on with kara swisher

Reid Hoffman on Why We Should All Want AI to Be Our Co-pilot

The LinkedIn co-founder and AI evangelist makes his optimistic case to Kara Swisher.

Photo: Patrick T. Fallon/Bloomberg via Getty Images
Photo: Patrick T. Fallon/Bloomberg via Getty Images

Kara Swisher has gotten to know a lot of tech-industry people over the years, and as she explains to producer Nayeema Raza in this episode of On With Kara Swisher, she knows “the difference between jerks and people who really actually do care about something bigger than themselves.” Kara wholeheartedly believes LinkedIn co-founder Reid Hoffman falls into the latter camp, even if the two of them don’t always agree about the benefits and harms of new technologies such as artificial intelligence. Hoffman is an AI evangelist who is knee-deep in that world (including, until he recently stepped down, being on the board of OpenAI, the nonprofit behind ChatGPT and GPT-4), while Kara looks at the current AI frenzy and sees storm clouds ahead.

During her conversation with Hoffman, Kara asks the longtime tech entrepreneur and investor for his thoughts on a range of topics, from the collapse of Silicon Valley Bank to his political advocacy and ongoing fears about Donald Trump. She also grills Hoffman about his seemingly unflinching tech optimism; in the condensed segment below, she asks him to make his best case for several new AI-based technologies as well as explain what does, in fact, worry him about how AI could go wrong.

On With Kara Swisher

Journalist Kara Swisher brings the news and newsmakers to you twice a week, on Mondays and Thursdays.

Kara Swisher: I want you to explain some of these technologies in a lightning round. Tell me what the use case is in your assessment of how revolutionary it’s been or will be.

ChatGPT, the old iteration of OpenAI’s language processing?

Reid Hoffman: So the simple description is it’s like an amazing research assistant, across a wide variety of things, that gives you an instant answer. And what that will mean for everyone who does any professional activity — and I think “professional activity” is I consume information, I make a decision, I might make a decision about investing, I might make a decision about a prescription, I might make a decision about where to go, etc. — is there will be a co-pilot that will be between useful and essential for you doing that activity, you doing that job, within two to five years.

Swisher: Okay, so that’s a helpful assistant.

DALL-E, which was also OpenAI and more focused on visual creation, meaning mushing up pictures and things like that?

Hoffman: Yeah. You know, I wrote essays about DALL-E saying, look, this isn’t replacing graphic-designer jobs. This is amplifying whether or not you’re no skill, low skill, medium skill, or high skill. It makes you more effective in what you’re doing, and it gives even no-skill people in graphic design, like me, some graphics, some powers, where I can send my friend Kara, you know, a birthday card and kind of craft it myself.

Swisher: All right.

GPT-4: This is the new multi-language model that can interpret images along with presumably better answers, is a bigger language model, etc. What about that?

Hoffman: Well I published a book, as you know, called Impromptu, and I authored it with GPT-4. It gives much deeper, richer language information, and it can range from anything, you know, like dad jokes to like deep analyses. Like in my fireside-chat box with ChatGPT, I talked about, well, how does language mean in Wittgenstein’s theory of language and language games? You know, you could do this whole range of things. And that’s a lens into the assistant I was talking about.

Swisher: Yeah, you used one with Jerry Seinfeld, which was better than Wittgenstein.

DeepMind, which was acquired by Google many years ago? I actually broke that story when that happened, and I don’t think I realized how important it was.

Hoffman: Well, DeepMind was one of the organizations that kicked off kind of the new massive-scale computing iteration. You know, there’s Demis Hassabis; my co-founder in Inflection, Mustafa Suleyman; Shane Leg; you know, kind of brought this together and said, look, this could be super-important for solving a bunch of the world’s problems. Like, they’re working on protein folding, which could mean all kinds of things for medicine and scientific advancement. And so it’s one of the original gang, you know, in the revolution.

Swisher: Search engines powered by AI like Bingbot or Google’s Big Bard?

Hoffman: So I think one of the reasons why I think it’s fairly straightforward that this will be a major step forward is that frequently what people want is an answer to something versus ten blue links that they have to hunt through and see any of that information.

Swisher: So that it will be search on steroids in a lot of ways.

Swisher: So I was reading a recent blog post in which you talked about elevating humanity through tech. And it reminded me of a piece you wrote in The Atlantic earlier this year about how technology makes us more human. And then it reminded me of all our conversations where we were on opposite sides many, many years ago. Where I was like, “This is gonna end badly, Reid.” And you’re like, “No, it’s not this, this, this, and this.” I thought you were a bit rosy — anonymous trolls, disinformation. Talk to me about why you keep saying this. Like, I don’t think you’re stupid. That’s not what I’m calling you.

Hoffman: Thank you.

Swisher: But I am saying you gotta be kidding given where we’ve gone, right?

Hoffman: Look, I think there’s a whole bunch of stuff where even a bunch of the technologies that we’re having troubles with and working through today have added a bunch of different value to our lives. And, you know, obviously I could bore you to tears with talking about where I think LinkedIn is super-valuable and, you know, helps everyone’s lives.

Swisher: I don’t have a problem with LinkedIn.

Hoffman: But it isn’t that I deny that there aren’t challenges with currently deployed technology or being-developed technology, but I think that shaping the technology is the solution and that the solution can be so much better. So, for example, let’s talk about AI. If you say, well, I have a line of sight right now to every smart phone having an AI tutor and an AI doctor, that could be for everybody. Everybody in the entire world who has a smart phone. Delaying that is a huge cost in human suffering. Like, that’s a hugely valuable —

Swisher: So as opposed to the links, I’ve got a cough, and this says, “You need to get to the hospital.” That kind of thing?

Hoffman: Yeah. “Could be, in order of probability it could be these three things. Uh, let me ask you some questions,” right? “Uh oh. Shit. This is dangerous,” right? Even Bing Chat today, I was uh, a guy who’s working on some of this medical stuff at Microsoft was showing me Bing Chat live and was showing me how it would go, “Oh, stop talking to me.” You know, “Call 911!”

Swisher: Right, right, it can do that. So I know you’re an investor, and I believe you’re a believer. I want you to argue the other side, the Black Mirror version of AI. In 2020, Elon, when he was speaking to me, told me, “AI doesn’t have to hate us to destroy us.” I want you to argue that side.

Hoffman: Well, it depends on exactly which one you wanna do. The existential threat tends to be very science fiction. So it tends to be, you know, Hollywood, Terminator, you know, Ex Machina, etc. But what I would say is, look, what are the real dangers? Real dangers are: AI can be a bunch of tools in the hands of bad human beings, cybersecurity, other kinds of weapons. Like, for example, in terms of bad governments, it could be an instrument of oppression. If not being paid attention to, it could be an instrument of systematizing various forms, injustices, you know, whether it’s racial or economic injustice, with credit scoring or paroling or other kinds of things. All of those things, I think, are realistically possible.

Now the kind of thing you were gesturing at was like, well, it’s the so-called paper clip [thought experiment]: AI is told to maximize the number of paper clips, and it eliminates human beings in trying to do that. And, you know, that’s like saying, okay, well there’s one kind of supercomputer that controls the whole world and so forth. And so I do pay attention to this. It’s part of the safety and alignment stuff that all of the different AI groups that I work with [do], and I don’t see anything close to that.

This interview has been edited for length and clarity.

On With Kara Swisher is produced by Nayeema Raza, Blakeney Schick, Cristian Castro Rossel, and Rafaela Siewert, with mixing by Fernando Arruda, engineering by Christopher Shurtleff, and theme music by Trackademics. New episodes will drop every Monday and Thursday. Follow the show on Apple PodcastsSpotify, or wherever you get your podcasts.

More From 'on with kara swisher'

See All
Reid Hoffman on Why We Should All Want AI to Be Our Co-pilot