donald Trump says of Huang that “he dealt with somebody, but you know, so does everybody else,” but—I actually don’t think this situation is that common! I’ve never come across it in my life. Or if there is a spy trying to infiltrate my marriage to obtain my trade secrets (you will claw my log-in verification codes from my cold, dead hands), that spy is not very good at his or her job. But maybe I’m out of the loop! Maybe people just live like this.

Now I am worried, based on Jensen Huang’s claim that this was “completely normal and not unusual” and an Nvidia spokesperson’s insistence that his interactions with the alleged spy were “not unusual” that, actually, we’ll be seeing multiple emails like this. Finally, organic demand for AI! If so, the template will get better and better—that is, if we can still use the chips.

Comments (0)

Comments (0)

No comments yet.