Reid Hoffman on the Future of AI: Why He's Optimistic

The Potential of AI to Enhance Human Capabilities

In his latest book, “Superagency: What Could Possibly Go Right with Our AI Future,” LinkedIn co-founder Reid Hoffman proposes that artificial intelligence has the capacity to augment human agency, providing us with increased knowledge, improved employment opportunities, and enhanced quality of life, rather than diminishing it.

Hoffman, along with Greg Beato, doesn’t disregard the potential drawbacks of this technology. He characterizes his perspective on AI, and technology in general, as one of “smart risk taking” instead of uncritical optimism.

Focusing on Positive Outcomes

“Generally speaking, people tend to concentrate excessively on potential negative consequences, while insufficiently considering the possibilities for positive developments,” Hoffman stated.

While supporting “intelligent regulation,” he emphasizes that an “iterative deployment” approach – introducing AI tools broadly and responding to user feedback – is even more crucial for achieving favorable results.

“The advancements in car safety, such as brakes, airbags, bumpers, and seat belts, demonstrate that innovation isn’t inherently unsafe; it actually contributes to safety,” Hoffman explained.

Hoffman’s Background and Expertise

Hoffman previously served on the OpenAI board, currently sits on the Microsoft board, and is a partner at Greylock. During a discussion about his book, he highlighted the benefits of AI he’s already observing, its potential impact on the climate, and the distinction between an “AI doomer” and an “AI gloomer.”

The following interview has been edited for brevity and clarity.

Expanding on Previous Work

You previously authored a book on AI, “Impromptu.” What new insights did you aim to convey with “Superagency”?

“Impromptu” primarily aimed to demonstrate AI’s ability to readily amplify intelligence, both through demonstration and explanation. “Superagency,” however, centers on how human agency can be significantly improved, not only through enhanced capabilities, but also through the transformation of industries and societies as more individuals gain access to these technologies.

The typical discourse surrounding these advancements often begins with pessimism before evolving into a vision of an elevated state for humanity. AI is simply the latest in a series of disruptive technologies. “Impromptu” didn’t fully address the concerns surrounding the path to this more human future.

Defining Perspectives on AI

You categorize different viewpoints on AI into “gloomers,” “doomers,” “zoomers,” and “bloomers.” Let’s begin with “bloomers,” the category you identify with. What defines a bloomer, and why do you consider yourself one?

A bloomer is fundamentally optimistic about technology, believing that its development can be highly beneficial for individuals, groups, and society as a whole. However, this doesn’t imply that every conceivable technology is inherently positive.

Bloomers advocate for navigating with calculated risks, engaging in dialogue, and steering development. This is why the book emphasizes iterative deployment, as it allows for ongoing engagement and adjustments to ensure optimal outcomes.

Iterative Deployment and Democratic Access

When discussing steering, you mention regulation, but seem to believe iterative deployment, particularly on a large scale, holds greater promise. Do you think the benefits are inherent – that giving AI access to more people is inherently democratic? Or do you believe products need to be designed to solicit user input?

The approach can vary depending on the specific product. However, the book illustrates that simply engaging with a product – using it, avoiding it, or using it in specific ways – constitutes interaction and contributes to its shaping. Developers analyze user engagement, listen to online feedback, and respond to both positive and negative comments.

This feedback is a significant source of steering, separate from data collection or direct feedback mechanisms.

Addressing Concerns About Overwhelming Feedback

However, with the widespread popularity of tools like ChatGPT, individual objections might be drowned out by the sheer number of users.

Having millions of users doesn’t mean every objection will be addressed. Some might argue against any speed exceeding 20 miles per hour, but that’s a matter of opinion.

It’s the collective feedback that matters. If multiple people express similar concerns, it’s more likely to be heard and addressed. Furthermore, competition between companies like OpenAI and Anthropic encourages them to listen to feedback and steer development towards desirable features while avoiding undesirable ones.

Societal Concerns and Iterative Feedback

AI may pose risks that aren’t immediately apparent to individual consumers. Can iterative deployment address broader societal concerns?

That’s precisely why I wrote “Superagency” – to encourage dialogue about societal concerns. For example, some fear AI will diminish human agency. However, if most users don’t experience this loss of agency, that argument weakens.

The Role of Regulation

You’re open to regulation in certain contexts, but concerned about stifling innovation. What might beneficial AI regulation look like?

There are a few key areas. First, when addressing specific, critical issues like terrorism or cybercrime, narrowly targeted regulations can be effective. Second, innovation itself drives safety improvements, as seen with advancements in car safety features.

I encourage articulating specific concerns in measurable terms and tracking those metrics. If a measurement indicates a growing problem, further investigation and potential intervention can be considered.

Doomers vs. Gloomers

You distinguish between “doomers,” concerned about existential risks, and “gloomers,” focused on short-term issues like job displacement. Your book seems to primarily address the concerns of the gloomers.

I’m addressing two groups: those skeptical of AI, including gloomers, and those curious about it. I also aim to inspire technologists and innovators to prioritize human agency in their designs, leading to even more agency-enhancing technology.

Examples of AI Extending Human Agency

Can you provide current or future examples of how AI could extend, rather than reduce, human agency?

People often focus on the “superpowers” AI might grant, but superagency also arises when many people gain these capabilities, benefiting everyone. Cars are a prime example – they enable individual travel, but also allow doctors to make house calls, extending access to healthcare.

Today’s AI tools already offer superpowers, such as the ability to learn complex topics explained at different levels, identify objects from images, and perform various language tasks. When writing “Superagency,” I utilized AI to gain insights from a historian’s perspective, enhancing my research process.

Job Transformation and Skill Development

You acknowledge that new technologies often lead to the obsolescence of old skills, but also the development of new ones. However, there’s concern that reliance on AI like ChatGPT might discourage critical thinking and independent research.

This is a valid concern, similar to those raised about Google, search engines, and Wikipedia. The key is to learn when to rely on these tools, when to cross-check information, and how to assess the level of certainty. We’ve seen inaccuracies in information sourced from various online platforms, and learning to critically evaluate sources is essential.

As AI agents become more accurate, they can also provide context and highlight conflicting information, empowering users to make informed decisions.

Balancing Risks and Opportunities

You emphasize the importance of asking “What could go right?” alongside “What could go wrong?”

That’s the essence of being a bloomer. While optimistic about potential benefits, it doesn’t mean ignoring potential risks. The problem is that people generally focus too much on the negative and not enough on the positive.

AI and Climate Impact

You’ve suggested that the climate impacts of AI are often misunderstood or overstated. Do you believe widespread AI adoption poses a climate risk?

Fundamentally, no, or the risk is minimal. AI data centers are increasingly powered by green energy, and companies like Microsoft, Google, and Amazon are investing heavily in renewable energy sources. AI applications can also contribute to climate solutions, such as optimizing energy consumption in data centers. Furthermore, the energy consumption of AI currently represents a small percentage of overall energy usage.

Addressing Concerns About Growth

However, the growth of data centers and AI could be significant in the coming years.

That’s possible, but the investment in green energy is crucial.

The McKinsey Mindset and Job Displacement

Ted Chiang’s essay highlights a concern that companies often view AI as a cost-cutting measure, leading to job elimination rather than unlocking new potential. Is this a worry for you?

I am concerned, particularly regarding the transition period. Historically, technological transitions have been challenging, and this one will likely be no different. “Superagency” aims to learn from past experiences and develop tools to navigate this transition more effectively. Job displacement is a real concern, especially in areas like customer service.

However, AI can also enhance productivity and create new opportunities. As AI makes employees more effective, companies may hire more people in other areas. The key is to focus on job transformation and provide individuals with the skills they need to adapt.

AI can also assist in learning new skills and finding suitable employment. While the transition will be difficult, we can strive to make it more graceful.