Robots That Jump

Robot Bodies Needed Before Robot Minds

Monthly Archives: November 2017

Too bad there wasn’t a beheading…

Even Forbes, which should know better, has its lame reporters reporting on the “mind” of Sophia, the first “robot citizen” of Saudi Arabia.

https://www.forbes.com/sites/zarastone/2017/11/07/everything-you-need-to-know-about-sophia-the-worlds-first-robot-citizen/#29420af346fa

This big electric puppet, with some small “ai” features useful in tricking children into thinking it is alive, is claimed to have mind and emotions.

After characterizing the robot as a publicity stunt, the article then moves on to beauty contest questions, e.g. “world peace”.

In practice, there isn’t any Ai on earth that could have reacted to the questions aimed at the robot during press conferences in an intelligent fashion. Instead, the robot uses ELIZA style reflection to appear intelligent (“what makes you think I am thinking about that”) along with canned answers that are supposed to seem wise. Apparently, the conservatives in Saudi Arabia don’t mind an idol-esque oracle being set up to answer our “deepest questions”. And since oracles are often blown out on drugs or gas seeping through cave walls, mayhaps that makes sense.

https://en.wikipedia.org/wiki/Sophia_(robot)

Business Insider has done a few articles that point out the clumsy voice recognition. But it makes a fascinating claim:

Sophia’s capacity for displaying emotion is still limited. It can show happiness — sort of.

Sophia fakes it

 

I put this image in because, to the robot (presumably a couple of Linux boxes) the nature of the “happiness” has been mapped directly to a collection of actuator movements. That, in turn implies that human happiness is simply a collection of muscles firing in a way to stretch the face. The amazin thing is that there’s no concern about a subjective – nobody wonders if Sophia actually feels “happy” or “sad” internally. In fact, the robot makes a face when something (e.g. a keyword) causes a sub-program to execute.

Whatever human emotion is, it is unlikely to be a Von Neuman machine ratcheting through alternative subprograms. The brain’s structure is so different from a computer that its is nearly impossible that this be true.

In Ai, however, analogy is often used to refute the argument. Flight, for example can be achieved by bird or insect wings in completely different ways. So the argument goes that emotions can be created in similarly analogous ways.

The problem is that nobody has adequately defined what “happy” or “sad” are in the first place so that we can construct analogous ways of achieving it. The argument that Sophia is “happy” is the same as saying a corpse with its face stitched into a smile is “happy” if we can make that smile by attaching electrodes to the muscles – from Wikipedia entry:

Giovanni Aldini (Luigi’s nephew) performed a famous public demonstration of the electro-stimulation technique of deceased limbs on the corpse of an executed criminal George Foster at Newgate in London in 1803.[5][6] The Newgate Calendar describes what happened when the galvanic process was used on the body:

On the first application of the process to the face, the jaws of the deceased criminal began to quiver, and the adjoining muscles were horribly contorted, and one eye was actually opened. In the subsequent part of the process the right hand was raised and clenched, and the legs and thighs were set in motion.[7]

I doubt if a modern electrophysiologist would conclude that a recently dead body was “happy” if the muscles twist. But replace that flesh with plastic, the electrical stimulus with a command out the serial port, and it is “feeling”.

Hoping that Hanson Robotics studies Behaviorist theory before making more silly claims about puppets.

 

Advertisements

Finally, a ROBOT THAT JUMPS!

I’ve been writing about the (incorrect) direction of much of robotics, along with the fakery which implies robots are more than they are for 15 years. The alternative I’ve pushed is building an agile body over “Ai” or “mind” popular with almost everyone. In my opinion, a mindless body that is physically agile – one that can reproduce the things even simple animals can do on a daily basis – is vastly more useful than a so-called “intelligent” robot.

Well now, Boston Dynamics, which has followed the “body first, then mind” approach, has finally come through. The newest version of the Atlas robot can jump and even do backflips in what looks to be an agile way:

The results are impressive. This is a huge, heavy metal robot showng agility. Small motions of the arms are being used to stabilize, which makes it look like a pretty complex control system. Though I am guessing this is pretty staged (change the box positions and the robot falls), it is a candidate for a true Robot that Jumps.

This follows on other robots designed to imitate living thins, in pursuit of agile behavior.

While these robots are finally meaningful (in terms of being real robots instead of fakey electric puppets), they are far from practical. The biggest issue is power supply – they use IMHO 20-100x times as much power as a comparable animal body to move. Electric battery density simply isn’t high enough for a human-sized robot to remain powered for more than a few minutes at a time. At present, you would probably need to use nuclear power to move these robots for reasonable operations – or big Diesel engines.

The other issue is muscles. Robots using electric motors aren’t very agile. Those using pneumatics can achieve lifelike motion, but it is impractical to power them standalone. I’m guessing the gasoline engines on the early BD robots supplied compression for the leg actuators – NOISY. These are not quiet, like the would be if a human did the same motions. The holy grail – elastic plastic and contractile muscle tissue – is a long way off. You see visions of such muscles in movies like Ghost in the Shell, but they don’t actually exist.

https://vitalybulgarov.com/ghost-in-the-shell/

Ghost in the Shell - major shelling sequence

The closest we have is something like this combined electro (heating?) hydraulic muscle:

https://www.sciencedaily.com/releases/2017/09/170919091000.htm

It took 30 years for walking robots to learn to jump, good muscles may easily take that long.

I’m guessing another 50 before something like Ghost in the Shell could be built.

However, a mental point might have been reached. With the debut of the BD robot, people may be less accepting of the stupid fake robots pedaled at SF conventions as somehow “real” or “the future”.  Mayhaps we’ll start ignoring the “mind without body robot”, typlified by the creepy sexist machines popular in the media:

 

In a word…this is a big fake puppet implying human abilities that don’t exist. And this machine can’t jump.