</hed/> Reid Hoffman—The AI Enthusiast > <
/by/ Sam Blum ---------------------: As an expert in network effects, the LinkedIn co-founder and now AI entrepreneur is only too happy to spread (mostly) good cheer about AI and its societal potential. ---><= investor and LinkedIn co-founder Reid Hoffman, 56, was one of the chief architects of the social media age. Now he’s turned his attention to artificial intelligence. Besides investing in OpenAI, he’s a co-founder, alongside Karén Simonyan and Mustafa Suleyman, of the Palo Alto, California-based Inflection AI. To date, the trio has raised $1.5 billion from Bill Gates, Eric Schmidt, and Nvidia, among others, to build Inflection’s personal intelligence AI solution, dubbed Pi.
In March, Simonyan and Suleyman departed Inflection AI for Microsoft, and Inflection announced a pivot: It would primarily focus on creating and testing AI models for commercial customers, rather than a consumer-facing smart assistant. Hoffman remains as the company’s evangelist-in-chief.
In just over a year, the entrepreneur has visited the White House to talk
AI with President Biden; launched an interview-style podcast called Possible, offering the “brightest version of the future” of AI and other technologies; and published the book Impromptu, which was co-authored by ChatGPT and extols the virtues of AI. Here, Hoffman shares why he’s so optimistic about AI—and why he thinks you should be too. < /Inc.:> What’s behind the charm offensive?
< /Hoffman:> There are five billion smartphones; there could be a medical assistant on each of them. The amplification for humanity is obvious. Then go to education. If you have an infinitely patient tutor for humans of all ages on all subjects on every smartphone, think of what the amplification of human capability could be. I wanted to get that positive message out for what we can make, because this is the amplification of humanity. It can make us all very broadly a lot better. < /Inc.:> And what about all of the concerns? Surely, you’ve needed to address the naysayers.
< /Hoffman:> I knew we were going to have critics, just given the direction of the tech-lash and what the popular idea of artificial intelligence would be— that it’s going to seriously damage our democracy, it’s coming for our jobs, and maybe it’s coming for our humanity.
My worry with all the critics is that they think they’re doing this very prohumanist thing by saying what the risks are. Actually, they’re doing harm to human beings and society, because it’s like preventing the automobile. You can’t fix every possible problem before you get it on the road and start understanding how it operates.
< /Inc.:> As someone who is now working in AI—you reportedly scaled back your role at Greylock Partners last summer to focus more on the technology—what are the problems?
< /Hoffman:> Siri, Alexa, and Google Home mostly provide an audio interface to basic electronic functions on your specific devices. But, while any of them can pull up a recipe for spaghetti Bolognese from the internet, none of them could give you a recipe that’s based on the ingredients in your refrigerator. While any of them could set up a timer or list the news headlines of the day, none of them could produce a summary of the news tailored to your request. These capabilities are part of how all of the large language models massively outperform the traditional agents.
There are still lots of limitations, however. The primary one is figuring out how to make it economically availVenture able to everybody, because there is a cost of serving. But you learn to make an AI better by deploying it.
< /Inc.:> That’s a good segue into the business application for AI. Besides using ChatGPT to write your book, how do you as an entrepreneur and investor most often interact with generative AI tools?
< /Hoffman:> When I’m thinking about something that could be an investment, I’ll put information into ChatGPT, or into Bard, to sometimes compare and contrast. Because the systems are not exactly the same. They’re trained in different ways.
One investment I might be looking at is this really interesting intersection of artificial intelligence with synthetic biology—it will make a big difference with mRNA [the protein-generating molecule that led to the Covid vaccine]. I would ask the AI questions related to the investment: How will AI be enhancing our ability to diagnose and our ability to create mRNA vaccines? What would be the key risks? What are the key limiters right now?’
It isn’t that it answers every question. But it gives you a strong starting base, the same kind of way that Wikipedia in some contexts can be a starting base.
< /Inc.:> How do you think business leaders should begin talking about AI with their staff—without rattling them?
< /Hoffman:> I remember my grandfather would use his typewriter because he refused to use the computer. That’s not how progress works with
technology. If I’ve spent a decade, two decades, or 30 years developing my skills with a tool set, I might feel uncomfortable about change. Unfortunately, that’s one of the things you have to navigate. One of the ways I describe AI is it’s like the steam engine of the mind. The steam engine brought this massive transformation of society. It’s how we got the broad middle class. That transformation was painful. We got the Industrial Age; we had to figure out child labor, we had to figure out labor unions to navigate it. We’re going to be having the same kind of transformation of a whole bunch of human tasks with a steam engine of the mind. It will have transition issues, which we will need to navigate.
I don’t mean to downplay the difficulties that a number of real human beings are going to have in the transition. But in any time of societal change, you have difficulty. So, how do we navigate it with as much grace and as much humanity as possible? < /Inc.:> And how does Inflection’s own large language model, Pi, factor in? < /Hoffman:> You can have conversations with Pi on subjects ranging from poetry to sports, from emotions to facts, from pragmatic problems to spiritual considerations. Pi can be your research assistant, your brainstorming partner, your tutor, your sounding board. We’ve designed it to respond to user input in a very conversational way, so that its responses feel more like exchanges with a coach or a brainstorming colleague than a Wikipedia page. < /Inc.:> Where is AI heading in the next five to 10 years?
< /Hoffman:> The certain way to look foolish in five or 10 years is to offer very specific predictions, because the future is usually sooner and stranger than you might think. I hope we’ll figure out how to have medical assistants and tutors for everybody who has access to a computational device, within three to five years. And hopefully in 10 years, we’ll have AIs that are helping workers answer career questions. And I hope we’ll have personal amplifiers for certain careers, like for doctors, lawyers, and scientists. < /Inc.:> And how about artificial general intelligence, and advancements in which AI surpasses human intelligence? Nvidia’s CEO recently said we were just five years out from this. Any reason to start quaking in our boots?
< /Hoffman:> There are concerns here from reasonable people, and we should think ahead on long-term human-AI alignment. Some worry about hostile robots; some worry about the new sentient masters of the earth; others worry that we’ll simply become squirrels running across the road.
Of course, I look to work with intelligent versus blind optimism. I look at each frontier system and consider if a different, possibly concerning form of superintelligence might emerge, the first of which is how human bad actors might use those systems in destructive ways, like creating new bioweapons, criminals incorporating AI into their activities, or rogue states or other foreign actors interfering with democratic election processes or waging war.
So far, though, these forms of superintelligence have been enormously helpful to human capabilities.