The Condition of AI: An outlook on the globe in 2030

(As a part of this series, tune in to MIT Technology Review’s editor in chief, Mat Honan, along with editor at large, David Rotman, for an exclusive discussion with Financial Times columnist Richard Waters on the transformation of the global economy by AI. Live on Tuesday, December 9 at 1:00 p.m. ET. This event is exclusive to subscribers, and you can register here.)

Will Douglas Heaven states: 

Whenever I’m inquired about the forthcoming developments, I can’t help but get a Luke Haines song stuck in my thoughts: “Please don’t ask me about the future / I am not a fortune teller.” But here we go. What will life look like in 2030? My response: similar yet distinct.

Opinions are vastly divided when predicting the imminent effects of generative AI. On one side lies the AI Futures Project, a small, donation-based research organization led by former OpenAI researcher Daniel Kokotajlo. This nonprofit created a significant buzz back in April with AI 2027, an imaginative projection of what the world will appear as two years from now.

The narrative follows the extraordinary advancements of an AI enterprise called OpenBrain (any resemblances are coincidental, etc.) leading to a choose-your-own-adventure-style prosperity or calamity conclusion. Kokotajlo and his coauthors are clear about their belief that within the next decade, the influence of AI will surpass that of the Industrial Revolution—a 150-year era of both economic and social turmoil so profound that we are still existing in the world it generated.

On the opposite end, we have the team behind Normal Technology: Arvind Narayanan and Sayash Kapoor, researchers from Princeton University and coauthors of the book AI Snake Oil, who contest not only most of AI 2027’s projections but, more critically, its foundational perspective. They argue that’s not how technology operates.

While rapid advancements at the forefront may transpire swiftly, change across the broader economy and society as a whole evolves at human pace. The widespread acceptance of new technologies can be slow; acceptance itself is even slower. AI will be no exception.

What should we conclude from these opposing views? ChatGPT emerged three years ago last month, yet it remains ambiguous just how proficient the latest iterations of this technology are in replacing lawyers, software developers, or (gulp) journalists. Additionally, recent updates no longer produce the transformative enhancements they once did.

And still, this groundbreaking technology is so newly developed that it would be imprudent to dismiss it prematurely. Just consider: No one truly comprehends how this technology functions—let alone what its true purposes are.

As the pace of progress in core technology decelerates, the applications of that technology will become the primary differentiators among AI businesses. (Observe the new browser wars and the chatbot variety already available.) Concurrently, high-end models are becoming less expensive to operate and more reachable. Anticipate this to be where the majority of the action occurs: Innovative methods to utilize existing models will keep them engaging, distracting those eager for new advancements.

In the meantime, progress continues beyond LLMs. (Remember—AI existed before ChatGPT, and it will persist beyond it too.) Approaches like reinforcement learning—the engine behind AlphaGo, DeepMind’s board-game-playing AI that triumphed over a Go grandmaster in 2016—are poised for a resurgence. There’s considerable excitement surrounding world models, a genre of generative AI with a better understanding of how the physical universe operates compared to LLMs.

Ultimately, I concur with the Normal Technology team that swift technological progress does not directly correspond to immediate economic or societal changes. There’s simply too much chaotic human interaction in the mix.

But Tim, I now pass the floor to you. I’m eager to know what your insights reveal.

Will, I hold greater certainty than you that by 2030, the world will appear significantly different. In five years, I foresee the AI revolution advancing swiftly. However, who reaps those benefits will create a divide between AI fortunate and unfortunate.

It seems unavoidable that the AI bubble will burst before the decade concludes. Whether a venture capital funding shakeup transpires in six months or two years (I believe the current frenzy still has some momentum), many AI app developers will vanish overnight. Some will find their contributions absorbed by the models they rely on. Others will come to realize that selling services priced at $1 for 50 cents is unfeasible without a deluge of VC backing.

The survival of many foundational model companies is uncertain, yet it appears evident that OpenAI’s web of interdependencies within Silicon Valley renders it too significant to fail. Nonetheless, a funding reassessment will compel it to increase its service pricing.

At its inception in 2015, OpenAI vowed to “advance digital intelligence in the manner most likely to benefit all of humanity.” That increasingly seems implausible. Eventually, the investors who entered at a $500 billion valuation will seek returns. Those data centers won’t finance themselves. By then, many companies and individuals will have become reliant on ChatGPT or other AI services in their daily operations. Those capable of paying will enjoy the productivity advantages, acquiring the excess computing resources as others are priced out of the market.

Being able to stack various AI services together will create a cumulative effect. One instance I encountered during a recent visit to San Francisco: Resolving the issues in vibe coding involves several iterations on the same challenge followed by employing additional AI agents to identify bugs and security concerns. This seems highly GPU-intensive, suggesting that actualizing AI’s current productivity promises will necessitate customers to invest significantly more than the majority do now.

The same principle applies to physical AI. I firmly anticipate that by the decade’s end, robotaxis will be prevalent in every major city, and I predict we will even witness humanoid robots in numerous households. However, while Waymo’s Uber-like fares in San Francisco and the affordable robots made by China’s Unitree give the impression that they will soon be accessible to all, the computation expenses involved in making them practical and widespread seem destined to render them luxuries for the affluent, at least in the short term.

The rest of us, in contrast, will be left with an internet brimming with low-quality content and lacking the means to access functional AI tools.

Perhaps an advancement in computational efficiency will prevent this outcome. However, the current AI surge implies that Silicon Valley’s AI firms lack the motivation to develop more efficient models or explore radically different chip architectures. This only heightens the possibility that the next wave of AI innovation will emerge from outside the United States, whether from China, India, or even further afield.

Silicon Valley’s AI boom will undoubtedly conclude before 2030, but the competition for global dominance over technological development—and the political discussions surrounding the distribution of its benefits—seem poised to persist well into the coming decade.

Will responds: 

I share your view that the expenses associated with this technology will lead to a divide between the fortunate and less fortunate. Even today, spending over $200 a month yields a markedly different experience for power users of ChatGPT or Gemini compared to those on the free version. This disparity in capability is bound to grow as model creators aim to recover their expenses.

We are going to observe immense global inequalities as well. In the Global North, adoption has skyrocketed. A recent report from Microsoft’s AI Economy Institute highlights that AI is the quickest-spreading technology in history: “In under three years, over 1.2 billion individuals have utilized AI tools, which is a faster adoption rate than the internet, personal computers, or even smartphones.” Yet, AI is ineffective without accessible electricity and internet; vast regions of the world still lack both.

I remain skeptical that we will witness anything resembling the revolution insiders claim (and investors hope for) by 2030. When Microsoft refers to adoption, it’s considering casual users rather than assessing genuine long-term technological establishment, which requires time. Meanwhile, casual users often lose interest and move on.

How about this: If I find myself coexisting with a domestic robot in five years, you can send your laundry to my house in a robotaxi any day of the week.

Just kidding! As if that would be within my budget.

Further reading

What is AI? It may seem like a silly question, but it has never been more pressing. In this profound exploration, Will dissects decades of narrative and speculation to uncover the essence of our collective technodream.

AGI—the notion that machines might achieve human-like intelligence—has taken over an entire sector (and perhaps the US economy). For MIT Technology Review’s recent New Conspiracy Age series, Will presents a thought-provoking examination of how AGI resembles a conspiracy.

The FT explored the economics surrounding self-driving vehicles this summer, questioning who will shoulder the multi-billion-dollar expense to procure enough robotaxis for a major city like London or New York.
A valid counterpoint to Tim’s argument on AI inequities is that freely available open-source (or more accurately, “open weight”) models will continue to drive down prices. The US may desire leading-edge models built on US chips, but it is already losing ground to China in the global south regarding software.