We get to choose the relevant technology horizon for our ideal future world vision, and I argue that a post-work future - brought about by ubiquitous, human-level artificial intelligence - is most interesting for us to consider. I believe that it is inevitable by definition, it will have the greatest transformational effect on society, and the mental bandwidth it will create for society is the greatest resource for us to truly imagine our 'ideal future' as a blank canvas.
“When looms weave by themselves, man’s slavery will end.”
Aristotle, 4th century BC
Much of this project is to determine how we want things in the future to look. It is vital therefore to better understand the technology that will enable - and by definition bring about - our envisioned future. 'Human-level' artificial intelligence. This won't be a primer on AI development - if you'd like a more basic understanding of the potential for AI capability, I recommend this primer from MorningBrew, but you should be able to enjoy this article with the definitions and explanations I'll provide.
There are many domains this transformational technology will upend. Our health will be transformed when AI can discover new drugs that are personalised to each recipient. Our communication will be transformed when AI can provide real-time translation between languages (though this one might not be too far away). Our transportation will be transformed when AI enables self-driving cars, ships, and planes. That said, in this article I will argue that the transformation that 'human-level' AI will have on the world of work will have the biggest impact on our 'ideal' future world. It should therefore be a foundational component of our future vision to ground us in a common understanding of what future horizon we are envisioning and designing.
The future of work is leisure
To put it another way, I believe that human-level artificial intelligence will create a society in which no human needs to work. I'll let that sink in, as many people disagree with the idea. My argument is simple. At some time in the future we will have developed an artificial intelligence at a level at least equal with humans. Humans are currently able to do all work. Therefore the artificial intelligence will be capable of all human work. Further, as humans will be capable of performing any 'new jobs' created by this technological development, the AI would be capable of this work too.
Maybe you immediately agree with the logic - if so...well you can probably skim the rest of the article. If instead you immediately thought of any number of arguments against this simplified perspective - read on. This simplified argument relies on particular definitions that I will explain, but given we are selecting the technology horizon to set the stage for our envisioned future, we are at liberty to set these definitions. Also keep in mind that it is in the interests of people in power - namely politicians and business leaders - to downplay the potential of job automation as - as we'll see later in the article - it would require a complete reimagining of the power structures they currently sit on top of.
The two most important definitions: we define 'human-level' AI as capable of all human capabilities, not simply matching human-level intelligence on a narrow set of tasks, and that the cost of the technology will fall to a point it is cheaper to use AI/robotics than humans for any task. What I will call 'ubiquitous, human-level AI'. I'll explain the rationale behind those, and other, arguments, and why it is most interesting for us to consider this particular technology horizon.
The cases against
Some will argue that we will never develop human-level artificial intelligence. They say it is simply beyond our technological capabilities as a species to either reverse-engineer the human brain or create something novel that replicates its capabilities. This argument has traction in that it certainly is an incredibly difficult problem to solve...but we've solved seemingly impossible problems in the past. Just ask an astronaut.
From first principles, we can say that developing human-level intelligence is a physical problem. We have brains and they obey the laws of physics. There are certainly open questions about the nature of consciousness and what makes a human special, but purely to recreate the intelligence capabilities of the brain requires the right atoms in the right order. I believe that with sufficient time we can solve any problem that obeys the laws of physics - that as long as we don't go extinct, we will eventually crack this problem . And it's worth saying that experts believe we will crack it significantly sooner than that.
It could prove true that human-level intelligence is the highest level possible. It seems unlikely as we know human intelligence has increased even over just the last 100 years that it has been measured. However, even if it proves true and we are reaching the limit of human intelligence, the broader argument still holds. We only need ubiquitous human-level artificial intelligence. Nothing more.
What is 'human-level' intelligence? Or, more precisely, how will we define 'human-equivalent' intelligence? If it is defined by completion of a selection of narrow tests, for example passing the Turing Test or scoring in the top 1% of a standardised test, it is fair to argue that there will be certain capabilities the AI cannot perform. AI might be declared 'human-level' in the media without possessing all of the qualities of human intelligence. This is important as this would mean there would be swathes of the labour force that the AI cannot supplant. For example, if the definition of 'human-level' AI focuses solely on knowledge work, the ability of the AI to manipulate robotics could be low, meaning the AI might not be able to perform fine motor skills at a human level.
Again, we're in a position where we can choose a definition that is most interesting to us, and I propose that defining 'human-level' intelligence as encompassing all human capabilities is the most interesting technological frontier for us to consider for our future vision. To pick an intermediate point on the path to full automation would leave our project bogged down in discussions of what jobs would be automated first and last, and how we would design a society with mass, but not full, unemployment. We will likely still cover these questions in our strategy for reaching our vision, as we will almost certainly want good answers to those inevitable and important questions, but for our purposes of unconstrained envisioning, a post-work world gives us both the greatest freedom and most clarity of a shared point in time.
It is easy to think of AI as a computer program, where all we have to do is 'copy/paste' the software and create a second instance, almost instantaneously. It is fair to argue that a more realistic analogy is one of software and hardware, for example to imagine the AI is a human brain.
What would it take to replicate the brain? It could be incredibly expensive. There will likely come a time where we have developed the technology, but it might cost millions or billions of dollars to produce each 'brain'. This better replicates our 'super computer' development path, with each new best-in-the-world model costing over a billion dollars. In my view this is the likely path of development, but is not a very interesting point to settle on for our envisioned future, because it will be less transformational for society. Our world would look very different if we all had a supercomputer in our pockets, much in the way that the world of 2010 and the iPhone 4 looks transformed compared to 1985 and the cray-2 supercomputer that the iPhone 4 was equivalent to. It is more interesting to focus on a point where human-level AI is ubiquitous. This could happen either as the natural path of technology development - with more time a cutting edge technology becomes cheap enough to be broadly accessible - or a plausible scenario where the first human-level intelligence to be developed is used to find a path to replicate itself as cheaply as possible. It would likely be a combination of the two. With 'ubiquitous, human-level' AI, we have a broader canvas on which to paint our ideal future
New job creation
This is the most common counter argument is as follows. History shows new jobs are always created with each technology wave, and we can't be expected to predict what they are ahead of time. 'Social media manager' didn't exist in 2000. Even in 2010 nobody would have predicted that 'Youtube influencer' would ever top the job wish list for kids.
As we've established, our definition means that will have human-level intelligence at a cost below that of a human worker. The 'basic argument' outlined above still stands. 'This time' is different by definition. Every previous technological wave, when viewed through the framework of the intelligence level of the technology, would inevitably create more work as humans would have a comparative advantage in some areas. Our scenario is the tipping point where that is no longer true.
The most interesting horizon
Ultimately, we get to decide what technology horizon is most interesting for us in envisioning our ideal future, and I posit that is a post-work horizon. Assuming a post-work future is possible, why is this the most transformational aspect?
The average worker will spend 90,000 hours of their life working. Average global life expectancy is ~72 years, or 630,000, a third of which is spent sleeping and at least 20 years of which cover pre- or post-working age. Work is the single biggest category of time after sleep. That is to say, the only technology more transformational in terms of affecting how people spend their time would be something that eliminated the human need for sleep. I think it's fair to assume we will develop ubiquitous human-level AI first.
A post-work society would require a transformational redistribution of income, and possibly wealth. Labour is the only source of income for 88% of the global population. It is the the primary source of income for the 98% of the global without sufficient wealth to live on above their country's poverty line, meaning an even smaller percentage who would actually be comfortable on such an income.
A world without work will necessitate a redistribution of income so that everyone has what they need. If not, 88% of the global population would have nothing and the next 10% would live in poverty - an untenable situation for any society. Moreover, there will be little-to-no basis on which to claim some are more 'deserving' than others - the claim that is used to justify any level of 'fair' income inequality. This would suggest there could be a much greater and fairer distribution of resources across the world than we see today. This raises important questions for our 'veil of ignorance' thought experiment to determine a 'fair' future world. What determines inequality in this world? Are there options to earn more than the baseline? Is income inequality purely a function of wealth ownership? How will capital gains be redistributed?
Setting our sights on a post-work future let's us explore these vitally important questions of fairness and equality without constraint.
So what do we do with all this time? That will be the focus of my next 'What Future Thoughts' article, where I'll argue that - despite it being an intimidating prospect - a post-work future is unambiguously a good thing. For now, we just need to accept the rationale behind the technology frontier for our ideal future vision being defined as a world with ubiquitous, human-level AI. It would be a world without the need for work. And nothing would be more transformational.
Please share your thoughts if you have any feedback on this article, or leave a comment below.