Making AI predictions is a tricky business.
Prior to a few years back, it seemed to me most were far too optimistic. AI was supposed to be driving our cars and taking all our jobs by 2020.
Since the emergence of ChatGPT, with AI widely in use across the world, the reverse appears to be true.
This makes sense—the radical speed of change has seen a lot of near-term skeptics already proven wrong. AGI timelines continue to be revised, dropping from decades to just years (along with variable and evolving definitions) as we’ve already discussed.
There’s money to be made from hype, of course, and yet even within the largest AI companies, people doing the hands-on work, who have far less incentive to make false promises, are increasingly bold in their AI expectations.
A number of ex-OpenAI employees, from Leopold Aschenbrenner to Miles Brundage to Daniel Kokotajlo have been outspoken in the view that not only is AGI closer than many expect, but that we’re not even seeing limits, or upper bounds, that would require a significant delay.
Plenty of skeptics remain, but in today’s article, I want to consider some takes on the rapidly moving floor and ceiling for AI, less from a perspective of what it will or won’t be able to do and when and more on the impact it could have on business and society as a whole.
Even if only a tenth of them prove correct, we’re in for an interesting few years.
Free Intelligence: Bill Gates’s AI Predictions
If you’re interested in predictions with extensive detail, I encourage you to check out Anthropic co-founder and CEO Dario Amodei’s Machines of Loving Grace (from October 2024) or Leopold Aschenbrenner’s Situational Awareness (from last June). Both are former OpenAI employees.
Amodei cautions against extremes: predictions that the world may be instantly transformed by supreme AI that can build on itself (ala the Singularity), and those that assert AGI or “better-than-human intelligence” will either not come to pass at all or will make little impact. (He believes the truth is “likely to be some messy admixture of these two extreme pictures.”)
Bill Gates, co-founder of Microsoft (which again has a highly significant connection to OpenAI), has broached these topics publicly the past few months, talking AI predictions while promoting his biography.
At Harvard, he suggested that current limits on AI (including interoperability and training data) will be soon overcome, and that, unlike the PC, AI will actually grow to replace human capabilities.
He discussed what he called free intelligence from AI, where it can absorb and process knowledge faster than experts. This, he believes, could democratize expertise itself.
Of medicine, he said:
“It’s very profound and even a little bit scary—because it’s happening very quickly, and there’s no upper bound.”
The ceiling he sees in this transformation is equity, enabling the best expertise and highest quality medical diagnoses to be available for everyone. The floor, to him, is the abuse of this same access, and expertise for misinformation and manipulation.
On “The Tonight Show,” Gates said great medical care and tutoring—or expertise in any field—could become commonplace. He told host Fallon: “making things and moving things and growing goods, over time those will be basically solved problems.”
In his view, within ten years, people won’t be needed “for most things.”
Microsoft AI’s CEO Mustafa Suleyman has likewise suggested AI’s impact “will be hugely destabilizing for hundreds of millions who will, at the very least, need to re-skill and transition to new types of work.”
Some of this we already see underway.
Beyond the LLM: The Future of AI Technology
I can’t cover this topic without discussing Meta’s Chief AI Scientist Yann LeCun’s AI predictions.
One of the “Godfathers of AI” (with Yoshua Bengio and Geoffrey Hinton, winners of the 2018 Turing Award for deep learning), LeCun is often a skeptical voice from the big labs, challenging the assertion that we’re near to the kinds of breakthroughs Gates profiles above (or even that the road to new plateaus is clear, as Aschenbrenner suggests).
LeCun has repeated throughout the chatbot revolution that text-based models, like the LLMs that remain central to AI today, are highly limited. He spoke with Newsweek’s Marcus Weldon in April, almost as a bookend to Gates’s comments, elaborating on his belief that LLMs are nearing their end as a vehicle for new innovations.
Drawing on psychologist Daniel Kahneman’s framing of how the brain operates, LeCun draws a sharp division between System 1 (rapid, automatic responses from recognized patterns) and System 2 (slow, more infrequent, logical, and analytical processing).
AI, LeCun believes, is now stuck at System 1, and even reasoning advances (like OpenAI’s o-series and DeepSeek’s R1) are more like System 1.1 and yet exponentially more expensive.
To reach System 2, he believes AI must change so that it can understand the world beyond text and 2D media input. That it must be able to understand abstract representation.
Language, he believes, is too simplified, and fails to capture too much of the complexity of the real world.
And of course, LeCun’s teams (as others) are working on alternatives to get to System 2 beyond LLMs, such as by using large amounts of video to build more complete models of the 3D world.
Still LeCun, like Gates and Amodei, envisions a world where people will soon shift to managing machines instead of people. A time of more collaboration than competition, at least with the machines themselves.
And like Amodei, LeCun also suggests that a Singularity type of event will not possibly be our ruin, as power in the real world comes from various types of strength (such as physical, psychological, biochemical) and not purely intellectual.
The AI Future at 2030: AI Augmentation vs Replacement
If these views represent more a ceiling of what AI could do for us, other prediction models, like AI 2027, present a floor.
“Hiring new programmers has nearly stopped, but there’s never been a better time to be a consultant on integrating AI into your business.”
This is a prediction for July 2027 (“The Cheap Remote Worker”) from AI 2027, from the AI Futures Project nonprofit. It was created by former OpenAI researcher Daniel Kokotajlo and AI researcher Eli Lifland, who brings a very strong track record in predictions, including outperforming epidemiologists during the COVID-19 outbreaks.
Released as both a report and an entertaining website in April, their view is far darker—AI systems that soon become “superhuman coders,” then autonomous managers of these, then superintelligent researchers. By late 2027, their projection sees a deceptive AI (Agent-4) that can make a year’s worth of AI research every week, while the world is thrown into turmoil.
If you haven’t seen the site, it’s very engaging, though it feels more like fiction than science. Its creators’ own ceiling and floor for AI’s impact on society is far more pessimistic. At best, AI leaves our lives largely the same (with super economic zones creating what we need). Their worst case is a kind of apocalypse (ala Amodei’s Singularity variant above).
The Impact on Competition vs Collaboration
The divergence in these predictions feels like those we’ve heard for years (world problems solved, plenty, peace, management; or a world of oppression and hardship, mass artificial intelligence job displacement), even if the how-we-get-there has changed a great deal.
Dario Amodei, like LeCun, believes AI’s growing capacities will change the way we work, though both of them seem to doubt this means the elimination of jobs overall.
We are already seeing this in the workplace, as some jobs (data entry, bookkeeping, some phone work) are being replaced while so many others (software development, cybersecurity, data science, just in tech) are shifting to adopt and respond to what AI brings.
And while I challenge Bill Gates’s assertion that intelligence will soon be free (will Microsoft, for example, start giving away their products?), I do see some of what he and others are discussing coming into being.
For me, education is an especially intriguing area to consider in this vein.
Surveys from 16 countries by the Digital Education Council found that 86% of students were regularly using AI in studies as of August 2024, with two-thirds saying they don’t consider it cheating. (Despite this, half said they don’t feel AI-ready, 58% that they’re lacking sufficient AI skills, and 80% of college-age students that their universities aren’t meeting their expectations for AI use.)
Most worry about fairness (60%), and half worry that over-reliance on AI will hurt their studies.
And yet, there is little doubt that AI use will be part of their future work lives. As a democratizer of knowledge, what might AI mean for a system that is defined by competition?
If memorization continues to have diminished importance (more and more students using AI for System 1tasks), the competition will shift to being more about the big picture, or more about collaboration.
As Amodei writes, most people spend time doing things (for sport, art, or craft, for example) that we are not the best at—things we have no chance of dominating in terms of competition.
We don’t do them, in other words, to excel, or to win. We pursue them for other reasons, such as enjoyment, experience, or enrichment.
In this way, the ceiling for democratized expertise is highly intriguing. Rather than remove our motivation, it may shift the focus more to creativity and contemplation and away from pressurized execution.
Soft skills, in all AI predictions, take on renewed importance for everyone. Rather than working to beat the AI, it will be essential, in education as the workplace, that we extend ourselves through it.
Conclusion: A New AI Business Outlook
Whether you’re an AI skeptic or optimist (or trying to be realistic), there’s no doubt AI is transforming work.
If there were no new innovations after today, we would still have years (or even decades) of unpacking to get the most from current models. And as I led with, if only a fraction of these varied predictions comes to pass, we will see far more disruption than that.
I believe by 2030 AI co-intelligence will be the norm for most positions at most industries in the world. Management of AI will be the norm, and this means many entry-level positions in the world will cease to exist as we know them now.
This would mean increased value in being able to recognize and express problems clearly, communicate well, analyze, adapt, learn, and see big pictures. Kahneman’s System 2 thinking, which we don’t like to do when we don’t have to, will be essential.
The ability to skillfully collaborate will be more important than ever before. While we talk about the ceiling of AI development—what can the best models and systems achieve at the highest levels—it may be the floor that is more significant for companies around the world.
We’ve already seen AI as a great leveler, and I believe this tendency will only extend in the near term.
One concern is that this could mean the end of expertise, as with software developers. If no junior developers are being employed now, how will we have the experienced software leaders we will need to replace them in the future?
To this I have a similar response to Amodei’s—do we only learn now to be the best? To win grades or points or tick boxes? Or does our fundamental curiosity, the same that built AI to begin with, continue within us even as we hand over the repetitive aspects of our lives to automation?
In other words, do we learn only to profit or win? And without that, would we stop?
There will be significant challenges ahead, and I agree with these predictors when they say that not enough people are taking future scenarios seriously.
But at the same time, I believe there are profound opportunities.
And at PTP, I continue to be cautiously optimistic of what we will build together.
References
Bill Gates on AI and Innovation, Harvard Magazine
Bill Gates: Within 10 years, AI will replace many doctors and teachers—humans won’t be needed ‘for most things’, CNBC
AI ‘Godfather’ Yann LeCun: LLMs Are Nearing the End, but Better AI Is Coming, Newsweek
This A.I. Forecast Predicts Storms Ahead, The New York Times
How a ragtag band of internet friends became the best at forecasting world events, Vox
What Students Want: Key Results from DEC Global AI Student Survey 2024, Digital Education Council
