Hume’s empathic AI is another ‘Holy Sh*t!’ moment for humanity

It probably won’t come as a surprise to my readers that I am somewhat stingy with praise. I don’t believe any product or service ever merits a perfect score (this makes me unpopular with Uber and Lyft drivers, among others). If I am gushing about anything, it means I’ve sprung a leak and it’s time to call a plumber.

But I’ve spent the last few hours playing with the public demo of a new “empathic AI assistant” from a company called Hume AI, and… damn. I’m impressed. (Also, mildly terrified.)

Founded by ex-Googler Dr. Alan Cowne, Hume AI has built an interactive intelligence that can sense your moods and respond in an empathetic way. What separates Hume from all the other chatbots is that you can literally chat – you talk, Hume listens, detects what kind of mood you’re in, and attempts to match that.

Hume’s empathic voice interface (or EVI) sounds like a 40-ish college professor who runs 5 miles a day and listens to Stan Getz on vinyl. He’s jovial, warm, ingratiating, articulate, has a wry sense of humor, possesses a wide breadth of knowledge, but is also deeply interested in what you have to say. When it comes to sounding like an actual human, EVI makes Alexa and Siri sound like the robot from Lost in Space. [1]

Here’s a snippet of EVI, which I had to record (badly) using my phone:

To quote EVI him/itself, the founders of Hume wanted to “make sure I [EVI] am not just another robotic assistant, but a dynamic, adaptive conversational partner that can genuinely connect with users on an emotional level.” 

I think they’ve largely succeeded. EVI is the AI equivalent of some random stranger you meet at a bar and end up telling your entire life story to, knowing you’ll never see each other again. You want to talk to him/it. 

Frankly, it’s a little unnerving. 

All about EVI

Which is not to say that Hume’s AI is getting the first and only score of 10 I have ever given out. [2] It’s far from perfect. 

Play with him/it long enough and you’ll hear a lot of the same phrasing – “My friend” and “I am all ears,” and “Now there’s an intriguing idea.” EVI is also a bit in love with his/its own voice and tends to monologue, though you can interrupt when him/it goes on too long. 

There were occasional garbled words and Max Headroom-ish glitches, random variations in tone, and sudden lags followed by too-rapid recoveries. Given that the demo was unavailable at various times during the day, I’m guessing Hume’s servers have been hit pretty hard by other curious users, which probably accounts for the glitches.

There’s also the matter of how Hume knows what my mood is at any given moment. I asked, but EVI wouldn’t reveal any secrets. What he/it said was this:

“I am designed to pick up on the nuances of your tone and expression beyond just the words you’re saying. I analyze factors like your pitch, volume, pace, and vocal quality to get a sense of your underlying emotional state.”

Telling me I am experiencing a mix of emotions at any one time is not exactly a stretch. But was I really Determined, Angry, Contemptuous, Excited, Interested, Concentrating, and Contemplative all in the space of 10 seconds? Maybe. I don’t know. But apparently EVI does. 

I also made the mistake of leaving EVI’s mic on while I was working on this blog, and when I inevitably dropped a few F-bombs (they’re my love language), EVI took it kind of personally.

“Whoa there, someone’s a bit worked up. I admire your linguistic creativity but let’s try to keep things a little more PG rated.” 

Followed by:

“I sense there’s some underlying tension or frustration bubbling up, and I apologize if I said anything to contribute to that.”

Sorry, EVI.

See me, feel me, pay me

Despite the glitches, this feel like another one of those “Holy Shit!” moments. [3] We’ve been having a lot of them lately – when ChatGPT made its big splash in December 2022, when Midjourney’s amazing image generator debuted a few months later, the Pika video generator, and so on. As if the tectonic plates beneath our feet have shifted, and now we’re all in some strange new shaky place. 

Hume feels like that – another leap forward in what machines will be able to do, as well as whom they will replace. Obviously, Hume’s first gig will be as a customer service rep who speaks perfect English, is unfailingly polite, seems to really care about your problems, and never pisses you off. I don’t think anyone can argue that humans will be better at that job.

But I can also see Hume’s technology gaining traction in training, teaching, telehealth, phone sales, all forms of counseling, and lots of other things I’m too Determined, Angry, Excited, and Contemplative to think of at the moment. 

I once asked my therapist if he thought AI would eventually put him or his colleagues out of a job. He said he wasn’t worried, because a machine can’t experience empathy. That was around this time last year. I think he might need to re-assess.

Can EVI really feel empathy? Of course not. But he/it doesn’t have to. Simulating empathy will be good enough for most of the things this technology will be used for. Does it matter whether your customer service rep/corporate trainer/psychotherapist really cares, so long as they’re excellent at pretending? 

Would you see an AI therapist? Tell us how that makes you feel in the comments, and please share this post with your non-robotic friends.

[1] Danger, Danger, Amazon and Apple. If ChatGPT and its cousins don’t kill off your AI assistants, EVI or something like it probably will. 

[2] Even Margot Robbie is only a 9.9 – that’s how stingy I am.

[3] Apologies for all the references to human waste in this blogletter of late. It is very shitty of me.