Elon Musk: A future worth getting excited about | Tesla Texas Gigafactory interview | TED
In general, I believe that the discussion revolves around problems, whether minor or significant.
Many people are concerned about the future and are pessimistic.
I believe this is not beneficial.
We should be eager to wake up in the morning and look forward to the day ahead.
We should be excited about what the future holds.
Life cannot be reduced to solving one misery after another.
If we look ahead 30 years, the year 2050 has been designated by scientists as a climate change "doomsday deadline." There is a consensus among scientists that if we do not completely eliminate greenhouse gases or entirely offset them by 2050, we will effectively invite climate catastrophe.
Do you believe that there is a way to avoid such a disaster?
Could you give me an explanation?
Yes, I am not one of the doomsday prophets, which may surprise you.
I believe we are on the right track.
However, I urge that we avoid becoming complacent.
As long as we are not complacent and maintain a strong sense of urgency about transitioning to a sustainable energy economy, I believe things will be fine.
I cannot overstate how important it is for us to push hard and avoid complacency;
the future will be bright.
Don't worry about it.
I mean, be concerned, but ironically, your concern will become a self-fulfilling prophecy.
There are three essential components to a sustainable energy future: sustainable energy generation, which is primarily derived from wind and solar, as well as hydro, geothermal, and I am pro-nuclear.
Nuclear power is acceptable in my opinion.
However, the primary sources of energy will be solar and wind.
The second component is that you require batteries to store solar and wind energy because the sun does not shine all the time and the wind does not blow all the time.
As a result, numerous stationary battery arrays are required.
Finally, you require electric transportation.
As a result, electric vehicles, electric planes, and boats are all required.
Electric rockets are not possible, but the propellant used in rockets can be produced using sustainable energy sources.
As a result, we can establish a fully sustainable energy economy.
These three things are the key: solar/wind, stationary battery packs, and electric vehicles.
What are the constraints on progress then?
Battery cell production will be the primary limiting factor.
As a result, that will be the fundamental driving force behind the rate.
The slowest component of the lithium-ion battery cell supply chain, which includes mining, refining, and finally creating a battery cell and packaging it, will be the limiting factor in the progress of sustainability.
We must discuss batteries further because I want to understand something.
There appears to be a scaling problem here that is both fascinating and alarming.
You have calculated that the amount of battery production required for sustainability globally is 300 terawatt hours.
Is that the ultimate goal?
These are extremely rough estimates, and I would encourage others to examine our calculations because they may reach different conclusions.
However, in order to transition not just current electricity production but also heating and transportation, which roughly triples the amount of electricity required, it amounts to approximately 300 terawatt hours of installed capacity.
We need to give people a sense of the scale of this undertaking.
We are currently located at the Gigafactory.
As you are aware, this is one of the world's largest structures.
What I've read, and please correct me if I'm incorrect, is that the current goal is to eventually produce 100 gigawatt hours of batteries per year here.
We will most likely outperform that, but yes, in a couple of years, we should hopefully get there.
I'm aware of that, but that is still just a small fraction of what is required.
How much of the remaining 100 do you anticipate Tesla will capture between now and 2030, 2040, when we truly need the scale-up to occur?
I mean, these are all merely educated guesses.
As a result, I hope people will not hold me accountable for them.
It's not as though this were a...
What often occurs is that I'll make some educated guess, and then five years later, some idiot will write an article claiming "Elon predicted this would happen, and it didn't happen.
He's a liar and an imbecile." It is really annoying when that occurs.
As a result, these are merely educated guesses;
this is merely a conversation.
We need to introduce a fully sustainable electric grid based on a mix of the sustainable energy sources you mentioned by 2050.
Because that grid will likely offer the world extremely low-cost energy compared to today, is that not accurate?
I'm also wondering if people are allowed to be a little enthusiastic about the possibilities of that world.
People should have a positive outlook on the future.
Humanity will prevail in the area of sustainable energy.
If we continue to push hard, the future for energy will be bright and positive.
It will then be feasible to use that energy for carbon sequestration as well.
In order to extract carbon from the atmosphere, a significant amount of energy is required because releasing carbon into the atmosphere releases energy.
As a result, you now understand that in order to extract it, you must invest a significant amount of energy.
However, if you have access to a lot of sustainable energy from wind and solar, you can capture carbon.
As a result, you can revert the carbon parts per million of the atmosphere and oceans.
Additionally, you can have as much fresh water as you want.
Earth is mostly covered in water.
As a result, we should refer to Earth as "Water." More than 70% of its surface area consists of water.
However, the majority of that is seawater, it is as though we simply happen to be on the section that is land.
And with energy, seawater can be transformed into...
Yes.
Irrigating water or whatever water you need.
At a very minimal cost.
Things will be great.
And things will be fine.
Additionally, there are other advantages to this non-fossil fuel world where the air is cleaner because as you burn fossil fuels, all these side reactions and various forms of hazardous gases are produced.
As well as little particulates that are detrimental to your lungs.
There are all sorts of negative things that are happening that will disappear.
And the sky will be brighter and more peaceful.
The future is going to be wonderful.
I'd like for us to now turn our attention to the subject of artificial intelligence.
You mentioned the annoyance you experience when people approach you with outdated predictions from the past when we transition to that subject.
As a result, I'm probably being irritating right now, but I'm curious about your timelines and how you predict and why some things turn out to be so startlingly accurate while others do not.
When it comes to predicting Tesla vehicle sales, for instance, I believe you have been incredibly successful.
For instance, in 2014, when Tesla had sold 60,000 vehicles that year, you predicted that "we will sell half a million per year in 2020." Yes, we came close to selling half a million units.
You came really close to selling half a million.
You were ridiculed in 2014 since no one since Henry Ford and the Model T had come close to that rate of growth for automobiles.
You were laughed at, but you actually reached 500,000 cars and then produced 510,000 or so.
However, five years ago, when you last appeared at TED, I questioned you about full self-driving, and you replied, "Yes, this very year, I'm convinced that we will have a car traveling from LA to New York without any human intervention." Yeah, I don't want to blow your mind, but I'm not always correct.
(Laughs) What distinguishes the two?
Why has predicting full self-driving in particular been so challenging?
I mean, what really got me, and I believe it will get a lot of other people, is that with self-driving, there are just so many false dawns, where you believe you have the problem under control, and then it turns out that you simply hit a ceiling.
This is so because if you were to plot the development, it would resemble a log curve.
As a result, it is a series of log curves.
I believe most people are unfamiliar with log curves.
With your hands, illustrate the form.
It ascends, you know, fairly straight, and then it begins to decline, and your returns begin to diminish.
And you're like, "Uh oh, it was trending up, and now it's curving over, and you're starting to reach these local maxima where you don't realize how stupid you were.
And then it occurs again.
And in the end...
These things, in retrospect, appear to be obvious, but in order to properly address full self-driving, you must actually solve real-world AI.
Because what are the road networks built to work with?
They are created to work with a biological neural network, our brains, and with vision, our eyes.
As a result, in order for it to function with computers, you must basically address real-world AI and vision.
Because we require cameras and silicon neural nets in order for self-driving to function with a system created for eyes and biological neural nets.
When you put it that way, I suppose it becomes rather obvious that the only way to solve complete self-driving is to solve real-world AI and advanced vision.
How do you feel about the present architecture?
Do you believe you now have an architecture that allows the logarithmic curve to avoid flattening down anytime soon?
Well, of course, these may be infamous last words, but I am really convinced that we will solve it this year.
Which will exceed...
What is the likelihood of an accident at which point you will surpass that of the average individual?
I believe we will surpass that this year.
What are you observing behind the scenes that gives you such confidence?
We are nearly at the point where we have a unified vector space of excellent quality.
We were initially attempting to accomplish this using image recognition on individual images.
However, if you extract a single image from a video, it can be extremely difficult to determine what is occurring without ambiguity.
However, if you examine a short video clip of several seconds, that ambiguity is resolved.
As a result, the first step was for us to connect all eight cameras so that they are synchronized, so that all frames are viewed and labeled simultaneously by a single person, as we still need human labeling.
As a result, they are not labeled at various times by different people in various ways.
It is therefore a sort of surround image.
Another important step is to add the time dimension.
So that you have a surround video that you are also labeling.
This is really hard to do from a software standpoint.
To write our own labeling tools, we had to create auto labeling software to increase the efficiency of human labelers because labeling is quite complex.
In the beginning, it took numerous hours to label a 10-second video clip.
This is not scalable.
So, what you essentially need is surround video, which must be primarily automatically labeled with humans just being editors and making little corrections to the labeling of the video and then feeding those corrections back into the future auto labeler, so you end up with this flywheel where the auto labeler is able to take in vast amounts of video and, with great accuracy, automatically label the video for cars, lane lines, and drive space.
What you are saying is...
...
that the result is that you are effectively providing the vehicle with a 3D model of all the objects that surround it.
It recognizes what they are and how quickly they are moving.
The remaining task is to determine what are the eccentric behaviors that, you know, a pedestrian walking down the road with a smaller pedestrian may or may not do unpredictably.
Before you can genuinely refer to it as safe, you must incorporate it into your design.
You must have memory both across time and space.
This is what I imply: Memory cannot be unlimited because it consumes a significant amount of the computer's RAM, essentially.
As a result, you must determine how much you will attempt to remember.
It is rather common for things to become occluded.
So, if we talk about a pedestrian crossing the street in front of a truck, where you saw the pedestrian begin on one side of the truck and then they are hidden by the truck.
You would instinctively know, "OK, that pedestrian is most likely going to pop out on the other side." A computer is unaware of this.
You need to slow down.
A skeptic will say that for the past five years, you have always stated, "Well, no, this is the year, we are confident that it will be there in a year or two or, you know, like it has always been about that far away." But we have a new architecture now, you are seeing enough improvement behind the scenes to make you not certain, but fairly certain, that by the end of this year, the car will be able to drive without interventions more safely than a human in most, not every city, and every circumstance, but in many cities and circumstances.
Yes.
I mean, the car currently drives me around Austin most of the time with no interventions.
So it's not like...
And we have over 100,000 people in our full self-driving beta program.
As a result, you can watch the movies they upload online.
I do.
And some are fantastic, while others are somewhat frightening.
I mean, the car occasionally veers off the road and terrifies people.
It is still in beta.
But you are behind the scenes examining the data and observing sufficient progress to conclude that a this-year timeline is realistic.
Yes, that appears to be the case.
I mean, we could be talking again next year, like, another year went by, and it didn't happen.
However, I believe this is the year.
So, in general, when people talk about Elon time, I mean it sounds like you can't just have a general rule that if you predict something will be done in six months, we should actually anticipate it will take a year or it is like two-x or three-x, it depends on the type of prediction.
Some things, I guess things involving software, AI, whatever, are fundamentally harder to predict than others.
Is there a component that you actively make aggressive prediction timelines in order to drive people to be ambitious?
Nothing is completed without it.
Well, I believe that in terms of internal deadlines, we should set the most ambitious deadline possible.
Because there is a kind of law of gaseous expansion where, for schedules, where whatever time you set, it will not be less than that.
It is incredibly uncommon for it to be shorter than that.
However, as far as our forecasts are concerned, what often happens in the media is that they will report all of the incorrect ones and disregard all of the correct ones.
Or, you know, when writing an article about me—I've had a long career in numerous industries.
If you list my sins, I'll sound like the worst person on Earth.
However, if you compare those to my accomplishments, it makes far more sense, you know?
As a result, the longer you do anything, the more errors you will accumulate over time.
Which, if you add up those mistakes, will give the impression that I am the worst predictor in history.
For example, for Tesla vehicle growth, I stated that I believe we would achieve 50%, and we have achieved 80%.
Yes.
However, they do not mention that one.
As a result, I'm not sure what my track record for predictions actually is.
They are more optimistic than pessimistic, but they are not all optimistic.
Some of them are probably exceeded more or later, but they do come true.
It is extremely rare for them not to materialize.
It is similar to a radical technology prediction, in that the point is not that it was a few years late, but that it occurred at all.
That is the more significant aspect.
So, it seems like at some point in the last year, seeing the progress on understanding, the Tesla AI understanding the world around it, led to a kind of, an aha moment at Tesla.
Because you recently truly surprised people when you said that the most important product development at Tesla this year is probably this robot, Optimus.
Yes.
Many companies out there have attempted to develop these robots and have been working on them for years.
However, no one has yet mastered it.
There are no widely used robots in people's homes.
There are some in manufacturing, but I'd argue that no one has truly cracked it.
Is the development of full self-driving what gave you the confidence to say, "You know what, we can do something special here?
" Yeah, that's it.
So, you know, it took me some time to realize that in order to solve self-driving, you really needed to solve real-world AI.
And at the point at which you solve real-world AI for a car, which is really a robot on four wheels, you can then generalize that to a robot on legs as well.
The two hard parts I think are -- like obviously companies like Boston Dynamics have shown that it's possible to make quite compelling, sometimes alarming robots.
Right.
You know, so from a sensors and actuators standpoint, it's certainly been demonstrated by many that it's possible to make a humanoid robot.
The things that are currently missing are enough intelligence for the robot to navigate the real world and do useful things without being explicitly instructed.
So the missing things are basically real-world intelligence and scaling up manufacturing.
Those are two things that Tesla is very good at.
And so then we basically just need to design the specialized actuators and sensors that are needed for humanoid robot.
People have no idea, this is going to be bigger than the car.
So let's dig into exactly that.
I mean, in one way, it's actually an easier problem than full self-driving because instead of an object going along at 60 miles an hour, which if it gets it wrong, someone will die.
This is an object that's engineered to only go at what, three or four or five miles an hour.
And so a mistake, there aren't lives at stake.
There might be embarrassment at stake.
As long as the AI doesn't take it over and murder us in our sleep or something.
Right.
(Laughter) So talk about -- I think the first applications you've mentioned are probably going to be manufacturing, but eventually the vision is to have these available for people at home.
If you had a robot that really understood the 3D architecture of your house and knew where every object in that house was or was supposed to be, and could recognize all those objects, I mean, that's kind of amazing, isn't it?
Like, the kind of thing that you could ask a robot to do would be what?
Like, tidy up?
Yeah, absolutely.
Make dinner, I guess, mow the lawn.
Take a cup of tea to grandma and show her family pictures.
Exactly.
Take care of my grandmother and make sure -- Take a cup of tea to grandma and show her family pictures.
Exactly.
Take care of my grandmother and make sure -- CA: It could obviously recognize everyone in the home.
It could play catch with your kids.
Yes.
I mean, obviously, we need to be careful this doesn't become a dystopian situation.
I think one of the things that's going to be important is to have a localized ROM chip on the robot that cannot be updated over the air.
Where if you, for example, were to say, “Stop, stop, stop,” if anyone said that, then the robot would stop, you know, type of thing.
And that's not updatable remotely.
I think it's going to be important to have safety features like that.
Yeah, that sounds wise.
And I do think there should be a regulatory agency for AI.
I've said that for many years.
I don't love being regulated, but I think this is an important thing for public safety.
Let's come back to that.
But I don't think many people have really sort of taken seriously the notion of, you know, a robot at home.
I mean, at the start of the computing revolution, Bill Gates said there's going to be a computer in every home.
And people at the time said, yeah, whatever, who would even want that.
Do you think there will be basically like in, say, 2050 or whatever, like a robot in most homes, is what there will be, and people will love them and count on them?
You’ll have your own butler basically.
Yeah, you'll have your sort of buddy robot probably, yeah.
I mean, how much of a buddy?
How many applications have you thought, you know, can you have a romantic partner, a sex partner?
It's probably inevitable.
I mean, I did promise the internet that I’d make catgirls.
We could make a robot catgirl.
Be careful what you promise the internet.
(Laughter) EM: So, yeah, I guess it'll be whatever people want really, you know.
What sort of timeline should we be thinking about of the first models that are actually made and sold?
Well, you know, the first units that we intend to make are for jobs that are dangerous, boring, repetitive, and things that people don't want to do.
And, you know, I think we’ll have like an interesting prototype sometime this year.
We might have something useful next year, but I think quite likely within at least two years.
And then we'll see rapid growth year over year of the usefulness of the humanoid robots and decrease in cost and scaling up production.
Initially just selling to businesses, or when do you picture you'll start selling them where you can buy your parents one for Christmas or something?
I'd say in less than ten years.
Help me on the economics of this.
So what do you picture the cost of one of these being?
Well, I think the cost is actually not going to be crazy high.
Like less than a car.
Initially, things will be expensive because it'll be a new technology at low production volume.
The complexity and cost of a car is greater than that of a humanoid robot.
So I would expect that it's going to be less than a car, or at least equivalent to a cheap car.
So even if it starts at 50k, within a few years, it’s down to 20k or lower or whatever.
And maybe for home they'll get much cheaper still.
But think about the economics of this.
If you can replace a $30,000, $40,000-a-year worker, which you have to pay every year, with a one-time payment of $25,000 for a robot that can work longer hours, a pretty rapid replacement of certain types of jobs.
How worried should the world be about that?
I wouldn't worry about the sort of, putting people out of a job thing.
I think we're actually going to have, and already do have, a massive shortage of labor.
So I think we will have ...
Not people out of work, but actually still a shortage labor even in the future.
But this really will be a world of abundance.
Any goods and services will be available to anyone who wants them.
It'll be so cheap to have goods and services, it will be ridiculous.
I'm presuming it should be possible to imagine a bunch of goods and services that can't profitably be made now but could be made in that world, courtesy of legions of robots.
Yeah.
It will be a world of abundance.
The only scarcity that will exist in the future is that which we decide to create ourselves as humans.
OK.
So AI is allowing us to imagine a differently powered economy that will create this abundance.
What are you most worried about going wrong?
Well, like I said, AI and robotics will bring out what might be termed the age of abundance.
Other people have used this word, and that this is my prediction: it will be an age of abundance for everyone.
But I guess there’s ...
The dangers would be the artificial general intelligence or digital superintelligence decouples from a collective human will and goes in the direction that for some reason we don't like.
Whatever direction it might go.
You know, that’s sort of the idea behind Neuralink, is to try to more tightly couple collective human world to digital superintelligence.
And also along the way solve a lot of brain injuries and spinal injuries and that kind of thing.
So even if it doesn't succeed in the greater goal, I think it will succeed in the goal of alleviating brain and spine damage.
So the spirit there is that if we're going to make these AIs that are so vastly intelligent, we ought to be wired directly to them so that we ourselves can have those superpowers more directly.
But that doesn't seem to avoid
