Physical AI

[JeffKeryk]

Perhaps some of you have seen the discussion @OVERKILL and I have had around "Physical AI".

What is this thing? From the Google:
"Physical AI involves machines using sensors to perceive and interact with the real world, employing AI techniques like computer vision and reinforcement learning to understand physics, react to environments, and perform physical tasks through actuators. Unlike digital AI, physical AI creates a closed-loop system where the system makes decisions and directly affects its physical environment, enabling applications such as autonomous vehicles, humanoid robots, and smart factories."

If you have any interest, here is a video from Morgan Stanley surrounding Physical AI. I think you might enjoy it and perhaps get you thinking...
Team, this is beyond thick vs thin. Way beyond.

 

[ctechbob]

Let's hope Asimov's 3 rules are onboard.

 

[wwillson]

Investing in robot tech and manufacturing companies could be the next big thing. Hold on to your hats, they are coming and fast.

 

[JeffKeryk]

Nvidia CEO Jensen Huang stated, "The next wave of AI is physical AI," and that this wave will involve "AI that understands the laws of physics, AI that can work among us". He described this new form of AI as possessing physical world common sense, such as understanding object permanence, friction, and cause and effect, enabling robots and autonomous systems to interact with and navigate the real world effectively.

I read somewhere Huang describes Physical AI something like the difference between AI and a dog is, if you roll a ball off the table, AI may have no sense of where it went; it in some Metaverse? A dog will recognize the table, deal with it and go around it to find the ball.

Physical AI is the dog.

 

[SonofaBoomer]

Just wait until it starts scheming to defeat it's predefined limits. In my view, impossible to avoid when military purposes are driving the tech, also inevitable.
https://techcrunch.com/2025/09/18/openais-research-on-ai-models-deliberately-lying-is-wild/

 

[PandaBear]

Got a presentation and a good meal from a chip company last week and one of their showcase is about physical AI. Without getting involved with NDA the general concept of physical AI is that, instead of using a lot of math to do actual simulation that is too expensive in calculation, AI will do educated guess in a way that is more "common sense" based on those training it has gotten before.

So, will it improve the world incrementally? Absolutely. Will it make the impossible happen from what we haven't done before? I don't think they are completely new like how FFT has changed wireless communication or floating gate leads to flash memory and changed digital camera.

BTW, they are renaming some of the chips that would have been called DSP into AI now. I was confused when they were presenting why AI module was significantly improving speed in some stuff until they start showing me how they were just doing FFT and to use those feature they have to limit to 32 or 64 bits, instead of custom stuff. Then suddenly I realized AI chip is just doing a bunch of multiply and add and they are literally the same as what we called DSP 2 decades ago...

All the hard things are in the training. Those chips are still really amazing though, literally what a whole rack of servers now down into one chip to support a whole cell tower, inside the antenna. I have also seen the presentation using Llama 3 locally and, without network connection, able to generate things on a workstation as operator type without waiting to finish the typing.

 

[PandaBear]

Just wait until it starts scheming to defeat it's predefined limits. In my view, impossible to avoid when military purposes are driving the tech, also inevitable.
https://techcrunch.com/2025/09/18/openais-research-on-ai-models-deliberately-lying-is-wild/

They have to outsmart Palentir on that. I highly doubt Open AI would win that arm race.

 

[Carmudgeon]

A guy whose primary interest isn't in pumping stocks disagrees.

Knows a thing or two about robots.

 

[alarmguy]

I think the bottom line is the POSSIBILTY we will go through a dangerous time when the only time you know you are talking to a person giving you real "safe" information is when they are right in front of you.
An example might be using online banks. We might one day feel more secure going to physical banks again as an example. Otherwise how might you know, the bank you think you are communicating with is actually the bank and not an online spoof of a bank stealing your money?
That is just an example, I would also think there will be the AI police, policing the internet for bad actors.

Another danger that can't be ignored is control over you. This example is only today, imagine the future.
In the simplest form in today's world. Tesla can deny you charging your vehicle. Every time you hookup your car to one of their chargers, they know who you are, where you are, what you are doing, including cameras all around the car. This is only the beginning.

This just happened on a Tesla vehicle. The perfect condition 2022 Model 3 someone bought used, was denied charging on Tesla's chargers because the vehicle had a Salvage Title. Tesla felt it might be a danger.

 

[Vil_Riker]

These type of robots will debut in the military arena (they actually might have already).

 

[alarmguy]

These type of robots will debut in the military arena (they actually might have already).

Yes, I know for fact we have robotic "dogs" at some installations. Unarmed that we know of right now.

 

[JeffKeryk]

These type of robots will debut in the military arena (they actually might have already).

Drones...

 

[SC Maintenance]

The issue is in the power transmission.

They tried to figure out how much it would cost to build a robot to clean a hotel room. The problem is that making a bed and washing a mirror require very different mechanics. They figured they would need 6 different robots. Wish I had saved that study - was a few years old.

Same with self driving cars. Driving the car part is easy. The sensing tech isn't there yet - not at an affordable cost.

And of course no one will accept a robot that is only as good as a human. A self driving car with as many errors as a human would be considered a dismal failure.

Now robots purpose built for specific tasks already exist and will likely get much better.

 

[JTK]

Will they play nice on BITOG though..

 

[JeffKeryk]

Will they play nice on BITOG though..

The bar is low... Ha!

 

[Skippy722]

This will work great.... Until they hit the grungy blue collar jobs. Then it'll be one robot will be on divorce 3, another will have $1200 in child support every week, ones gonna smoke 3 packs a day.... And despite all 3 taking turns car pooling they'll be late everyday from lunch because "someone was buying lotto tickets at our favorite gas station and we needed cigs!"

 

[eljefino]

I think the bottom line is the POSSIBILTY we will go through a dangerous time when the only time you know you are talking to a person giving you real "safe" information is when they are right in front of you.
An example might be using online banks. We might one day feel more secure going to physical banks again as an example. Otherwise how might you know, the bank you think you are communicating with is actually the bank and not an online spoof of a bank stealing your money?

How do we know that now? I put a sizeable sum in a HYSA in a bank I found via some online rate comparison; never set foot in a brick and mortar. "Researched" them for about five minutes to find a real street address (in Texas) and that they're an offshoot of a real B&M bank.

As far as talking to a person, they could just be an avatar with an AI backend. Could bring a cellphone frequency jammer but it would probably just switch the bot to "local", like S4E8 of "Star Trek: TNG" and try to BS its way through without the "hive" available for help.

 

[alarmguy]

How do we know that now? I put a sizeable sum in a HYSA in a bank I found via some online rate comparison; never set foot in a brick and mortar. "Researched" them for about five minutes to find a real street address (in Texas) and that they're an offshoot of a real B&M bank.

As far as talking to a person, they could just be an avatar with an AI backend. Could bring a cellphone frequency jammer but it would probably just switch the bot to "local", like S4E8 of "Star Trek: TNG" and try to BS its way through without the "hive" available for help.

I didn’t say we know I said possibility.
I’ve been banking online for well over a decade. I don’t even know the location of two of my banks.

I’m not entirely sure of your response, except the possibility of someone being fooled by AI in the future when they open that bank account and deposit their money. Then find out that the bank really isn’t there. Or you log into your bank thinking you’re really logging into your bank, but you’re not.
AI with criminals in charge will be able to create anything anytime that looks entirely real, including real voices.
There’s no doubt my mind the future holds good AI that is policing bad AI

All possibilities no one knows yet. They also say wars might be over in a matter of seconds between two competing AI from different countries