Elon Musk has said several times recently that desalinisation is “absurdly cheap”.1 

This was surprising to me. When I was younger I was taught the mantra that desalinisation “uses lots of energy and is really expensive”. And to be honest, I hadn’t thought about it much since then.

Cheap desalinisation would be very welcome indeed; not just to make societies more resilient to periods of water scarcity, but to relieve pressure on groundwater resources, and provide clean water to those that are already lacking.

I thought it was time to update my perspective and see what the numbers had to say about the prospects for desalinisation.

A few caveats to start:

• I’ll focus on energy use and costs here. I’m not analyzing the amount of brine water that’s generated and how to manage that in an environmentally-sustainable way. Not because that’s not important, but because this post would '“absurdly” long.

• To get a sense of the numbers, I’ll make pretty unrealistic tests like “every household in the US needs to get its domestic water from desalinisation”. Or “everyone in the world needs desalinated drinking water”. These scenarios are extremely unlikely; they’re just useful “what ifs” to understand scale.

There are two key types of desalinisation technology: thermal desalinisation, and reverse osmosis.

Thermal desalinisation uses heat to evaporate the water (separating it from the salts and impurities) before condensing it back into a liquid. The salty brine water is left behind.

In reverse osmosis, pressure is applied to push water through a semi-permeable membrane, which removes the salt and impurities to leave freshwater on the other side. The figure below shows the “natural” process of osmosis where fresh and saltwater would want to mix to make the concentration of salts equal on both sides of the membrane.

In reverse osmosis (shown on the right), pressure is applied to overcome and reverse this.

Both technologies work, but thermal desalinisation is much more energy-intensive (and therefore expensive). It uses around three to five times as much energy per cubic metre of water.2

Most desalinisation plants today use reverse osmosis, and this is the process I’ll focus on in this article.

Reverse osmosis technologies have become much more energy-efficient over the last few decades. You can see this in the chart below (although it ends in 2008).

In the 1970s, it would take around 20 kilowatt-hours (kWh) to desalinate one cubic metre of water.

Today that figure is around 2.5 to 3.5 kWh. It’s often suggested that the theoretical minimum for reverse osmosis is around 1 kWh. Think of this as the floor and the absolute best that can be achieved. No one is there yet, and seems unlikely that we’d get the process as efficient as this.

If we wanted to get energy use lower than 1 kWh per m³ we’d need to innovate and develop a different process.

For all of the calculations that follow, I’m going to assume that reverse osmosis uses 3.5 kWh per m³. I’m being quite (deliberately) harsh here because some desalinisation plants use less than this.

What would happen to electricity demand if countries were to get all of their domestic water from desalinisation?

The average American household uses around 1135 litres of water per day. Getting all of this water from desalinisation would increase its domestic electricity demand by around 13%.3

The average UK household uses less water: 349 litres of water per day. But it also uses less electricity, so desalinisation would increase its electricity demand by a similar share: 15%.

Let’s put this electricity demand into context for the United States. To do this, I’ll use data from the Energy Information Administration (EIA)’s Residential Energy Consumption Survey. I looked at what American households use electricity for in a previous article.

You can see how the demand for desalinisation — 1450 kWh — stacks up against other household power users in the chart below. Note that this is based on the average across the US, and is the electricity for households that have that given technology or end use. So, it shows the average electricity use of a family with an electric vehicle.4

As you can see, it would use a decent — but not crazy — amount of electricity. In the US, it’s on a similar level to many other important technologies like dehumidification; and much less than the energy used to heat water, or to heat or cool their home.

But for households in poorer countries, 1450 kWh is a lot of electricity.5 More than an entire family in lower-income countries would use in total. Granted, households in the US use a lot more water than those in more water and energy-constrained settings would.

In the chart below I’ve compared total electricity demand per capita for all uses in many lower-income countries (and those on slightly higher incomes, such as India) to desalinisation of US levels of water consumption, and the WHO’s minimum guideline of 50 litres per person per day.

Providing 50 litres per person per day — the WHO minimum — would need 64 kWh of electricity. That’s about equal to what the average Malawian uses for everything today. Even for those that use more than 64 kWh, electricity demand for desalinisation would be equal to a 50% increase in demand, or more. Achieving American levels of water usage would be unattainable.

To me, the main message of this chart is not that desalinisation is incredibly energy-intensive. It’s that many people in the world still live in dire energy poverty.

Only a fraction of the domestic water we use is for drinking. Guidelines often recommend drinking 2 to 2.5 litres of water per day. There is some nuance to this number, which I’ll leave in the footnote, but let’s take that as a rough estimate.6 We’ll be generous and assume that there are some spillages, so everyone needs 3 litres per day.

If we were to use desalinisation to provide this water for everyone — 8 billion of us — how much electricity would we need?

31 TWh of electricity per year.7 The world currently produces around 30,000 TWh of electricity per year, so our demand would increase by 0.1%. Not a lot!

Of course, not everyone in the world would need drinking water from desalinisation, but you can then quickly slice the numbers. Imagine an incredibly catastrophic and extreme drought left one-third of the world without water: around 10 TWh would be needed per year to provide the most basic drinking water supplies.

The WHO has a guideline that people should have a minimum of 50 litres per day to meet all of their domestic needs, which is not just drinking water but also cleaning, showering etc. As we saw earlier, people in richer countries use far more than this: in the UK, three times as much, and in the US, six times.

If we were to provide the 50-litre minimum for 8 billion people, we’d need 511 TWh of electricity. That’s 1.7% of what the world currently produces.

See the chart below.

Here I’m focusing on the unit cost. Large-scale desalinisation plants would also need significant capital investment, which will probably be the biggest barrier in many countries. It’s not dissimilar to some renewable projects: the cost per unit of electricity is often very low, but people or communities still need to find the capital up-front.

I’ve tried to find a range of projects across the world that would give insight into how much desalinisation costs, and how this might vary. Most quoted figures are in the range of $1 to $2.50 per cubic metre. I couldn’t find many estimates higher than this. 

There are also some more optimistic figures: the Sorek B plant in Israel, for example, is contracted to produce water at $0.41 per cubic metre. A study looking at 107 desalinisation plants found a minimum cost of $0.27.8

One way to stress-test these figures is to estimate how much the electricity alone would cost. Electricity in the US in 2023 cost around $0.13 per kWh (industry is subsidised, so it sells at around $0.09 per kWh). If it takes 3.5 kWh to produce one cubic metre of water, then it costs $0.45 per m³. Electricity prices in states like California are much higher — roughly twice the national average — so there it could cost $0.90 per m³.

Energy is just one part of the full cost; estimates are around one-third. Multiply by three, and we get around $1.50 per m³. In states like California, the energy cost might be $0.50 higher, bringing the total to $2.

Let’s put these numbers into context. How much would desalinated water cost per person per year?

If the average person in the United States uses 310 litres of water per day for domestic uses, it would cost them $0.42 per day.9 Or $154 per year.

In the UK, it would be similar: $0.43 per day, and $159 per year. British electricity is more expensive, but we also use less water so the costs balance out.

To get the WHO’s “minimum” domestic supply of 50 litres per day would cost around $0.11 per day and $38 per year.10

Here’s the surprising figure. Producing enough drinking water for someone — assuming 3 litres per day — costs just $2.30 for the entire year. That’s less than the cost of a single bottle of water in many countries.

This was far cheaper than I’d have guessed, and in a water crisis I would agree that this is “absurdly cheap”. More expensive than many of us get from the tap today, but pretty cheap for an essential resource.

Again it’s worth putting the $38 that would be needed to meet the WHO’s minimum guidelines of 50 litres per day, into context for the world’s poorest.

Some people still live on as little as $2 per day. Getting 50 litres of domestic water per day from desalinisation would eat up more than half of a month’s income every year.11

Farming is where the costs of desalinisation start to get unfeasible.12

Most of the world’s freshwater withdrawals — around 70% — are used for agriculture. In some countries — particularly in the tropics and subtropics — this can be more than 90%. Making sure that people have drinking water and basic supplies at home is crucial during periods of water scarcity. But when suffer from drought, it’s usually because crops fail and food supplies run out.

Let’s look at how much freshwater is withdrawn from aquifers for agriculture across different countries, and how much electricity would be needed to replace that with desalinisation.13 Note that this is not all of the water used in agriculture, since a lot of it is rainfed.

As you can see, desalinisation to replace just a subset of water used for agriculture would increase electricity demand by around 50% to two-thirds for many countries. For some — often the countries where water stress is already high — total electricity generation would need to more than double. Current desalinisation technologies are not going to be the agricultural safety net for large-scale national droughts.

But the bigger constraint for desalinisation in farming is cost. Using desalinated water would either bankrupt the farmer or make food much more expensive.

Let’s take an example of wheat. It takes around 650 litres to produce one kilogram of wheat.14 It would take around 2.3 kWh of electricity to produce the water for this. Take electricity costs in the UK and the water costs $0.66.

The current market price for wheat is around $0.22 per kilogram. The water would be three times as expensive as the current final product.

For maize production in the US, I estimate that desalinated water would cost $0.20 per kilogram.15 In some states with higher electricity prices, it could be double. The market price for corn is around $0.20.

The economics for staple crops that are cheap and have low margins just don’t work.

It might just be economic in fringe cases for high-value crops, grown in conditions that are much more water-efficient — such as indoor farming — but we’re still pretty far from having solutions that could make a big dent in meeting water demand for staple crops.

Support Sustainability by numbers

Using data and research to understand what really makes a difference

The principles of thermodynamics are cornerstones of our understanding of physics. But they were discovered in the era of steam-driven technology, long before anyone dreamed of quantum mechanics. In this episode, the theoretical physicist Nicole Yunger Halpern (opens a new tab) talks to host Steven Strogatz (opens a new tab) about how physicists today are reinterpreting concepts such as work, energy and information for a quantum world.

Listen on Apple Podcasts (opens a new tab), Spotify (opens a new tab),  TuneIn (opens a new tab) or your favorite podcasting app, or you can stream it from Quanta.

Transcript

[Theme plays]

STEVEN STROGATZ: In the mid-1800s, engineers were grappling with questions at the forefront of the Industrial Revolution: how to convert steam into mechanical work, translate rushing streams into electrical energy or pump water out of mines. Their inquiries and observations built the groundwork of a new science: thermodynamics.

By the early 1900s, we had not one, but three laws of thermodynamics. These laws have since become ubiquitous and proven fundamental to our understanding of physics in everyday life. But as our knowledge of the physical world continues to grow, the limits of these old mechanical notions become more apparent — especially as we approach the quantum scale.

I’m Steve Strogatz, and this is “The Joy of Why,” a podcast from Quanta Magazine where I take turns at the mic with my cohost, Janna Levin (opens a new tab), exploring the biggest unanswered questions in math and science today.

In this episode, we’re going to ask, what do concepts like work and heat mean on an atomic or even subatomic level? And can our laws of thermodynamics be reinterpreted in quantum terms?

[Theme fades out]

We’re joined by Nicole Yunger Halpern. She is a theoretical physicist at the National Institute of Standards and Technology and an adjunct assistant professor at the University of Maryland. Her research lies at the intersection of quantum physics, information processing and thermodynamics. She’s also the author of an award-winning book, Quantum Steampunk: The Physics of Yesterday’s Tomorrow (opens a new tab).

Nicole, it’s a great pleasure to have you with us on “The Joy of Why.”

NICOLE YUNGER HALPERN: It’s a delight to be here. Thanks for having me.

STROGATZ: Well, thank you for joining us. I really am excited about this. I was just asking my wife on the drive over to the studio about the word “steampunk.” I have to admit that I’m not familiar with this. She mentioned to me that even jewelry can be done in steampunk style.

YUNGER HALPERN: Steampunk comes up in costumes and conventions and jewelry and film, in books and short stories all over the place. It’s combines the aesthetic of the 1800s. So the Victorian-era people in waistcoats and petticoats and top hats, and also the American wild, wild west and Beijing, Japan, as well as futuristic technologies.

STROGATZ: Huh, excellent. Well, it’s a very intriguing title, Quantum Steampunk. I hope our listeners will check out the book.

So let’s start talking about the Victorian era and the thermodynamics ideas that grew out of that. We now think of it as classical thermodynamics, and I mentioned the laws of thermodynamics. There are so many things to unpack — the three laws, ideas like work, heat, energy, entropy, efficiency. Take us through some of those ideas, at least, to start with.

YUNGER HALPERN: Thermodynamics is something that we all have a sense of, but maybe we kind of take it for granted. And maybe that’s because thermodynamics is so general. It’s the study of energy, period. The forms that energy can be in and the transformations amongst those forms.

Energy can be transmitted in the form of heat and in the form of work. Work is coordinated, directed energy that can be directly harnessed to do something useful, like power a factory or charge a battery.

Heat is random, uncoordinated energy. It’s the energy of particles jiggling about randomly. Heat engines turn this random heat into coordinated, useful work.

And heat and work feature in the laws of thermodynamics. The number has been growing a bit, depending on whom you ask. There might be three, there might be four, there might be five.

The zeroth law of thermodynamics was actually developed after the first three, but people thought it was so important that it should be given precedence. And so it tells us that there are thermometers. Suppose that you have a cup of tea and I have a cup of tea. We want to be able to compare their temperatures. How can we do that? We can do it using a thermometer.

The first law tells us that the total amount of energy in the world remains constant. The second law tells us that the entropy of a closed, isolated system remains constant or increases only, at least on average. And the third law tells us that you can’t actually cool any system down to the lowest conceivable temperature, absolute zero — zero Kelvin — in any finite number of steps. So that is a very brief history of thermodynamics.

STROGATZ: Excellent. Great. That is a good summary. The second law always feels to me like the really deep one.

YUNGER HALPERN: Yes.

STROGATZ: Right? I mean, the concept of entropy, it’s often phrased as some measure of disorder in a system. Do you want to talk to us about entropy just for a minute?

YUNGER HALPERN: Sure, I think of entropy as a measure of uncertainty. And all sorts of things can have entropies associated with them. For instance, the weather on any given day in the Boston area is extremely random. It could be sunny or rainy or cloudy or snowy. And suppose that we learn on some given day what the weather is. So we’ve learned some amount of information and the amount of information we learn can be seen as an entropic quantity. And then suppose that we average this amount of information that we learn over all of the days. That’s another entropic quantity, a pretty common one.

And we can translate the story into thermodynamics by saying we have physicists’ favorite thermodynamic system — a classical gas in a box. And suppose that we know only large-scale properties of a gas, like the total number of particles and the volume. There are lots of different microstates or configurations associated that are consistent with this large-scale macro state. By a microstate, I mean a list of all of the particles’ positions, all of their momentum, and maybe some other properties, depending on what sorts of particles we have.

So, if we know just these large-scale properties, how ignorant are we about the microstates? And that’s essentially the thermodynamic entropy.

STROGATZ: It’s amazing, this idea of the gas in a box. Because it’s true, that is the universal example. And I’m actually, at this very moment, sitting in a box called a studio. The door is closed. There is gas in here. It’s the air around me.

YUNGER HALPERN: I’m very glad.

STROGATZ: I mean, as far as I can tell. Now, I suppose it’s conceivable that all the air molecules could spontaneously go into the corner, and you might hear me gasping. But that would be a very rare event.

Would that be a state, if all the molecules were in the corner? That would be, what? Very low entropy, I suppose?

YUNGER HALPERN: Right, so there’s a lot of debate, especially in the philosophical community, about how to define thermodynamic entropy. But the way that we’re often taught to reason in statistical physics classes, we would say tend to say, yes, the state in which the particles are all clumped together in the corner of the box is indeed a low-entropy state.

STROGATZ: There’s a concept that comes up a lot: equilibrium. Can you remind us, what does that mean? Like when we speak of a system being at thermodynamic equilibrium, what is that? Why does it matter?

YUNGER HALPERN: Equilibrium is a rather quiet state of a system. It’s a state in which large-scale properties like the total temperature and total volume remain approximately constant over time, and there’s no net flow of anything like heat or particles into or out of the system.

So suppose that we have had a hot cup of tea. We have let it sit on the counter for a long time. It has come to have the same temperature as the rest of the room, and a little bit of the water has evaporated away. At this point, the tea is at thermal equilibrium with its environment.

STROGATZ: And so it always seems kind of like an artificial thing that happens in a chemistry lab or in this famous tea cooling off on the kitchen counter. Whereas in real life, you know, I’m eating food all day long, it seems — just had a cookie before coming to the studio. I can’t relate to thermodynamic equilibrium very well. Is that fair to say? In our everyday life, what things are at equilibrium and what things are not?

YUNGER HALPERN: A great deal in our lives, including life itself, as you point out, is far out of equilibrium.

Organisms keep themselves far out of equilibrium by doing just what you said, by eating so that they consume energy in a well-organized form and expel it in a very highly entropic form. So you radiate lots of heat. This helps keep us far out of equilibrium.

If you have run a bath and let it sit around too long, you might have experienced equilibrium unpleasantly. Or made a cup of coffee and gotten distracted by your work, so that you end up having to drink cold coffee, you might have experienced equilibrium.

STROGATZ: I see. So it does seem like a sort of final state. It’s like after everything settles down. There’s no drive for anything to change anymore, it sounds like.

YUNGER HALPERN: Exactly, there is no drive.

STROGATZ: So when you mentioned the different laws of thermodynamics, what kinds of systems do the laws apply to? What other caveats do we need to make about those systems in order for the laws to apply?

YUNGER HALPERN: Well, the laws of thermodynamics were originally formulated by people who had in mind large classical systems. They didn’t necessarily think of these systems as consisting of many, many particles. The theory of atomism was not entirely accepted by the Victorian era. But they were thinking of systems that, at least now we will all acknowledge, consist of lots and lots of particles.

Around the turn of the 20th century, people discovered Brownian motion, which is random jiggling of particles that’s observable with a microscope, and it led people to accept very broadly that, in fact, materials do consist of very small particles. They jiggle around randomly, and occasional jiggling in the wrong direction led to some minor changes in at least the second law of thermodynamics.

But what’s really surprising to me is that the laws of thermodynamics seem to be going strong, even though we’ve learned a great deal since even the turn of the 20th century about small systems, biological systems, chemical systems and even quantum systems.

STROGATZ: Well, so we’ve been talking so far from the point of view of these particles that you keep mentioning — the atoms or molecules — systems made up of enormous numbers.

But now we’re going to start to get into the quantum aspects of thermodynamics with you. Is the main novelty conceptually the idea that we have just very few particles now? Or is it that we’re using quantum ideas instead of classical ideas?

YUNGER HALPERN: I see quantum thermodynamics as involving the extension of conventional thermodynamics to small systems, quantum systems and far-from-equilibrium systems.

Although not all these categories have been addressed only by quantum thermodynamicists. For instance, there was a lot of really amazing work done in far-from-equilibrium statistical mechanics during the 20th century in the field of non-equilibrium statistical mechanics, which is kind of adjacent to and admired by and friends with quantum thermodynamics.

STROGATZ: Interesting. So if I heard you right, you said there are going to be three kinds of things to think about: far-from-equilibrium, quantum and small numbers. All three of those we could think of as at the edge of what was traditional thermodynamics. By traditional, I mean like the subject that [Josiah Willard] Gibbs (opens a new tab) and [James Clerk] Maxwell (opens a new tab) and people like that helped develop in the late 1800s. They didn’t have the math or the physical concepts to really handle small systems, far-from-equilibrium systems, or they wouldn’t have even known about quantum systems at that point.

YUNGER HALPERN: That’s a good way of putting it.

STROGATZ: I’m surprised by your answer. It’s interesting. I thought you would just say quantum, but you’ll deal with large numbers. But so you can allow for small systems, too.

YUNGER HALPERN: One of the reasons is, suppose that we want to address quantum thermodynamics. That is complicated to do, so it can be simpler to make a model that is amenable to small systems, solve some problems using this model, and after you’ve solved those problems, you know, add in some more quantum features — like coherences, which sometimes I describe as the wavelike nature of quantum particles.

And this is, in fact, what happened a number of years ago in the intersection of quantum thermodynamics and quantum information theory. Some colleagues of mine created a model for certain systems. They wanted for this model to describe quantum thermodynamic systems. But just solving the classical version of the small-scale problem was complicated enough. After that problem was solved, then people could make progress on the really quantum features. So the classical small-scale system problem was in service of the quantum thermodynamics.

STROGATZ: Hmm, alright, so we have a lot to discuss here. [laughing] I’m a bit daunted because it’s conceptually very rich and feels to me very new.

YUNGER HALPERN: The recent wave of quantum thermodynamics that has become widely accepted as a subfield is very new. There’s been this trend over the past 10 to 15 years in which quantum thermodynamics has grown a great deal.

Quantum thermodynamics first started being thought about during the 1930s. A quantum engine was proposed in the 1950s and ’60s. There was work in the ’80s. But these pieces of work were not always accepted by the wider community, and quantum thermodynamics itself was sometimes called an oxymoron, because thermodynamics was developed for large classical systems, so people just couldn’t understand what it could possibly have to say about quantum systems.

But in the early 2010s or so, as I was starting my graduate work, quantum thermodynamics started to grow a great deal, I think for two reasons. First, the field of quantum information science had matured in the early 2000s, and we could use it as a mathematical, conceptual and experimental toolkit for understanding quantum systems through how they store and process information. We could use those tools in quantum thermodynamics.

And second, some people managed to secure a very large grant for quantum thermodynamics. And so over the past 10 to 15 years, quantum thermodynamics has really boomed, and I think that’s why it feels so new.

STROGATZ: Well, let’s unpack some of the words that you’ve been using here. We keep saying “quantum,” but maybe we should just offer a quick reminder of what are some of the key features that distinguish quantum phenomena or quantum systems we keep speaking of from classical systems. Like, what’s the hallmark of something being fundamentally quantum mechanical?

YUNGER HALPERN: Some features are, quantum systems tend to be small. They can have wavelike and particle-like natures. They can be disturbed a great deal by measurement in a way that classical systems aren’t. They can entangle with each other, [and] so form very strong relationships, which lead to really strong correlations. A quantum particle can have only certain amounts of energy, not absolutely any possible amount of energy from zero on upward.

STROGATZ: Yeah, that’s really where the word came from, isn’t it?

YUNGER HALPERN: Right, “quantum” literally means a small packet of something. And so an atom, say, can receive quanta of energy, and so jump between kind of rungs on their energy ladder to go from one discrete amount of energy to another discrete amount of energy.

STROGATZ: Yeah, it’s all very deep and mysterious. Is it fair to say that we discovered quantum ideas in the context of physics originally, in the quanta of quantized energy levels in atoms, but really, quantum ideas are more general than physics. For instance, could quantum theory be a kind of generalization of probability theory and information theory, divorced from applications to atoms?

YUNGER HALPERN: Quantum information science indeed has spread into computer science and mathematics and engineering, and it’s also inherently in chemistry.

I think of quantum information science in two ways. On the one hand, I think of quantum information science as, as I mentioned, a mathematical, conceptual and experimental toolkit for understanding quantum systems through how they store and process information, such as these strong correlations that I mentioned earlier — entanglement. And on the other hand, I think of quantum information science as the study of how we can use quantum phenomena, like disturbance by measurement, to process information — so solve computational problems, secure information, communicate information and so on — in ways that are impossible if we have just classical technologies.

STROGATZ: One of the ideas that I ran across while preparing to talk to you struck me as pretty mind-blowing, and I’m hoping you can help enlighten me and our listeners about it, was the relationship between information and work. That work can sometimes be done by changing the amount of information in a system? Am I getting that right? That, like, erasing information is tantamount to doing work, or something like that?

YUNGER HALPERN: So, work is a resource in thermodynamics because if we have work, then we can push a rock up a hill, or charge a battery, or power a car. Just as there are thermodynamic tasks, like charging batteries, there are information-processing tasks like storing information and solving computational problems.

And in information theory, information is a resource. What I find really interesting is that information can also serve as a resource in thermodynamics. If we have information and heat, we can kind of combine those to obtain work, which we could use to, say, power a car. And also the reverse is true. Work can be a resource in information processing.

You mentioned erasure. So suppose that you did a calculation and you’ve filled up a whole piece of scrap paper and now I need to do a calculation. I need scrap paper. And suppose you hand your calculation to me. I don’t know what your handwriting looks like, so it might be beautiful, so no offense — suppose it’s just really bad handwriting, so I can’t read anything.

[Both laugh]

YUNGER HALPERN: Then, I need to erase that information in order to have useful scrap paper for performing my computation. That erasure process is going to cost thermodynamic work. There’s a fundamental limit, a fundamental lower bound, on the amount of work required to erase information. So both information and work can serve as resources in both thermodynamics and information processing.

STROGATZ: Isn’t there some way of thinking about the relation between information and work by thinking about our old friend, the box filled with gas?

YUNGER HALPERN: At least two stories grow up out of the setting that you’ve described. One story goes by the name of Szilard’s engine. Leo Szilard (opens a new tab) was a great Hungarian-American physicist. He said, suppose that we have a gas in a box. Let’s just suppose for convenience and simplicity that the gas consists of one particle, to make things really easy. And suppose that we know which side of the box the gas is in: the right-hand side rather than the left-hand side. We can think of this whole system as storing a bit of information.

So a bit, the basic unit of information, is something that can have one of two possible values. You know, one or zero, right or left. So if we know that the gas particle is in the right-hand side of the box, we have one bit of information. We know it’s in the right-hand side rather than left-hand side.

And we can turn this bit of information into work. We can slide a partition into the box down the center, so we trap our gas in the right-hand side. And suppose that we let the gas interact with some fixed temperature environment through the walls of the box, so heat can flow into the box or out of the box. And we can hook up a little weight to the partition, and let the partition slide. The gas is going to hit the partition, and it’s going to keep punching the partition until the partition reaches the left-hand side of the box.

So now, the gas can be anywhere in the entire box. The gas has expanded, and as it has expanded and moved the partition, the partition has lifted the weight. So, we’ve lifted a weight, we’ve done useful thermodynamic work on the weight.

But we no longer know where in the box the particle is. We lost our bit of information. So we traded information for work, using heat that flowed into the gas from the environment. It’s that heat which we transformed into work with help from our bit of information.

STROGATZ: Huh, it’s really very vivid. I love that explanation that you just gave. Let me see if I really got it, I think I did. One molecule or one particle in a box. We think of it on the right. We’ve got this partition, a sort of sliding wall, potentially. The whole box is in a room at a certain temperature, and then somehow that’s enough to get this particle bouncing around randomly in its available space. And occasionally it will hit the partition. Is that the idea? That I’m only going to hit the partition from one side. So the partition moves unidirectionally, right to left, like a piston expanding.

In some jerky way, it’s going to be sliding, gradually expanding the available volume for this particle. So far so good, right?

YUNGER HALPERN: Yes.

STROGATZ: And so we’re losing information as that’s happening, in the sense that we’re increasing the volume, and so we have less information about where the particle is, until ultimately, when we’ve jammed the partition all the way over to the left wall, we now know nothing. The particle could be anywhere.

I guess we have zero bits of information at this point about where it is.

YUNGER HALPERN: Exactly.

STROGATZ: Yet we did useful work through this Rube Goldberg setup that you described. I’ve done work by losing information.

YUNGER HALPERN: Exactly.

STROGATZ: Unbelievable, that’s crazy.

YUNGER HALPERN: One thing I really love about Szilard’s engine is, this was described in a paper by Leo Szilard that was based on his Ph.D. thesis, which was praised by Einstein. And we’re still talking about it about a hundred years later. So I present this story as a motivation to grad students for something they can aspire to.

STROGATZ: Aha. So for listeners who wouldn’t know this name Leo Szilard, he’s pretty famous in the history, not just of science, but the history of the world because he’s the person who wrote the letter that Einstein signed to tell President Roosevelt that the U.S. needed to build an atomic bomb before Germany did.

YUNGER HALPERN: And he was also heavily involved after the Manhattan Project in causes of peace.

STROGATZ: Yes, right.

We’ll be right back after this message.

[Break for ad insertion]

STROGATZ: Welcome back.

So, this is Szilard’s engine, explaining how information can be traded for work. But you said there was a second example, about a box and a partition? Is there another one that we should talk about?

YUNGER HALPERN: I believe you alluded to this earlier: Landauer erasure.

STROGATZ: OK. Yes. Please tell me about those two words. Landauer. That’s Rolf Landauer (opens a new tab)?

YUNGER HALPERN: An information scientist at IBM.

STROGATZ: Uh huh.

YUNGER HALPERN: He basically reversed Szilard’s engine. Suppose that there is a gas in a box. Again, let’s think of it as being one particle. And suppose we don’t know where the gas particle is. Its position in the box is totally random, and we want to reset the particle’s position to a nice clean state: the right-hand side of the box.

This is like taking that sheet of scrap paper that has been scribbled on totally randomly and erasing it to a nice clean state. We can do that by sliding a partition into the box, now right next to the left-hand wall, and pushing the partition to the box’s center. The gas is going to be trapped in the right-hand side.

Now, in order to slide the partition, we have to compress the gas, and this compression process requires us to spend work. At the end of the day, we have a bit of information. We know the gas is in the right-hand side of the box, rather than the left-hand side. But we have spent work. So, in this case, we could trade work for information.

STROGATZ: Hmm. And why is this a conceptually important thought experiment? You’re saying this is another way of understanding that there is a tradeoff between work and information?

YUNGER HALPERN: In part, yes. As you said, there’s a tradeoff. Both work and information are useful, in information processing and thermodynamics. And you alluded to something deeper — I agree about that impulse. If we want to compute and keep computing and keep computing, we’re going to run out of scrap paper sometime. The universe doesn’t have an infinite supply. So we’re going to have to erase some.

Charlie Bennett (opens a new tab), somebody else at IBM, has argued that just ordinary computations can be performed at zero-energy cost. It would be very, very difficult to do and very impractical, but in principle possible. However, when it comes to erasure, there is this unavoidable work cost. So, there’s some part of computation — namely, erasure — that has an intrinsic fundamental thermodynamic cost.

When I first learned this, I was… I was thrilled and so surprised because computation and thermodynamics seem like they don’t necessarily have to have anything to do with each other, but they’re very closely bound up.

STROGATZ: Hmm, it’s really quite an astonishing idea. It seems like at least some aspects of thermodynamics very much bear on the information science of the future. I mean, if we’re going to do quantum computers soon, it sounds like we have to know something about thermodynamics. Did you say “Landauer limit” or something?

YUNGER HALPERN: Yes, it’s sometimes called the Landauer bound or the Landauer limit on the amount of work required. And there is a movement, maybe one could say, called the thermodynamics of information. And quantum thermodynamicists and our friends in adjacent fields really love putting together thermodynamics and information.

STROGATZ: So since I mentioned quantum computers, and I’m sure listeners are wondering: You know, we’re doing all these thought experiments with gases in boxes and moving partitions and things. This is not so far from reality, like maybe these things could become real at some point. So, should we shift gears now to that question of applications? Is there any hope in the near future or the long-term future to make quantum machines based on these principles? Or what should we think of in terms of coming applications?

YUNGER HALPERN: We could approach this question in a number of ways.

One is, experimentalists have recently started being able to investigate this Landauer limit on the work required for erasure. A group, for instance, in Finland, they used these electrons to store a bit of information, and they checked how much work they needed to erase the information.

You mentioned equilibrium earlier in the conversation. If we keep a system close to equilibrium, we don’t disturb it a whole lot. We don’t waste energy on riling it up. And so if we perform a process very slowly, such as the system is always in equilibrium or close to equilibrium, we can get away with spending relatively small amounts of work. And so they ran their process more and more slowly and saw what is the value of work, that seems like just below the amount of work that they had to pay, that they were approaching. And they found that their observations were consistent with Landauer’s bound.

This Landauer bound, this minimal amount of work required to erase a bit of information, that is pretty far from the amount of work that is actually spent to erase information in today’s hardware. We operate quite far from this fundamental limitation.

Some people are interested in trying to reach toward this bound to waste less energy. Heating of small parts of computers is a significant problem in both the classical realm of computing and the quantum realm of computing. That’s one motivation that people have for studying such fundamental limits.

And it does seem that quantum systems also obey an ultimate limit on the work cost of erasure. But the operations that we can perform, say in computation, can be different. And also, we can ask, as you did earlier, what if we have a quantum gas in our box rather than a classical gas? Does anything change?

Quite a few features change, and sometimes we can use quantum phenomena like entanglement as resources to help us out.

One of my favorite examples was proved by colleagues including Lídia del Rio (opens a new tab), a Portuguese physicist. And they said, suppose that our piece of information that we want to erase is a piece of quantum information stored in a quantum particle that’s entangled. It has this set of strong correlations, more or less, with another quantum system. And so there’s this kind of reference system and, you know, we can manipulate the entanglement. But what we want to do is erase the quantum information in this particle of interest while keeping the reference system in its same state. We don’t want to disturb it too much.

It turns out that we can erase the particle of interest while on the whole gaining work instead of spending work, which is counterintuitive because we’re supposed to spend work in order to erase information.

The trick is to kind of burn the correlations, the entanglement between the particle of interest and the reference in the presence of heat. So entanglement together with heat serves as this kind of thermodynamic quote-unquote “fuel” that we can use to erase while extracting work. This doesn’t violate Landauer’s principle because Landauer wasn’t actually thinking about a quantum particle that’s entangled. This entanglement is kind of an extra resource that we’re adding to the Landauer story after making it quantum.

So everything’s consistent, but it could be a little surprising that we can use entanglement as a resource in this decades-old erasure story.

STROGATZ: Ooh, it’s so interesting and weird what you just described. So there’s a particle in a box and it’s entangled, let’s say, with another particle outside the box, is that right?

YUNGER HALPERN: In this case, it might be useful to think of a different kind of platform or physical system. In classical information science, we think of bits. You know, a transistor in an ordinary computer encodes a bit. It can be in the zero state or the one state.

When it comes to quantum information science, the basic unit of information is the qubit — the quantum bit. And we can store a qubit, for example, in a property of an electron. The electron doesn’t have to be in a box, but if we bring two electrons together and perform some operation on them, as just one example, then we can have two qubits that are entangled with each other.

STROGATZ: Uh huh. So, hmm, remind me how that goes. Like what particles, what properties are entangled? Just to make it concrete.

YUNGER HALPERN: So one example is the spin of an electron.

STROGATZ: OK.

YUNGER HALPERN: It can store a qubit. There are also all sorts of other platforms that people are using to store qubits. For instance, superconducting qubits is a really tiny circuit printed on a chip and a current can flow in one direction in the circuit or flow in the other direction. These are kind of the two options that help define a qubit.

STROGATZ: Uh huh. So maybe now we have two entangled superconducting circuits in your setup. OK. And then — but going back to the thing that was blowing my mind, which is that you could somehow “burn entanglement” to provide —

YUNGER HALPERN: With scare quotes.

STROGATZ: Yeah, scare quotes. It is scary. I mean, entanglement is itself so spooky and amorphous in the way that an average person thinks about it, that the idea that you could burn it and use it as a resource, this is the first I’ve ever heard of this idea.

Entanglement gets destroyed sort of on its own very commonly, right? That’s the reason we’re not familiar with it in our ordinary macroscopic lives.

So it sounds in a way like what you’re describing, in a way, could be commonplace. Like if, if the destroying the entanglement is the key, then well, we do that all the time.

YUNGER HALPERN: Merely destroying the entanglement in absolutely any way probably won’t do the trick, but if you can control and manipulate the entanglement in the right way, then you can consume it in order to get out your work.

STROGATZ: I see. So, if you manipulate the entanglement in the right way, that in some sense destroys it, it could be destroyed in a productive way.

YUNGER HALPERN: Yes.

STROGATZ: And does this whole thought experiment that you’re describing now, are you saying we’re close to being able to do this? Or we can at least imagine we can do this?

YUNGER HALPERN: I wouldn’t be surprised if someone had tried it. I don’t think I’ve seen a paper but increasing numbers of quantum thermodynamics experiments are being performed, so I wouldn’t be too surprised if someone performed this experiment in the near future.

The original theory of thermodynamics went hand in hand with the Industrial Revolution, which was useful in a very different sense. And now people are starting to pivot to try to make quantum thermodynamics, maybe quantum thermal machines, actually useful for us, and not only really cool curiosities.

STROGATZ: [laughs] Yeah. I do think that’s a nice, honest answer that you’re giving, that so much of the pleasure of this subject seems to be the light it sheds on two very deep subjects, thermodynamics and quantum theory, which continues to be fascinating. And it seems that by combining those two, we’re getting even deeper understanding of both fields. It’s starting to become imaginable to actually make things using these ideas.

I know that you’ve been interested in something called autonomous quantum machines. Could you tell us a little about that? What is an autonomous quantum machine, in your mind or in reality?

YUNGER HALPERN: Sure. Quantum thermodynamicists have designed quantum engines, quantum refrigerators, quantum batteries, quantum ratchets, and some of them have been realized experimentally. I think the experiments are very impressive, they demonstrate that the experimentalists have excellent control. And just the idea of making a quantum engine is very fun.

On the other hand, you wouldn’t want to invest in a company that provides quantum engines because a quantum engine is so small, it outputs very little energy but cooling a system down so that it behaves in a quantum fashion and then manipulating the engine costs loads of energy. So the engine isn’t worth it — except for the fun factor.

Now there are, even in classical thermodynamics, engines and refrigerators, machines, that are autonomous. They can run on their own without the need for control that changes over time. You just give the machine access to some energy in its environment. The machine will extract the energy from its environment and do its own thing.

We can also make quantum versions of these autonomous thermodynamic machines. And since autonomous machines don’t require lots of control, they offer some hope for actually making useful quantum thermodynamic machines.

I recently collaborated with the lab of Simone Gasparinetti (opens a new tab) at Chalmers University in Sweden. Chalmers University is building a quantum computer from superconducting qubits. These superconducting qubits are in a large classical refrigerator called a dilution refrigerator, which cools the qubits down to, I think, tens of milliKelvin.

But suppose that this quantum computer has just finished a calculation. It’s used some qubits up as scrap paper, which we keep returning to in this conversation. And if we want to perform the next quantum computation, we need to clean off the scrap paper. And in experimental language, that means we need to cool down the qubits even more in order to reset them. Even more than this classical refrigerator can manage.

One can design a chip to stick inside the classical refrigerator, that consists of more superconducting qubits that act as an autonomous quantum refrigerator. You can kind of hand over your computational qubits to this quantum refrigerator and let it do its thing, and then take the qubits away and use them in your computation. In order to get the quantum refrigerator to behave in a quantum way, you have to make it cold. But you just stick it inside this classical refrigerator, which is already cold because you’re already performing quantum computations.

So, the experiment that was done in Simone’s lab was a proof-of-principle experiment, but it performed a lot better than I expected. So we’re hoping that it’s the beginning of making autonomous quantum machines useful.

STROGATZ: Hmm. Uh huh. It makes me wonder, what is it about what you do and what you get to think about that brings you joy?

YUNGER HALPERN: Good question. I have always loved dealing in abstract ideas. I have always loved reading because that’s always given me the opportunity to build universes in my head. In high school, I was very attracted to philosophy and mathematics, computer science, physics and so on. And quantum thermodynamics gives me joy because it enables me to play with these abstract ideas. I get to build universes in my head, you know, models of all sorts of different systems for a job. And I get to engage with all these different subjects and all their ideas, but there’s also a sense of balance because there are applications of quantum information theory and quantum thermodynamics to quantum technologies like quantum computers and cryptography and so on. So, I get to, you know, play in the realm of ideas and also feel like maybe I’m doing something useful.

STROGATZ: [laughs] Very impressive. Nice. We’ve been speaking with theoretical physicist Nicole Yunger Halpern about the ins and outs of quantum thermodynamics. It’s really been fun, Nicole. Thank you so much for joining us.

YUNGER HALPERN: Thank you. It’s been a lot of fun.

[Theme plays]

STROGATZ: Thanks for listening. If you’re enjoying “The Joy of Why” and you’re not already subscribed, hit the subscribe or follow button where you’re listening. You can also leave a review for the show — it helps people find this podcast.

[Theme continues]

“The Joy of Why” is a podcast from Quanta Magazine, an editorially independent publication supported by the Simons Foundation. Funding decisions by the Simons Foundation have no influence on the selection of topics, guests or other editorial decisions in this podcast or in Quanta Magazine.

“The Joy of Why” is produced by PRX Productions (opens a new tab). The production team is Caitlin Faulds, Livia Brock, Genevieve Sponsler, and Merritt Jacob. The executive producer of PRX Productions is Jocelyn Gonzales. Morgan Church and Edwin Ochoa provided additional assistance. From Quanta Magazine, John Rennie and Thomas Lin provided editorial guidance, with support from Matt Carlstrom, Samuel Velasco, Arleen Santana and Meghan Willcoxon. Samir Patel is Quanta’s editor in chief.

Our theme music is from APM Music. Julian Lin came up with the podcast name. The episode art is by Peter Greenwood and our logo is by Jaki King and Kristina Armitage. Special thanks to the Columbia Journalism School and Bert Odom-Reed at the Cornell Broadcast Studios.

I’m your host, Steve Strogatz. If you have any questions or comments for us, please email us at quanta@simonsfoundation.org (opens a new tab). Thanks for listening.

[Theme fades out]

It was early January 2016, and I had just joined Google X, Alphabet’s secret innovation lab. My job: help figure out what to do with the employees and technology left over from nine robot companies that Google had acquired. People were confused. Andy “the father of Android” Rubin, who had previously been in charge, had suddenly left. Larry Page and Sergey Brin kept trying to offer guidance and direction during occasional flybys in their “spare time.” Astro Teller, the head of Google X, had agreed a few months earlier to bring all the robot people into the lab, affectionately referred to as the moonshot factory.

I signed up because Astro had convinced me that Google X—or simply X, as we would come to call it—would be different from other corporate innovation labs. The founders were committed to thinking exceptionally big, and they had the so-called “patient capital” to make things happen. After a career of starting and selling several tech companies, this felt right to me. X seemed like the kind of thing that Google ought to be doing. I knew from firsthand experience how hard it was to build a company that, in Steve Jobs’ famous words, could put a dent in the universe, and I believed that Google was the right place to make certain big bets. AI-powered robots, the ones that will live and work alongside us one day, was one such audacious bet.

Eight and a half years later—and 18 months after Google decided to discontinue its largest bet in robotics and AI—it seems as if a new robotics startup pops up every week. I am more convinced than ever that the robots need to come. Yet I have concerns that Silicon Valley, with its focus on “minimum viable products” and VCs’ general aversion to investing in hardware, will be patient enough to win the global race to give AI a robot body. And much of the money that is being invested is focusing on the wrong things. Here is why.

The Meaning of “Moonshot”

Google X—the home of Everyday Robots, as our moonshot came to be known—was born in 2010 from a grand idea that Google could tackle some of the world’s hardest problems. X was deliberately located in its own building a few miles away from the main campus, to foster its own culture and allow people to think far outside the proverbial box. Much effort was put into encouraging X-ers to take big risks, to rapidly experiment, and even to celebrate failure as an indication that we had set the bar exceptionally high. When I arrived, the lab had already hatched Waymo, Google Glass, and other science-fiction-sounding projects like flying energy windmills and stratospheric balloons that would provide internet access to the underserved.

What set X projects apart from Silicon Valley startups is how big and long-term X-ers were encouraged to think. In fact, to be anointed a moonshot, X had a “formula”: The project needed to demonstrate, first, that it was addressing a problem that affects hundreds of millions, or even billions, of people. Second, there had to be a breakthrough technology that gave us line of sight to a new way to solve the problem. Finally, there needed to be a radical business or product solution that probably sounded like it was just on the right side of crazy.

The AI Body Problem

It’s hard to imagine a person better suited to running X than Astro Teller, whose chosen title was literally Captain of Moonshots. You would never see Astro in the Google X building, a giant, three-story converted department store, without his signature rollerblades. Top that with a ponytail, always a friendly smile, and, of course, the name Astro, and you might think you’d entered an episode of HBO’s Silicon Valley.

When Astro and I first sat down to discuss what we might do with the robot companies that Google had acquired, we agreed something should be done. But what? Most useful robots to date were large, dumb, and dangerous, confined to factories and warehouses where they often needed to be heavily supervised or put in cages to protect people from them. How were we going to build robots that would be helpful and safe in everyday settings? It would require a new approach. The huge problem we were addressing was a massively global human shift—aging populations, shrinking workforces, labor shortages. Our breakthrough technology was—we knew, even in 2016—going to be artificial intelligence. The radical solution: fully autonomous robots that would help us with an ever-growing list of tasks in our everyday lives.

We were, in other words, going to give AI a body in the physical world, and if there was one place where something of this scale could be concocted, I was convinced it would be X. It was going to take a long time, a lot of patience, a willingness to try crazy ideas and fail at many of them. It would require significant technical breakthroughs in AI and robot technology and very likely cost billions of dollars. (Yes, billions.) There was a deep conviction on the team that, if you looked just a bit beyond the horizon, a convergence of AI and robotics was inevitable. We felt that much of what had only existed in science fiction to date was about to become reality.

It’s Your Mother

Every week or so, I’d talk to my mother on the phone. Her opening question was always the same: “When are the robots coming?” She wouldn’t even say hello. She just wanted to know when one of our robots would come help her. I would respond, “It’ll be a while, Mom,” whereupon she’d say, “They better come soon!”

Living in Oslo, Norway, my mom had good public health care; caregivers showed up at her apartment three times daily to help with a range of tasks and chores, mostly related to her advanced Parkinson’s disease. While these caregivers enabled her to live alone in her own home, my mother hoped that robots could support her with the myriad of small things that had now become insurmountable and embarrassing barriers, or sometimes simply offer her an arm to lean against.

It’s Really Hard

“You do know that robotics is a systems problem, right?” Jeff asked me with a probing look. Every team seems to have a “Jeff”; Jeff Bingham was ours. He was a skinny, earnest guy with a PhD in bioengineering who grew up on a farm and had a reputation for being a knowledge hub with deep insights about … kinda everything. To this day, if you ask me about robots, one of the first things I’ll tell you is that, well, it’s a systems problem.

One of the important things Jeff was trying to reinforce was that a robot is a very complex system and only as good as its weakest link. If the vision subsystem has a hard time perceiving what’s in front of it in direct sunlight, then the robots may suddenly go blind and stop working if a ray of sun comes through a window. If the navigation subsystem doesn’t understand stairs, then the robot may tumble down them and hurt itself (and possibly innocent bystanders). And so on. Building a robot that can live and work alongside us is hard. Like, really hard.

For decades people have been trying to program various forms of robots to perform even simple tasks, like grasping a cup on a table or opening a door, and these programs have always ended up becoming extremely brittle, failing at the slightest change in conditions or variations in the environment. Why? Because of the lack of predictability in the real world (like that ray of sunlight). And we haven’t even gotten to the hard stuff yet, like moving through the messy and cluttered spaces where we live and work.

Once you start thinking carefully about all this, you realize that unless you lock everything down, really tight, with all objects being in fixed, predefined locations, and the lighting being just right and never changing, simply picking up, say, a green apple and placing it in a glass bowl on a kitchen table becomes an all but impossibly difficult problem to solve. This is why factory robots are in cages. Everything from the lighting to the placement of the things they work on can be predictable, and they don’t have to worry about bonking a person on the head.

How to Learn Learning Robots

But all you need, apparently, is 17 machine-learning people. Or so Larry Page told me—one of his classic, difficult-to-comprehend insights. I tried arguing that there was no way we could possibly build the hardware and software infrastructure for robots that would work alongside us with only a handful of ML researchers. He waved his hand at me dismissively. “All you need is 17.” I was confused. Why not 11? Or 23? I was missing something.

Boiling it down, there are two primary approaches to applying AI in robotics. The first is a hybrid approach. Different parts of the system are powered by AI and then stitched together with traditional programming. With this approach the vision subsystem may use AI to recognize and categorize the world it sees. Once it creates a list of the objects it sees, the robot program receives this list and acts on it using heuristics implemented in code. If the program is written to pick that apple off a table, the apple will be detected by the AI-powered vision system, and the program would then pick out a certain object of “type: apple” from the list and then reach to pick it up using traditional robot control software.

The other approach, end-to-end learning, or e2e, attempts to learn entire tasks like “picking up an object,” or even more comprehensive efforts like “tidying up a table.” The learning happens by exposing the robots to large amounts of training data—in much the way a human might learn to perform a physical task. If you ask a young child to pick up a cup, they may, depending on how young they are, still need to learn what a cup is, that a cup might contain liquid, and then, when playing with the cup, repeatedly knock it over, or at least spill a lot of milk. But with demonstrations, imitating others, and lots of playful practice, they’ll learn to do it—and eventually not even have to think about the steps.

What I came to believe Larry was saying was that nothing really mattered unless we ultimately demonstrated that robots could learn to perform end-to-end tasks. Only then would we have a real shot at making robots reliably perform these tasks in the messy and unpredictable real world, qualifying us to be a moonshot. It wasn’t about the specific number 17, but about the fact that big breakthroughs require small teams, not armies of engineers. Obviously there is a lot more to a robot than its AI brain, so I did not discontinue our other engineering efforts—we still had to design and build a physical robot. It became clear, though, that demonstrating a successful e2e task would give us some faith that, in moonshot parlance, we could escape Earth's gravitational pull. In Larry’s world, everything else was essentially “implementation details.”

On the Arm-Farm

Peter Pastor is a German roboticist who received his PhD in robotics from the University of Southern California. On the rare occasion when he wasn’t at work, Peter was trying to keep up with his girlfriend on a kiteboard. In the lab, he spent a lot of his time wrangling 14 proprietary robot arms, later replaced with seven industrial Kuka robot arms in a configuration we dubbed “the arm-farm.”

These arms ran 24/7, repeatedly attempting to pick up objects, like sponges, Lego blocks, rubber ducklings, or plastic bananas, from a bin. At the start they would be programmed to move their claw-like gripper into the bin from a random position above, close the gripper, pull up, and see if they had caught anything. There was a camera above the bin that captured the contents, the movement of the arm, and its success or failure. This went on for months.

In the beginning, the robots had a 7 percent success rate. But each time a robot succeeded, it got positive reinforcement. (Basically meaning, for a robot, that so-called “weights” in the neural network used to determine various outcomes are adjusted to positively reinforce the desired behaviors, and negatively reinforce the undesired ones.) Eventually, these arms learned to successfully pick up objects more than 70 percent of the time. When Peter showed me a video one day of a robot arm not just reaching down to grasp a yellow Lego block but nudging other objects out of the way in order to get a clear shot at it, I knew we had reached a real turning point. The robot hadn’t been explicitly programmed, using traditional heuristics, to make that move. It had learned to do it.

But still—seven robots working for months to learn how to pick up a rubber duckling? That wasn’t going to cut it. Even hundreds of robots practicing for years wouldn’t be enough to teach the robots to perform their first useful real-world tasks. So we built a cloud-based simulator and, in 2021, created more than 240 million robot instances in the sim.

Think of the simulator as a giant video game, with a model of real-world physics that was realistic enough to simulate the weight of an item or the friction of a surface. The many thousands of simulated robots would use their simulated camera input and their simulated bodies, modeled after the real robots, to perform their tasks, like picking up a cup from a table. Running at once, they would try and fail millions of times, collecting data to train the AI algorithms. Once the robots got reasonably good in simulation, the algorithms were transferred to physical robots to do final training in the real world so they could embody their new moves. I always thought of the simulation as robots dreaming all night and then waking up having learned something new.

It’s the Data, Stupid

The day we all woke up and discovered ChatGPT, it seemed like magic. An AI-powered system could suddenly write complete paragraphs, answer complicated questions, and engage in an ongoing dialog. At the same time, we also came to understand its fundamental limitation: It had taken enormous amounts of data to accomplish this.

Robots are already leveraging large language models to understand spoken language and vision models to understand what they see, and this makes for very nice YouTube demo videos. But teaching robots to autonomously live and work alongside us is a comparably huge data problem. In spite of simulations and other ways to create training data, it is highly unlikely that robots will “wake up” highly capable one day, with a foundation model that controls the whole system.

The verdict is still out on how complex the tasks will be that we can teach a robot to perform with AI alone. I have come to believe it will take many, many thousands, maybe even millions of robots doing stuff in the real world to collect enough data to train e2e models that make the robots do anything other than fairly narrow, well-defined tasks. Building robots that perform useful services—like cleaning up and wiping all the tables in a restaurant, or making the beds in a hotel—will require both AI and traditional programming for a long time to come. In other words, don’t expect robots to go running off outside our control, doing something they weren’t programmed to do, anytime soon.

But Should They Look Like Us?

Horses are very efficient at walking and running on four legs. Yet we designed cars to have wheels. Human brains are incredibly efficient biological computers. Yet chip-based computers don’t come close to performing like our brains. Why don’t cars have legs, and why weren’t computers modeled on our biology? The goal of building robots, I mean to say, shouldn’t just be mimicry.

This I learned one day at a meeting with a group of technical leaders at Everyday Robots. We were sitting around a conference table having an animated conversation about whether our robots should have legs or wheels. Such discussions tended to devolve more into religious debates than fact-based or scientific ones. Some people are very attached to the idea that robots should look like people. Their rationale is good. We have designed the places in which we live and work to accommodate us. And we have legs. So maybe robots should too.

After about 30 minutes, the most senior engineering manager in the room, Vincent Dureau, spoke up. He simply said, “I figure that if I can get there, the robots should be able to get there.” Vincent was seated in his wheelchair. The room went quiet. The debate was over.

The fact is, robot legs are mechanically and electronically very complex. They don’t move very fast. They’re prone to making the robot unstable. They’re also not very power-efficient compared to wheels. These days, when I see companies attempting to make humanoid robots—robots that try to closely mimic human form and function—I wonder if it is a failure of imagination. There are so many designs to explore that complement humans. Why torture ourselves reaching for mimicry? At Everyday Robots, we tried to make the morphology of the robot as simple as possible—because the sooner robots can perform real-world tasks, the faster we can gather valuable data. Vincent’s comment reminded us that we needed to focus on the hardest, most impactful problems first.

Desk Duty

I was at my desk when one of our one-armed robots with a head shaped like a rectangle with rounded corners rolled up, addressed me by name, and asked if it could tidy up. I said yes and stepped aside. A few minutes later it had picked up a couple of empty paper cups, a transparent iced tea cup from Starbucks, and a plastic Kind bar wrapper. It dropped these items into a trash tray attached to its base before turning toward me, giving me a nod, and heading over to the next desk.

This tidy-desk service represented an important milestone: It showed that we were making good progress on an unsolved part of the robotics puzzle. The robots were using AI to reliably see both people and objects! Benjie Holson, a software engineer and former puppeteer who led the team that created this service, was an advocate for the hybrid approach. He wasn’t against end-to-end learned tasks but simply had a let’s-try-to-make-them-do-something-useful-now attitude. If the ML researchers solved some e2e task better than his team could program it, they’d just pull the new algorithms into their quiver.

I’d gotten used to our robots rolling around, doing chores like tidying desks. Occasionally I would spot a visitor or an engineer who had just joined the team. They’d have a look of wonder and joy on their face as they watched the robots going about their business. Through their eyes I was reminded just how novel this was. As our head of design, Rhys Newman, would say when a robot rolled by one day (in his Welsh accent), “It’s become normal. That’s weird, isn’t it?”

Just Dance

Our advisers at Everyday Robots included a philosopher, an anthropologist, a former labor leader, a historian, and an economist. We vigorously debated economic, social, and philosophical questions like: If robots lived alongside us, what would the economic impact be? What about the long-term and near-term effects on labor? What does it mean to be human in an age of intelligent machines? How do we build these machines in ways that make us feel welcome and safe?

In 2019, after telling my team that we were looking for an artist in residence to do some creative, weird, and unexpected things with our robots, I met Catie Cuan. Catie was studying for her PhD in robotics and AI at Stanford. What caught my attention was that she had been a professional dancer, performing at places like the Metropolitan Opera Ballet in NYC.

You’ve probably seen YouTube videos of robots dancing—performances where the robot carries out a preprogrammed sequence of timed moves, synchronized to music. While fun to watch, these dances are not much different than what you’d experience on a ride at Disneyland. I asked Catie what it would be like if, instead, robots could improvise and engage with each other like people do. Or like flocks of birds, or schools of fish. To make this happen, she and a few other engineers developed an AI algorithm trained on the preferences of a choreographer. That being, of course, Catie.

Often during evenings and sometimes weekends, when the robots weren’t busy doing their daily chores, Catie and her impromptu team would gather a dozen or so robots in a large atrium in the middle of X. Flocks of robots began moving together, at times haltingly, yet always in interesting patterns, with what often felt like curiosity and sometimes even grace and beauty. Tom Engbersen is a roboticist from the Netherlands who painted replicas of classic masterpieces in his spare time. He began a side project collaborating with Catie on an exploration of how dancing robots might respond to music or even play an instrument. At one point he had a novel idea: What if the robots became instruments themselves? This kicked off an exploration where each joint on the robot played a sound when it moved. When the base moved it played a bass sound; when a gripper opened and closed it made a bell sound. When we turned on music mode, the robots created unique orchestral scores every time they moved. Whether they were traveling down a hallway, sorting trash, cleaning tables, or “dancing” as a flock, the robots moved and sounded like a new type of approachable creature, unlike anything I had ever experienced.

This Is Only the Beginning

In late 2022, the end-to-end versus hybrid conversations were still going strong. Peter and his teammates, with our colleagues in Google Brain, had been working on applying reinforcement learning, imitation learning, and transformers—the architecture behind LLMs—to several robot tasks. They were making good progress on showing that robots could learn tasks in ways that made them general, robust, and resilient. Meanwhile, the applications team led by Benjie was working on taking AI models and using them with traditional programming to prototype and build robot services that could be deployed among people in real-world settings.

Meanwhile, Project Starling, as Catie’s multi-robot installation ended up being called, was changing how I felt about these machines. I noticed how people were drawn to the robots with wonder, joy, and curiosity. It helped me understand that how robots move among us, and what they sound like, will trigger deep human emotion; it will be a big factor in how, even if, we welcome them into our everyday lives.

We were, in other words, on the cusp of truly capitalizing on the biggest bet we had made: robots powered by AI. AI was giving them the ability to understand what they heard (spoken and written language) and translate it into actions, or understand what they saw (camera images) and translate that into scenes and objects that they could act on. And as Peter’s team had demonstrated, robots had learned to pick up objects. After more than seven years we were deploying fleets of robots across multiple Google buildings. A single type of robot was performing a range of services: autonomously wiping tables in cafeterias, inspecting conference rooms, sorting trash, and more.

Which was when, in January 2023, two months after OpenAI introduced ChatGPT, Google shut down Everyday Robots, citing overall cost concerns. The robots and a small number of people eventually landed at Google DeepMind to conduct research. In spite of the high cost and the long timeline, everyone involved was shocked.

A National Imperative

In 1970, for every person over 64 in the world, there were 10 people of working age. By 2050, there will likely be fewer than four. We’re running out of workers. Who will care for the elderly? Who will work in factories, hospitals, restaurants? Who will drive trucks and taxis? Countries like Japan, China, and South Korea understand the immediacy of this problem. There, robots are not optional. Those nations have made it a national imperative to invest in robotics technologies.

Giving AI a body in the real world is both an issue of national security and an enormous economic opportunity. If a technology company like Google decides it cannot invest in “moonshot” efforts like the AI-powered robots that will complement and supplement the workers of the future, then who will? Will the Silicon Valley or other startup ecosystems step up, and if so, will there be access to patient, long-term capital? I have doubts. The reason we called Everyday Robots a moonshot is that building highly complex systems at this scale went way beyond what venture-capital-funded startups have historically had the patience for. While the US is ahead in AI, building the physical manifestation of it—robots—requires skills and infrastructure where other nations, most notably China, are already leading.

The robots did not show up in time to help my mother. She passed away in early 2021. Our frequent conversations toward the end of her life convinced me more than ever that a future version of what we started at Everyday Robots will be coming. In fact, it can’t come soon enough. So the question we are left to ponder becomes: How does this kind of change and future happen? I remain curious, and concerned.

Let us know what you think about this article. Submit a letter to the editor at mail@wired.com.

Stephanie Harrison is an expert in the science of happiness and founded a company called The New Happy, where she teaches millions of people how to be happier. Through it, she hosts The New Happy podcast. She is also a Harvard Business Review and CNBC contributor, and her work has been featured in other publications such as Fast Company, Forbes, and Architectural Digest. She regularly speaks at Fortune 500 companies, advising on employee wellbeing and company culture.

Below, Stephanie shares five key insights from her new book, New Happy: Getting Happiness Right in a World That’s Got It Wrong. Listen to the audio version—read by Stephanie herself—in the Next Big Idea App.

Audio Player

1. Everything you know about happiness is a lie.

When I was in my early twenties, I had everything that I thought would make me happy. I had a prestigious job, lived in New York City, and had complete freedom. Yet, I was absolutely miserable. At first, I ignored my emotions. Then, over time, I started to experience more challenges: getting physically ill, struggling with my mental health, and feeling lonely. One day, I found myself lying on my bedroom floor sobbing hysterically, wondering why I was so desperately unhappy.

Then, I had a moment of clarity. What if there wasn’t something wrong with me? What if I had been lied to by the world around me? Perhaps everything I had been told about what I needed to do to be happy was wrong.

At that moment, I didn’t know exactly what the lies were, but now, after ten years of research, I do. I call it Old Happy: our society’s incorrect definition of happiness and the culture we’ve created around it. Old Happy begins with the messages we receive as children from our families, all the way through to the media we consume and the institutions that enforce it. It comes from three cultural forces of individualism, capitalism, and domination, which tell us that to be happy, we must perfect ourselves, do more and more, and do it all by ourselves.

The devastating truth is that pursuing these objectives won’t make you happy. In fact, both research and experience show that it will actually make you miserable.

2. A happier life starts with unwinding Old Happy.

Due to Old Happy, Americans are struggling with unprecedented levels of unhappiness, illness, burnout, and loneliness, with no idea what’s wrong or what they need to do to feel better again. The evidence I’ve amassed about the harms of Old Happy is astounding. To live truly happy lives, we start by letting go of our Old Happy beliefs and adopt new ones by undergoing three key shifts:

• Old Happy taught you that you’re not good enough and that there’s something wrong with you. Instead, you need to start seeing that you are worthy exactly as you are.
• Old Happy taught you that, to prove how good you are, you must achieve a certain set of external goals and succeed. Instead, you need to focus on expressing yourself and growing as a person in whatever way feels authentic to you. You are not defined by your successes or failures.
• Old Happy taught you that you have to do everything by yourself. Instead, you need to see that you are connected to others and that no one does anything alone. We are social creatures who are wired to need support. You are inextricably connected to others.

The best way to do this is by naming Old Happy when you see it pop up in your life. When you feel the pressure to overwork, say to yourself, “That’s Old Happy, not me.” When you judge your appearance, remind yourself, “I’m comparing myself to Old Happy’s made-up standards.” When you feel like you can’t ask for help, tell yourself that no one ever does anything alone, and it’s perfectly human to need support.

3. The real secret to happiness is counterintuitive.

Once we’ve named Old Happy and started unwinding it from our lives, we can discover the real secret to happiness. If you want to be happy, you need to help other people be happy. This is the proven path to happiness, supported by my research across multiple fields.

Everyone wants to live a happy life. The way to experience that is through finding ways to be of service to one another. Helping others is scientifically proven to benefit our wellbeing; it connects us to one another and helps us find a greater purpose in life. It doesn’t just improve your mental health but your physical health, too. Just like we have a need for food and shelter, we also have a profound need to go beyond ourselves and help others.

“If you want to be happy, you need to help other people be happy.”

Many people have a narrow definition of helping: we think of it as going out and volunteering. While that’s a wonderful way to help, we need a more expansive understanding. You help by listening to your loved ones, holding the door for someone, collaborating at work, sharing your ideas and unique perspective, and encouraging others to be their best. Every day, there are countless ways to help, meaning there are countless opportunities to experience happiness.

4. You possess unique gifts that need to be shared.

I argue that the best way to help others comes from sharing your unique gifts with those around you—whether in your family, communities, at work, or for the broader world.

There are three types of gifts that all human beings possess: humanity, talent, and wisdom:

• Your humanity is who you are as a person. It’s your character, your best qualities, your good and loving nature. When you call a friend to listen to them talk about a challenge, take time to play with your kids after work, or smile at a stranger on the street, you are using your humanity gifts.
• Your wisdom is what you have learned. Each of us possesses a completely unique life packed with experiences that teach us important and meaningful lessons. That wisdom can be used to help people in countless ways—from helping others through hard times to preventing them from ever happening.
• Your talent is what you can do. Talents are cultivated through time, energy, and effort. Every single one of us has the power to either develop new talents or deepen existing ones, using them to inspire others and make powerful contributions.

Your gifts are what make you you. When you use them in service of others, you’ll experience profound joy, purpose, and contentment. That’s what New Happy is all about: being yourself and giving of yourself.

5. Your happiness has the power to change the world.

When we live by Old Happy, we are not only making ourselves miserable, but we’re contributing to creating a world that makes the collective unhappy, too. It only leads to competition, judgment, disempowerment, burnout, and isolation. No one wins when Old Happy is our dominant understanding of happiness.

But when you adopt New Happy, all of that changes. Through your daily actions, you’re now contributing to making the world a better place. By helping others experience happiness and by devoting your incredible gifts to the problems we face, you are slowly but surely transforming the world into a place where more and more people get to be happy. Isn’t that what we all long for? A better, more just, more compassionate world?

I often hear from people in my community that they feel so helpless about the state of the world. But you can start making it better right now simply by changing your definition of happiness and living in alignment with it. Working for the greater good facilitates your highest good.

To listen to the audio version read by author Stephanie Harrison, download the Next Big Idea App today: