Re: AI friendships, as a contrast example, how many people have extremely close friendships with non humans right now? How many people proudly display bumper stickers about superiority of their friendship with small language models named Fido, compared to humans?
Oh sorry I thought you mean "small language models named Fido" literally. I do think there will be people who don't particularly like interacting with other people—these are the people today who like to stay in and watch Netflix by themselves while eating DoorDash. But you don't need everyone going out to eat to have a thriving restaurant industry.
A very nice job of laying out the issues. I wonder how many of these will be permanent however.
#1 (physical bodies) and #2 (flexible + self-repairing) boil down to "we don't have really good robots yet". I agree that this won't happen soon, but... is a decade "soon"? Are we confident that (given the increasing level of motivation / investment) it will take more than a decade? I'm less confident than I used to be...
#3 (humans like to interact with other humans), #4 (humans care what other humans think)... as AIs improve and attitudes shift (possibly helped along by robots with humanoid bodies), this advantage might start to diminish? Some people now spend a lot of time talking with virtual AI companions, and supposedly a mere drawing of a pair of eyes is enough to reduce shoplifting (https://www.atlasobscura.com/articles/why-sticking-a-pair-of-eyeballs-on-a-sign-actually-changes-behavior). I'm certainly not confident these factors will fade entirely, but they probably won't be complete blockers to adoption of AI in the job categories you mention, either?
#5 (humans are scarce), #6 (humans are independent), #7 (humans form relationships with other humans): similar to the previous grouping, how confident should we be that these advantages won't fade over time? For example, "Maybe the AI vendor would promise not to mess with the AI CEO without the approval of the board. But the very fact that it’s possible to modify the software or view its chat histories is going to influence the way other people interact with an AI CEO" – it's possible that AI vendors will have so much reputational capital on the line that an AI CEO might eventually be viewed as *more* trustworthy.
Good points here, though I would question to what degree we would have AIs formally occupying management roles vs managers just making extensive use of AI but still being accountable for final high-level decisions.
I'm sure we'll go through many permutations of this, though I wouldn't attempt to predict the pace of change. At some point I could imagine the human accountability becoming a vestigal formality and ultimately fade away entirely, just as (I think?) we kept human elevator operators around for a while even after the actual operation of the elevator had been automated. And of course this might happen more slowly (or not at all) in some cases, while proceeding more rapidly for lower-stakes roles.
Hi, I agree with most of this. I definitely wouldn't rule out robots eventually being able to do all the "stuff manipulating" jobs. I'm skeptical that will happen in the next 20 years but I certainly can't prove it won't.
I expect jobs based on human scarcity and the value of face-to-face interaction to be around for a long time. A key point here is that it's not necessary for everyone—or even most people—to prefer human-provided services. As long as a healthy minority of people prefer human waiters, doctors, teachers, etc. there will be enough jobs to keep workers busy.
The independence one is the hardest to game out I think. I definitely think humans have an edge over today's AIs because today's AIs aren't even trying to exercise this kind of independence. But if a company set out to create a robot lawyer that people really believed would behave independently, maybe they'd succeed. After all, people trust Google won't read people's gmail messages.
What a great breakdown of the different types of work that we'll continue to need humans for even in an era of powerful AI. I wrote up a forecast a few weeks ago of three archetypal jobs that will probably grow more than the BLS expects over the next decade:
- Jobs where people are needed to be personally accountable for factual accuracy in fields where there will continue to be emerging knowledge (e.g. biologists, legal experts, etc.)
- Jobs where having a human is inherently valuable to the customer (therapists/counselors, and personal service workers among others, similar to the many examples you cite here)
- Jobs that require significant management decisions to be made by a directly accountable individual (managers, whether directly by regulation or downstream of it, and entrepreneurs)
I left out "jobs that require physical work, specific expertise and can perform in unpredictable environments" as I see this as a bit of a question mark over the very long run. As you mention, it will be interesting to see what the ROI of humanoid robots looks like over time.
I also appreciate your breakdown of employment across occupational categories-- if you do more detailed projections of these at some point, would be happy to collaborate!
In 20 years, humans are going to be mostly unemployed. Eventually, there will not be any need to work as the robots will provide everything the remaining humans need. Too many people writing about this subject are vested in finding ways to make our current social and economic organization continue to work into the future, which is misguided.
Science fiction over the past 100 years or so has examined almost every possible future of robots and sentient AI intelligence interacting with meat. Some example everyone here might want to read:
All of the above occur where humanity has made its way to the stars.
But what if we fail to do that? What if we don't make it off this planet? I'm sure some SF writer(s) have examined this possibility but nothing comes to my mind. In this case, I feel certain that the AI Overlord will have to limit human population significantly an deventually, may not see any value in allowing humans to continue to exist.
I wholeheartedly disagree with these predictions. We could see some higher degree of technological unemployment that still radically alters our paradigm, but for the variety of reasons that Tim describes here (and I describe in my comment, and in more detail in my most recent piece), I think that many will continue to find value in some forms of human labor, though the proportion is debateable.
You seem to be citing science fiction as if it were fact. But here is a thought exercise for you: try imaging what the transition to a future without human labor looks like. What are some of the earliest economic indicators of such a shift and how will markets and governments respond?
You may find that the real world is much messier than science fiction!
My hope (by the end of this decade) is to start writing a work of science fiction that describes this messy utopia of the future.
I wholeheartedly disagree with your disagreement! I believe both of you are victims of believing that the state of technology progress forward will be at the same slow glacial pace as in the past.
AI in its current and future forms is vastly accelerating the pace of discovery and implementation. Everywhere you turn are stories of how scientists are making more advancements far faster than previously with the help of AI's.
As to human workers, they are becoming too expensive to employ, while again, the pace of the development for machines that can replace humans is accelerating. 2025 is going to bring some significant replacements of humans by machines. I contend that the replacement is going to be exponential.
I have many hundreds of articles I could reference to support my viewpoint. Here's a few:
=======
Robot that watched surgery videos performs with skill of human doctor
Breakthrough training system utilizing imitation learning opens 'new frontier' in medical robotics
Science fiction has predicted virtually everything that we use today. SF CAN be a learning tool. The 3 recommendations I made above are made to be a learned from, but you'll have to read between the lines.
I've certainly read more than 1000 SF books in my lifetime and that is but a small number of those that exist. If you are going to write an SF book, I hope you have read deeply in the subject because your vision is likely not unique.
In the end, "we will see, said the blind man to his deaf friend".
Indeed, the AI revolution will take a lot longer than what people think, for many nitty-gritty reasons. Folks on Twitter who say in 3 years we'll be obsolete or dead are seriously off the mark.
Longer-term though, many of these human advantages will erode.
I can see why we'll keep teachers around, and have a loving human nanny, personal trainer friend, human artists we can talk to, etc., but 90% of the jobs will likely go away.
I mean 150 years ago half the population worked on farms. Most of those jobs went away, but the result wasn't mass unemployment. I expect a similar effect here—many jobs will eventually go away but there will be plenty of demand for people to do the remaining jobs.
It is true that society adapted before to jobs being eliminated on the large scale.
Machines with more empathy than people will be a big deal though, and plenty of people will prefer relationships with machines than with other people. They would be selfless, understanding, and infinitely focused on us alone. We shall see.
Any data to support this ‘Artificial systems are unlikely to gain these characteristics no matter how rapidly machine learning and robotics progress.’?
That sentence was specifically referring to these characteristics:
"Uniqueness, vulnerability, independence, and connections to other human beings."
I'm not sure data is the right way to think about this, but:
* Uniqueness: we can't mass-manufacture human beings in a factory. Every human being is unique and the growth of the human population is limited by the slow and labor-intensive process of raising children.
* Vulnerability: Human beings are vulnerable to injury and death in a way that computers and robots are not (you can damage a robot but it can always be replaced by new hardware running the same software). I think this is closely related to the "humans care what other humans think" section. Part of the reason people are reluctant to disobey a security guard is that it could lead to a physical confrontation where one or both parties could be injured.
* Independence: Human brains are self-contained and no one can read or modify their contents. This makes them different from computer systems, where software can easily be read or modified.
* Connections to other human beings: humans are hard-wired to care what other human beings think. Unless and until we develop Blade Runner-level androids, humans are going to have an edge here.
This reminds me of a scene in William Gibson's near-future novel "Virtual Light". The main character checks into a swanky hotel and is impressed by the ro it's scurrying around cleaning and performing other menial tasks.
His more worldly companion is unimpressed, though: "That just means they can't afford to pay people to do it."
How long before products have labels proudly proclaiming, "No Bots! No robotic labor was involved in the production of this product," as a seal of high quality?
“Hand made” has been a luxury item selling point for quite a while now. Though I think it is meaningful, in that generally there are fine details in a product that can be polished by skilled hands and are left rough in automated manufacturing processes. Or maybe that’s just my mechanical watch lover’s delusion!
I enjoyed reading and considering your arguments. Still, I think there is unwarranted optimism propping them up. For each example you cite, I can give a personal anecdote about how our capitalist system is more than willing to use AI to devalue each human asset. The incentive is property by yo inequality that benefits those who control AI at the expense of those who do not. It’s sad to say, but those in control of capital just don’t give a damn about the rest of humanity. My work is high touch and fairly AI proof, but access to jobs to which I can bring unique assets has frown alarmingly inhuman over the past two years, and it is getting worse. Society as we know it will collapse before the human golden age you dream of dawns again on Earth.
I don't agree with the statement that "it is hard to give an AI system this kind of independence, and even harder to convince users that they have it."
Humans are always attributing motives to other humans. So many controversies are "A is in the pocket of B" and "C is out to get D." Those are the plots for every soap opera and the majority of other stories.
In contrast, people think of computers as impartial and algorithmic. That can be reinforced with independent reviews and journalism that compares AI models. Because AI can be repeatable, a reviewer can compare the performance of each model in the exact same situation.
I also disagree with the statement that "every AI system runs on a server that’s owned and operated by somebody."
AI does not need to run on someone else's server. Today it requires a beefy computer. But, in the next 5 years LLM's and other agents will be able to run on your phone.
Even if it is running on someone else's server, you can run the same model on servers run by different providers and compare the results. You can do the same thing and compare different AI models.
Finally, there is hardware that makes it impossible for the server's owner or hypervisor to observe the software or OS running on it. The technology is broadly called confidential computing. It is an extension of the concepts used for DRM.
These statements are all true technically. But if you get arrested, and your in a cell and given access to a chatbot lawyer, what would it take for you to trust that it wasn't giving everything to the cops?
> I’d like to see Boston Dynamics make a robot horse that was powered only by hay. Humans like other animals are just very energy efficient.
Actually, humans are only about 25% energy efficient - all the rest is waste heat. Robots dominate us in efficiency by far. Even relatively wasteful things like combustion engines clock in at >35% efficiency. Electric servos and motors are 75-90% efficient.
Maybe "efficiency" is not what I was looking for, but requiring less energy to begin with? Is there a robot that can survive a week on batteries, moving around and doing robot stuff, never charged, in the way that a human can fast for a week? A Big Mac has about the same number of calories as about 3 oz. of gasoline but we seem to do ok? Regardless of efficiency, also, a robot that could just eat hay would be pretty cool
This essay feels like it started with a conclusion (that humans must have permanent advantages over AI) and worked backwards. I don’t see any of the human advantages listed as being insurmountable as the technology rapidly improves. As for the humans-prefer-humans advantage, we’ll see how much of a premium consumers are actually willing to pay for a “made by humans” stamp. 2x? 10x?
Yes, maybe the elite are still going to check into human-cleaned hotels, human-served restaurants, and use human doctors and whatever. The vast majority are going to be using the AI / Zoom versions of these things that are 1000%+ cheaper.
In the next 20 years or so, we will have text, news and social feeds beamed directly into our brains with chips that are inserted either when we are born or early in our lives. We will have permanent contact lens applied over our eyes that will be like a computer screen. We will see the world if we want, or a video feed. Incoming msgs will be in one corner, sports scores in another, etc. Access and payment for everything you do or consume every day will be keyed off of this embedded chip.
You may say, I will not participate, but if you choose not to, then your life will be much more difficult. You'll have to carry physical ID and show it many times daily. It may not be accepted. You'll have to type in access passwords over and over. Very few will put up with the inconvenience.
Compare the cost of a in-person workout to a workout app. A typical workout might cost $20 for a single lesson, compared to $10 to use a workout app for a whole month. So that's at least a 10x difference.
Or think about coffee: a keurig machine can make you coffee for around 50 cents per cup. Or you can go to Starbucks and pay $5. People are paying roughly 10x for the human-made version.
Or music: a Spotify subscription is $12 for a month of music. Or you can go to an in-person concert and pay $50 for a couple of hours of music. At least a 10x premium.
Or tax preparation: Turbotax charges $50 to $100 to do your taxes. A human accountant will charge $500 to $1,500. Again a roughly 10x premium.
Although all your price ratios are true, if you look at the "number of users" at each price point, I think you'll see that more than 10x as many use the 10x cheaper options.
Spotify has 700M+ active monthly listeners. How many people do you think go to a concert every month? I'm betting a lot less than 70M, 7M is almost certainly the right OOM.
How many coffee drinkers make coffee at home vs going to Starbucks? I'm betting way more than 10x. If you step outside the US, it's *certainly* way more than 10x.
How many people use accountants vs self filing or turbo tax? Once again, I'd bet more than a 10x gap.
Now put us in a world where people are making a lot less because lawyering and doctoring and software developing have been automated, and the majority of jobs are crappy service jobs like "personal trainer" or "mcdonalds slave" or whatever - how many people are going to be doing your 10x more expensive stuff then?
Like I said, "human serviced" stuff is going to be an elite thing, the vast majority are going to be using the 1000x cheaper options, just like today, but moreso.
Humans do currently have some advantages over machines. Those are likely to contribute to delaying the machine takeover.
It is funny how people like to talk about the "foreseeable future". That's what: two or three years out at the moment? After that, predictions diverge fairly rapidly about what is going to happen.
So: yes, most humans will probably still have jobs in two or three years. While there are still humans around, they will probably find productive work to do - though the pay may not always be so great. It may not be because humans are useful workers, but rather because unemployed humans get into trouble - and it is possible to avoid that by paying to keep them off the streets.
In the end, it depends on us, average people en masse, whether A.I. will overwhelm us or not. Don't buy it, so that it does not sell. Reject content created with A.I. Choose a human over a robot.
Am I a Luddite, opposing technology? Not at all. If A.I. can predict hurricanes and earthquakes, find proteins for better medicines - employ it! But get out, A.I., once and for good, from my personal life and human interactions.
> Installing scooping robots eliminates scooping jobs, but it simultaneously creates new jobs operating and repairing the robots. And those jobs that are likely to be more interesting (and pay better) than the old scooping jobs.
It’s hard to imagine that this is true in general. The jobs created are ones that computers have a hard time accomplishing, which is not perfectly in line with what humans find interesting (or with what pays a lot). An example here is data labeling, which is just intrinsically difficult for current models to accomplish but at the same time is trivial, cheap, and uninteresting for humans.
Re: developing relationships and friendships, I agree with many of your points, but the way people (especially younger people) interact with people online is drastically different than in-person interactions, and the success of eg Character AI makes me question a lot of the assumptions I had about what the future of interacting with AI will look like
Re: AI friendships, as a contrast example, how many people have extremely close friendships with non humans right now? How many people proudly display bumper stickers about superiority of their friendship with small language models named Fido, compared to humans?
I am honestly not sure. Do you know of people who do this?
I have seen hundreds of bumper stickers with a variation of this , but no real data. https://www.redbubble.com/i/sticker/Hate-people-love-dogs-by-PaletteDesigns/97885148.EJUG5?country_code=US&utm_source=google&utm_medium=organic-shopping&utm_campaign=organic+shopping
Oh sorry I thought you mean "small language models named Fido" literally. I do think there will be people who don't particularly like interacting with other people—these are the people today who like to stay in and watch Netflix by themselves while eating DoorDash. But you don't need everyone going out to eat to have a thriving restaurant industry.
You could argue we have more in common with a dog than a chatbot! They are highly social, playful, communicative mammals after all.
Right, but some of us dog or cat whisperers aside, I think we probably spend more time talking to AIs already
A very nice job of laying out the issues. I wonder how many of these will be permanent however.
#1 (physical bodies) and #2 (flexible + self-repairing) boil down to "we don't have really good robots yet". I agree that this won't happen soon, but... is a decade "soon"? Are we confident that (given the increasing level of motivation / investment) it will take more than a decade? I'm less confident than I used to be...
#3 (humans like to interact with other humans), #4 (humans care what other humans think)... as AIs improve and attitudes shift (possibly helped along by robots with humanoid bodies), this advantage might start to diminish? Some people now spend a lot of time talking with virtual AI companions, and supposedly a mere drawing of a pair of eyes is enough to reduce shoplifting (https://www.atlasobscura.com/articles/why-sticking-a-pair-of-eyeballs-on-a-sign-actually-changes-behavior). I'm certainly not confident these factors will fade entirely, but they probably won't be complete blockers to adoption of AI in the job categories you mention, either?
#5 (humans are scarce), #6 (humans are independent), #7 (humans form relationships with other humans): similar to the previous grouping, how confident should we be that these advantages won't fade over time? For example, "Maybe the AI vendor would promise not to mess with the AI CEO without the approval of the board. But the very fact that it’s possible to modify the software or view its chat histories is going to influence the way other people interact with an AI CEO" – it's possible that AI vendors will have so much reputational capital on the line that an AI CEO might eventually be viewed as *more* trustworthy.
Good points here, though I would question to what degree we would have AIs formally occupying management roles vs managers just making extensive use of AI but still being accountable for final high-level decisions.
I'm sure we'll go through many permutations of this, though I wouldn't attempt to predict the pace of change. At some point I could imagine the human accountability becoming a vestigal formality and ultimately fade away entirely, just as (I think?) we kept human elevator operators around for a while even after the actual operation of the elevator had been automated. And of course this might happen more slowly (or not at all) in some cases, while proceeding more rapidly for lower-stakes roles.
Hi, I agree with most of this. I definitely wouldn't rule out robots eventually being able to do all the "stuff manipulating" jobs. I'm skeptical that will happen in the next 20 years but I certainly can't prove it won't.
I expect jobs based on human scarcity and the value of face-to-face interaction to be around for a long time. A key point here is that it's not necessary for everyone—or even most people—to prefer human-provided services. As long as a healthy minority of people prefer human waiters, doctors, teachers, etc. there will be enough jobs to keep workers busy.
The independence one is the hardest to game out I think. I definitely think humans have an edge over today's AIs because today's AIs aren't even trying to exercise this kind of independence. But if a company set out to create a robot lawyer that people really believed would behave independently, maybe they'd succeed. After all, people trust Google won't read people's gmail messages.
What a great breakdown of the different types of work that we'll continue to need humans for even in an era of powerful AI. I wrote up a forecast a few weeks ago of three archetypal jobs that will probably grow more than the BLS expects over the next decade:
https://www.2120insights.com/p/three-jobs-that-might-grow-more-than
At a high level, these are:
- Jobs where people are needed to be personally accountable for factual accuracy in fields where there will continue to be emerging knowledge (e.g. biologists, legal experts, etc.)
- Jobs where having a human is inherently valuable to the customer (therapists/counselors, and personal service workers among others, similar to the many examples you cite here)
- Jobs that require significant management decisions to be made by a directly accountable individual (managers, whether directly by regulation or downstream of it, and entrepreneurs)
I left out "jobs that require physical work, specific expertise and can perform in unpredictable environments" as I see this as a bit of a question mark over the very long run. As you mention, it will be interesting to see what the ROI of humanoid robots looks like over time.
I also appreciate your breakdown of employment across occupational categories-- if you do more detailed projections of these at some point, would be happy to collaborate!
That is cool! I'm gonna subscribe to your newsletter.
Thank you! I really admire your work, this means a lot to me 😄
Working on a more fleshed-out version of this comment which I hope to publish tomorrow.
In 20 years, humans are going to be mostly unemployed. Eventually, there will not be any need to work as the robots will provide everything the remaining humans need. Too many people writing about this subject are vested in finding ways to make our current social and economic organization continue to work into the future, which is misguided.
Science fiction over the past 100 years or so has examined almost every possible future of robots and sentient AI intelligence interacting with meat. Some example everyone here might want to read:
1. Iain M. Banks' Culture series - https://www.perplexity.ai/search/how-would-you-summarize-the-ro-4yb.2eAOSsKlCZNikh0cqA#1
2. Neal Asher's Polity Universe - https://www.perplexity.ai/search/how-would-you-summarize-the-ro-4yb.2eAOSsKlCZNikh0cqA#0
3. ## Jack Williamson's Humanoid Novels - https://www.perplexity.ai/search/summarize-and-explain-jack-wil-VSaHxXkESFSvKg0Iq6Gu1g#0
All of the above occur where humanity has made its way to the stars.
But what if we fail to do that? What if we don't make it off this planet? I'm sure some SF writer(s) have examined this possibility but nothing comes to my mind. In this case, I feel certain that the AI Overlord will have to limit human population significantly an deventually, may not see any value in allowing humans to continue to exist.
I wholeheartedly disagree with these predictions. We could see some higher degree of technological unemployment that still radically alters our paradigm, but for the variety of reasons that Tim describes here (and I describe in my comment, and in more detail in my most recent piece), I think that many will continue to find value in some forms of human labor, though the proportion is debateable.
You seem to be citing science fiction as if it were fact. But here is a thought exercise for you: try imaging what the transition to a future without human labor looks like. What are some of the earliest economic indicators of such a shift and how will markets and governments respond?
You may find that the real world is much messier than science fiction!
My hope (by the end of this decade) is to start writing a work of science fiction that describes this messy utopia of the future.
I wholeheartedly disagree with your disagreement! I believe both of you are victims of believing that the state of technology progress forward will be at the same slow glacial pace as in the past.
AI in its current and future forms is vastly accelerating the pace of discovery and implementation. Everywhere you turn are stories of how scientists are making more advancements far faster than previously with the help of AI's.
As to human workers, they are becoming too expensive to employ, while again, the pace of the development for machines that can replace humans is accelerating. 2025 is going to bring some significant replacements of humans by machines. I contend that the replacement is going to be exponential.
I have many hundreds of articles I could reference to support my viewpoint. Here's a few:
=======
Robot that watched surgery videos performs with skill of human doctor
Breakthrough training system utilizing imitation learning opens 'new frontier' in medical robotics
Jill Rosen
10 Nov 2024
https://hub.jhu.edu/2024/11/11/surgery-robots-trained-with-videos/
=======
=======
Where’s My Robot?
Here’s how we could finally build humanoid robots that do all our domestic chores
By Erico Guizzo & Randi Klett
11 Nov 2024
https://spectrum.ieee.org/ai-robots
=======
=======
Taking stock
Machines of Loving Grace1
How AI Could Transform the World for the Better
Dario Amodei
October 2024
https://darioamodei.com/machines-of-loving-grace
=======
Science fiction has predicted virtually everything that we use today. SF CAN be a learning tool. The 3 recommendations I made above are made to be a learned from, but you'll have to read between the lines.
I've certainly read more than 1000 SF books in my lifetime and that is but a small number of those that exist. If you are going to write an SF book, I hope you have read deeply in the subject because your vision is likely not unique.
In the end, "we will see, said the blind man to his deaf friend".
Indeed, the AI revolution will take a lot longer than what people think, for many nitty-gritty reasons. Folks on Twitter who say in 3 years we'll be obsolete or dead are seriously off the mark.
Longer-term though, many of these human advantages will erode.
I can see why we'll keep teachers around, and have a loving human nanny, personal trainer friend, human artists we can talk to, etc., but 90% of the jobs will likely go away.
Then, one can make the argument that an AI being will love you and understand you a lot more than a human ever could or be willing to.
I mean 150 years ago half the population worked on farms. Most of those jobs went away, but the result wasn't mass unemployment. I expect a similar effect here—many jobs will eventually go away but there will be plenty of demand for people to do the remaining jobs.
It is true that society adapted before to jobs being eliminated on the large scale.
Machines with more empathy than people will be a big deal though, and plenty of people will prefer relationships with machines than with other people. They would be selfless, understanding, and infinitely focused on us alone. We shall see.
"many jobs will eventually go away but there will be plenty of demand for people to do the remaining jobs."
---
WHy? That makes no sense. There are going to be many more people an d many less jobs for humans, so how will all the humans have work?
The truth is that ALL jobs will eventually be done by computer intelligence.
Think about an exponential curve. That is how AI intelligence will grow.
Any data to support this ‘Artificial systems are unlikely to gain these characteristics no matter how rapidly machine learning and robotics progress.’?
That sentence was specifically referring to these characteristics:
"Uniqueness, vulnerability, independence, and connections to other human beings."
I'm not sure data is the right way to think about this, but:
* Uniqueness: we can't mass-manufacture human beings in a factory. Every human being is unique and the growth of the human population is limited by the slow and labor-intensive process of raising children.
* Vulnerability: Human beings are vulnerable to injury and death in a way that computers and robots are not (you can damage a robot but it can always be replaced by new hardware running the same software). I think this is closely related to the "humans care what other humans think" section. Part of the reason people are reluctant to disobey a security guard is that it could lead to a physical confrontation where one or both parties could be injured.
* Independence: Human brains are self-contained and no one can read or modify their contents. This makes them different from computer systems, where software can easily be read or modified.
* Connections to other human beings: humans are hard-wired to care what other human beings think. Unless and until we develop Blade Runner-level androids, humans are going to have an edge here.
This reminds me of a scene in William Gibson's near-future novel "Virtual Light". The main character checks into a swanky hotel and is impressed by the ro it's scurrying around cleaning and performing other menial tasks.
His more worldly companion is unimpressed, though: "That just means they can't afford to pay people to do it."
How long before products have labels proudly proclaiming, "No Bots! No robotic labor was involved in the production of this product," as a seal of high quality?
“Hand made” has been a luxury item selling point for quite a while now. Though I think it is meaningful, in that generally there are fine details in a product that can be polished by skilled hands and are left rough in automated manufacturing processes. Or maybe that’s just my mechanical watch lover’s delusion!
NOTHING human done will be better than what a robot can do in the not distant future. Sorry.
I enjoyed reading and considering your arguments. Still, I think there is unwarranted optimism propping them up. For each example you cite, I can give a personal anecdote about how our capitalist system is more than willing to use AI to devalue each human asset. The incentive is property by yo inequality that benefits those who control AI at the expense of those who do not. It’s sad to say, but those in control of capital just don’t give a damn about the rest of humanity. My work is high touch and fairly AI proof, but access to jobs to which I can bring unique assets has frown alarmingly inhuman over the past two years, and it is getting worse. Society as we know it will collapse before the human golden age you dream of dawns again on Earth.
I don't agree with the statement that "it is hard to give an AI system this kind of independence, and even harder to convince users that they have it."
Humans are always attributing motives to other humans. So many controversies are "A is in the pocket of B" and "C is out to get D." Those are the plots for every soap opera and the majority of other stories.
In contrast, people think of computers as impartial and algorithmic. That can be reinforced with independent reviews and journalism that compares AI models. Because AI can be repeatable, a reviewer can compare the performance of each model in the exact same situation.
I also disagree with the statement that "every AI system runs on a server that’s owned and operated by somebody."
AI does not need to run on someone else's server. Today it requires a beefy computer. But, in the next 5 years LLM's and other agents will be able to run on your phone.
Even if it is running on someone else's server, you can run the same model on servers run by different providers and compare the results. You can do the same thing and compare different AI models.
Finally, there is hardware that makes it impossible for the server's owner or hypervisor to observe the software or OS running on it. The technology is broadly called confidential computing. It is an extension of the concepts used for DRM.
These statements are all true technically. But if you get arrested, and your in a cell and given access to a chatbot lawyer, what would it take for you to trust that it wasn't giving everything to the cops?
I’d like to see Boston Dynamics make a robot horse that was powered only by hay. Humans like other animals are just very energy efficient.
> I’d like to see Boston Dynamics make a robot horse that was powered only by hay. Humans like other animals are just very energy efficient.
Actually, humans are only about 25% energy efficient - all the rest is waste heat. Robots dominate us in efficiency by far. Even relatively wasteful things like combustion engines clock in at >35% efficiency. Electric servos and motors are 75-90% efficient.
Maybe "efficiency" is not what I was looking for, but requiring less energy to begin with? Is there a robot that can survive a week on batteries, moving around and doing robot stuff, never charged, in the way that a human can fast for a week? A Big Mac has about the same number of calories as about 3 oz. of gasoline but we seem to do ok? Regardless of efficiency, also, a robot that could just eat hay would be pretty cool
This essay feels like it started with a conclusion (that humans must have permanent advantages over AI) and worked backwards. I don’t see any of the human advantages listed as being insurmountable as the technology rapidly improves. As for the humans-prefer-humans advantage, we’ll see how much of a premium consumers are actually willing to pay for a “made by humans” stamp. 2x? 10x?
Yup, this article is obvious cope.
Yes, maybe the elite are still going to check into human-cleaned hotels, human-served restaurants, and use human doctors and whatever. The vast majority are going to be using the AI / Zoom versions of these things that are 1000%+ cheaper.
In the next 20 years or so, we will have text, news and social feeds beamed directly into our brains with chips that are inserted either when we are born or early in our lives. We will have permanent contact lens applied over our eyes that will be like a computer screen. We will see the world if we want, or a video feed. Incoming msgs will be in one corner, sports scores in another, etc. Access and payment for everything you do or consume every day will be keyed off of this embedded chip.
You may say, I will not participate, but if you choose not to, then your life will be much more difficult. You'll have to carry physical ID and show it many times daily. It may not be accepted. You'll have to type in access passwords over and over. Very few will put up with the inconvenience.
Compare the cost of a in-person workout to a workout app. A typical workout might cost $20 for a single lesson, compared to $10 to use a workout app for a whole month. So that's at least a 10x difference.
Or think about coffee: a keurig machine can make you coffee for around 50 cents per cup. Or you can go to Starbucks and pay $5. People are paying roughly 10x for the human-made version.
Or music: a Spotify subscription is $12 for a month of music. Or you can go to an in-person concert and pay $50 for a couple of hours of music. At least a 10x premium.
Or tax preparation: Turbotax charges $50 to $100 to do your taxes. A human accountant will charge $500 to $1,500. Again a roughly 10x premium.
Although all your price ratios are true, if you look at the "number of users" at each price point, I think you'll see that more than 10x as many use the 10x cheaper options.
Spotify has 700M+ active monthly listeners. How many people do you think go to a concert every month? I'm betting a lot less than 70M, 7M is almost certainly the right OOM.
How many coffee drinkers make coffee at home vs going to Starbucks? I'm betting way more than 10x. If you step outside the US, it's *certainly* way more than 10x.
How many people use accountants vs self filing or turbo tax? Once again, I'd bet more than a 10x gap.
Now put us in a world where people are making a lot less because lawyering and doctoring and software developing have been automated, and the majority of jobs are crappy service jobs like "personal trainer" or "mcdonalds slave" or whatever - how many people are going to be doing your 10x more expensive stuff then?
Like I said, "human serviced" stuff is going to be an elite thing, the vast majority are going to be using the 1000x cheaper options, just like today, but moreso.
Great article. I would add:
Humans are legal persons, bound and privileged by law.
Humans are accountable, trustworthy, have reputations, and are able to do costly signaling. (Overlaps with your 4 & 7)
Humans do currently have some advantages over machines. Those are likely to contribute to delaying the machine takeover.
It is funny how people like to talk about the "foreseeable future". That's what: two or three years out at the moment? After that, predictions diverge fairly rapidly about what is going to happen.
So: yes, most humans will probably still have jobs in two or three years. While there are still humans around, they will probably find productive work to do - though the pay may not always be so great. It may not be because humans are useful workers, but rather because unemployed humans get into trouble - and it is possible to avoid that by paying to keep them off the streets.
In the end, it depends on us, average people en masse, whether A.I. will overwhelm us or not. Don't buy it, so that it does not sell. Reject content created with A.I. Choose a human over a robot.
Am I a Luddite, opposing technology? Not at all. If A.I. can predict hurricanes and earthquakes, find proteins for better medicines - employ it! But get out, A.I., once and for good, from my personal life and human interactions.
Nice article. I do wonder about a couple points:
> Installing scooping robots eliminates scooping jobs, but it simultaneously creates new jobs operating and repairing the robots. And those jobs that are likely to be more interesting (and pay better) than the old scooping jobs.
It’s hard to imagine that this is true in general. The jobs created are ones that computers have a hard time accomplishing, which is not perfectly in line with what humans find interesting (or with what pays a lot). An example here is data labeling, which is just intrinsically difficult for current models to accomplish but at the same time is trivial, cheap, and uninteresting for humans.
Re: developing relationships and friendships, I agree with many of your points, but the way people (especially younger people) interact with people online is drastically different than in-person interactions, and the success of eg Character AI makes me question a lot of the assumptions I had about what the future of interacting with AI will look like
Hip hip hooray for humans!