Can AGI become a tide that lifts all boats, not just the yachts?

I still remember that one random summer day when the YouTube algorithm queued me to this video and Ray Kurzweil’s work. He talked about the S curve law of evolution, present in biology and technological progress. And this curve seems to be just as relevant in AI.  

Life on Earth began with single-celled organisms that dominated for billions of years. For a long time, evolution seemed slow. But then came the Cambrian Explosion—around 541 million years ago—when complex, multicellular life burst onto the scene in a geological blink of an eye. From there, the pace quickened. Mammals emerged, then primates, followed by hominids. Homo sapiens only arrived around 300,000 years ago, and yet, in just the past few thousand years, human civilization has rocketed forward.

The story is similar in technology, particularly in the evolution of computing. Kurzweil often cites Moore’s Law—the observation that the number of transistors on an integrated circuit doubles roughly every two years—as one of many exponential trends. Each of these innovations had its own S-curve:

1.     a slow beginning

2.     a period of explosive improvement

3.     and then a plateau as physical or theoretical limits were reached.

But just as one technology began to falter, a new one would emerge to take its place, as shown in the graph below. When the First Technology plateaus, it is already the beginning of the Second Technology.

You can see this happening within the timeline of, for example, Netflix.

And then applied to the concept of sustainability in fashion.

And a wealth of other innovations.

He predicts that we will see the same thing happen in AI, and that is what he calls the Singularity.

The Singularity is the moment when machines become smarter than humans and begin improving themselves autonomously, leading to an intelligence explosion.

But if we are also in a S curve now, where in the S curve are we?

The slow adoption period? The fast acceleration? Or are we already at the plateau?

Most would predict we are in early acceleration stage. But as history tells us, you really only know when we are zooming out.

No matter where in the S curve we are, we are on one. And even if we have reached the tail end of this one (the multitude of reasons include: we are running out of training data, reaching capacity in compute power, rising energy costs and lack of energy to support computation), Ray’s theory tells us we might just be at the beginning of the next S curve.

But first, what is AGI?

Before we reach the stage of machines smarter than us (Artificial Super Intelligence), we first need to reach Artificial General Intelligence. Unfortunately, if we ask ten AI researchers for a definition of AGI, we get fourteen responses.

There are so many definitions out there but one we can use is one from Nils J. Nilsson (Stanford AI pioneer) who calls it:

“A machine that can perform any intellectual task that a human being can.”

This classic definition frames AGI as general-purpose intelligence, not bound to narrow domains.

Summed up in a table, this is what AGI represents.

Trait	Description
Generality	Not task-specific—can reason across many domains
Autonomy	Operates independently with minimal human input
Learning and Adaptation	Learns from experience and improves over time
Goal-Directed Behavior	Able to pursue objectives, plan, and make decisions
Transfer Learning	Applies knowledge from one task to another
Human-Level or Beyond	Matches or exceeds human performance in cognitive tasks

This contrasts with Narrow AI, of which generative AI is a part of, which are AI systems that are bound to narrow domains and can only operate on a specific number of tasks. This is what dominates the AI field already and technology that is existent.

AGI is usually defined relative to the total capabilities of an average human, not the sum total of all specialized humans. But who is this average human? Is this person from America, Nigeria or China? Is this person someone with a decade of work experience or someone who has just graduated college? This is why AGI is so hard to define and might make it into a cultural discussion instead of a technical one. In many ways, the AI we have today is multiples smarter than a human. It really depends on which “human” you’re picking.

In the end, it may not really matter because we may define AGI as a phase. For some countries, it will come sooner, for some it might come later. When we are in the future, we might be able to look back and decipher that it was in the year 20XX that we did achieve AGI collectively.

So we’ll have AGI, and then what?  

A few months ago, a paper was released by a couple of AI researchers. It talked about AI 2027. In three short years, an AI race between US and China erupts. At the end of the paper, there is a choose-your-own-adventure ending where you can pick from slowdown to acceleration. (Obviously, the authors are vouching for a slowdown). If you like to read about doomsday scenarios, have a read. I personally think this takes thing a little too extreme. Here I add a long footnote: The problem with papers that are published by AI researchers is that I think they forget about the rest of the world sometimes. And I’ve lived in the San Francisco bubble before when everyone around you does work out of an Apple Vision Pro (for a while) or uses a fleet of AI agents to complete their W2 jobs. I think this group overestimates the pace with which the world shifts. Acceptance of the mere use of AI still is an uphill climb. So the change is advancing, just not at the extreme pace that this paper predicts. (Well you can check back this article in 2027 and ping me about the results.) 

The Transition

These days, as I do my keynotes, I joke with the audience, hang on to those spreadsheets that you hate doing because in the future, you will reminisce them! Those pieces of technology might bring about nostalgia like tape recorders and gramophone.

There is a wealth of evidence that points to a decrease in entry level jobs. College graduates are fumbling over employment because the things that normally fell on an entry level worker were tasks that the senior employees never wanted to do. Now these tasks are simply handed to AI. There is a conundrum in the job market in that it’s hard to get hired without experience and most college grads don’t have experience. After all, the marker for success for many companies now isn’t headcount. It’s revenue per employee. So the fewer heads you have the better the company looks.

This is also the first time in history where MBAs from elite universities like Harvard have trouble finding jobs.

Recently Bill Gates came out with the notion that in the future, there are three roles that will remain unscathed by the destruction of AI. The first is surprisingly, programmers. While AI is capable of generating code and automating certain programming tasks, it still lacks the ability to handle complex software development, debugging, and refining code with the same precision and problem-solving skills as human programmers. AI tools can assist coders, but human expertise is still needed to guide AI's evolution in coding and software development. The second are Energy Experts. The energy sector, encompassing oil, nuclear, and renewable energy, is a complex and critical industry that requires strategic decision-making and human expertise to manage infrastructure, navigate industry challenges, and innovate for the future. While AI can assist in analyzing data and improving efficiency in the energy sector, it cannot fully replace human expertise in decision-making and crisis management. The last is Biologists. AI is becoming a powerful tool in biological research, particularly in areas like disease diagnosis and DNA analysis, but it currently lacks the creative and intuitive thinking necessary for major scientific breakthroughs. Biologists are still crucial for formulating hypotheses, designing experiments, interpreting results, and making the kind of leaps in scientific understanding that lead to discoveries.

As AGI permeates knowledge work, there will be several waves of roles and industries affected first. Apart from the obvious jobs such as administrative support jobs, customer service jobs and entry-level jobs, what other jobs are going to be affected in the first wave? In fields like finance, insurance, and analytics, AGI will excel at data-crunching, pattern recognition, and prediction. Tasks performed by analysts, accountants, and bookkeepers are highly automatable. An AGI can instantly analyze financial reports, audit records, or market data – work that occupies armies of junior analysts today. Roles such as bookkeeping and accounting clerks and entry-level financial analysts will be pruned or radically redefined.

What can we learn from the Industrial Revolution?

To predict the future, maybe we first need to look back at history. What’s happening to knowledge workers now with AI is similar with the automation felt by factory workers during the Industrial Revolution.

In the late 1800s the United States raced ahead on the back of railroads, steel, and mass-production factories. A handful of industrialists—names like Rockefeller, Carnegie, and Vanderbilt—became unimaginably wealthy, while many ordinary workers toiled long hours in dangerous conditions for low pay. Because there were no real antitrust rules, no income tax, and hardly any labor protections, the gains from all that new technology and efficiency flowed almost entirely to the owners of capital.

Inequality soared.

We might be on the same path when it comes to AGI. Power might be concentrated in the hands of the few who know how to use AGI to their advantage. AGI-driven automation favors workers with advanced skills, education, or the ability to quickly learn and adapt. High-skill professionals (engineers, data scientists, managers, etc.) will see their productivity enhanced by AGI and likely command even higher salaries as their value-add grows. There will also be strong demand for AI-specialized talent, driving their wages up. In contrast, workers in roles that don’t require higher education – factory workers, clerks, service employees – face job erosion or stagnating wages due to labor surplus.

In the U.S., we could see something like the early Industrial Revolution pattern: soaring productivity and wealth creation, but labor’s share of that wealth falling – until social and political responses forced a rebalance. In concrete terms, the top 10% of earners (especially the top 1%) might capture a greater share of national income, whereas the bottom 50% might see their share decline. If no interventions occur, social stratification could become more extreme, potentially leading to unrest or political upheaval in later years.

Income Polarization

We can expect a continuation (and amplification) of the polarization trend: high-wage jobs grow in number and pay, low-wage jobs also grow but often pay poorly, and middle-wage jobs shrink. Displaced middle-class workers may be forced into lower-paying work. For example, a laid-off administrative assistant might compete for jobs as a retail supervisor or medical receptionist, increasing competition in those lower-wage fields. McKinsey finds that workers in lower wage quintiles are 10–14 times more likely to need to transition to new occupations by 2030 than those in the top quintile– a stark indicator that the lowest rungs will feel the brunt of change. Women and certain minorities might also be disproportionately affected, since they are overrepresented in some automatable roles (e.g. women in clerical and service jobs).

Regional Disparities

Certain regions of the U.S. could face more acute disruption. Areas heavily reliant on manufacturing or routine back-office jobs (e.g. some Midwest towns with call centers or Southern regions with warehouses) might see concentrated job losses. Meanwhile, tech-centric regions (Silicon Valley, Seattle, Research Triangle, etc.) or cities with diversified, high-skill economies could thrive. If left unaddressed, AGI may widen the gap between dynamic metro hubs and struggling rural or rust-belt areas, fueling geographic inequality. This echoes what Brookings Institution found with earlier automation: it hits some communities much harder than others, exacerbating regional inequality.

No Immediate Mass Unemployment, More gig work

It’s important to emphasize that AGI’s impact is cumulative and accelerating, not a sudden one-time shock. We are unlikely to see 30% unemployment overnight. Instead, the labor market will initially absorb some displaced workers through new job creation and redeployment – for instance, some workers will find jobs in the expanding healthcare, green energy, or tech sectors. The unemployment rate might remain reasonably low in the first few years if the economy is well-managed (especially if there’s growth in other sectors offsetting losses).

If a job can be partially automated, employers gain bargaining power to pay less or make roles part-time/contract. Without strong labor institutions or new standards, more workers could face gig-like employment conditions. AGI might also spur job growth in new areas that we can’t fully predict – possibly creating a new middle class of AI-era technicians or creative entrepreneurs. Additionally, if the work week could be shortened (splitting the available work among more people) without loss of pay due to higher productivity, quality of life could increase for everyone. But such outcomes will not happen automatically. They require deliberate action to redistribute and reinvest the gains of AGI.

Can AGI become a tide that lifts all boats, not just the yachts?

There are numerous experiments being run for life after AGI. You might have heard of Universal Basic Income as one that is most common. Experiments have been run in Canada, Wales, Norway and California. Across very different settings, UBI/guaranteed-income pilots repeatedly improve mental health, reduce poverty stress and sometimes even raise full-time employment. The results were opposite to stimulus checks handed out during COVID, where most people simply took the money and bought a fancier coffee machine.

Robot Tax or “AI Automation Tax”

Bill Gates has been talking about this one – to ensure companies share the benefits of automation with society, policymakers could introduce a tax on extreme gains from labor-displacing AI. One proposal is a modest tax on the use of AI/robots that replace human workers. For instance, if a company automates away 100 jobs, it might pay a tax equivalent to a portion of those lost wages into the “AI Transition fund”. This both generates revenue for retraining and slightly tempers the incentive to fire humans purely for profit. As a concrete idea, companies benefiting from AI-driven automation could be taxed to help finance UBI or retraining programs. Another approach is to create an “AI dividend”: if AGI boosts GDP significantly, some of that new wealth is distributed per capita to all citizens (similar to Alaska’s oil dividend, but for AI productivity). This would directly mitigate inequality by providing everyone a basic income floor. In the nearer term, even without full UBI, the government could expand the Earned Income Tax Credit or offer wage insurance – for example, if an auto worker laid off due to AI takes a lower-paying job, a wage insurance program could pay part of the difference for a time. Countries like South Korea have already considered taxes on automation, and the idea is gaining traction globally as AI advances.

A Nationwide “AI Transition Workforce Fund” and Retraining Initiative

This program would provide income support, retraining, and job placement services to workers displaced by automation. For example, if a call center in Iowa lays off 500 workers due to an AI system, these individuals could receive unemployment benefits coupled with free tuition for in-demand skills training (e.g. in healthcare, IT, or skilled trades), plus stipends for relocation if needed. Basically, any form of buffer that would soften the blow that lets society a gradual transition into an automated world. Since AGI’s impact will be widespread, the scale of this program must be large and well-funded. It can take inspiration from past programs: Trade Adjustment Assistance served over 5 million workers, and evidence shows participants ultimately earned more over the subsequent decade than those who didn’t get support.

Massively Expand Lifelong Learning & Upskilling Programs

Current education regimes makes it seem like our education careers are over when we reach 22 years old whereas in the age of AI, education and career pivots might last a lifetime. A core long-term solution is education reform to promote lifelong learning. Government, in collaboration with industry, should subsidize and facilitate continuous skills training. This includes making community college or technical training tuition-free for displaced workers, providing grants to employers that invest in training their staff in new skills, and creating certification programs for AI-related competencies that are widely accessible (including online). Employers must be encouraged to hire for skills, not just formal credentials and to consider candidates from nontraditional backgrounds who have retrained. Think DEI programs adjusted for the AI-affected.

Conclusion

As we get closer and closer to AGI, we will likely see all sorts of ways of course correction. Unemployment might rise quickly in some roles, but displaced workers might also find work quickly in others. Gig work will increase. The worst case scenario would be a sudden wave of unemployment which will cause mass unrest, though this should be unlikely to happen. Government, in the mean time, will respond with various measures, from UBI to robot taxes, to the erection of AI Transition Funds. We might even see companies receive tax breaks if they adopt an AI-Affected Program, as an incentive to bring on workers who were affected by AI, but re-trained. What is certain though is inequality will increase. Those who invested in AI from a capital perspective will see a surge in wealth formation and exacerbate the problem of a missing middle class. Despite all this, I remain hopeful, because the changes aren’t over yet. I think the best course of action is for more people to get to know what might unfold in the next couple of years.

So join in. If you think someone might benefit from this article, please send it to them. The more informed our population, the more resilient our society would become. AI can rise all boats, not just the yachts.

--

Hello! I’m Sharon Gai, a keynote speaker on AI and its effects on workers and society. If you were forwarded this article, here is here it came from https://sharongai.substack.com/