Dr. Lance Eliot, AI Insider
[Ed. Note: For reader’s interested in Dr. Eliot’s ongoing business analyses about the advent of self-driving cars, see his online Forbes column: https://forbes.com/sites/lanceeliot/]
“Their hands did it.”
That’s what my children told me when they were quite young and had managed to put their hands onto wet paint on a freshly painted fence in our neighborhood.
They had no control over their hands.
Somehow, magically, mysteriously, their hands had decided on their own to touch that paint, without consulting with the rest of their bodies and minds, and that’s how it happened.
It’s a funny story now and one that I remember vividly, while they today as young adults don’t seem to remember it at all.
The Alien Limb Syndrome
What makes the story particularly notable too is that they unknowingly landed on an actual aliment that exists.
There is an actual documented phenomena of people that are unable to control their limbs.
It is typically referred to as the alien limb syndrome.
My kids didn’t actually have it.
For those of you that are movie buffs, you might remember that in the movie Dr. Strangelovethe main fictional character is unable to control his arm and hand, and he flails them uncontrollably around at times, making the character seem grotesque and befitting with the role. We ought to not consider the movie as any kind of supporter for considering this as a serious ailment and a medical condition that we should give careful and due consideration for.
Some people even refer to the alien limb syndrome as the Dr. Strangelove syndrome.
There are also some that refer to this aliment as the alien hand syndrome, though the condition can impact arms, legs, feet, and essentially all of the limbs.
It is not exclusively for the hands, though the hand as the focus does seem to be the more popular affected limb (for those of us left handed, it also seems to be primarily the left hand!).
Brain And Body Entanglement
We can’t know for sure what is happening in the person’s brain. Maybe a part of their brain told their arms and hands to do the act, while another part of the brain was unaware that the other part was acting to do so. It could be that the person is only aware of the part of the brain that was unaware and so they tell us that their brains did not command their arms and hands.
Some assert that alien limb syndrome is a disentanglement of the mind and the body.
The limbs genuinely are acting on their own. The mind is not involved at all.
Some say that the disentanglement is within the mind, causing parts of the mind to become disentangled, such as the part that controls the motor functions of the limbs and the part that does action planning for the body.
Another intriguing element of the alien limb syndrome is that sometimes one limb will seemingly try to purposely counteract the other limb.
Dangers Associated With The Alien Limb
There are obviously dangers involved in having the alien limb syndrome.
Suppose your limb acts up when you least want it to do so.
If you are interested in the alien limb syndrome, there are lots of fascinating studies trying to pin down what causes it and what can be done about it. In a recent study done at Vanderbilt University, researchers seemed to trace the aliment to connections in the brain involving the precuneus. The precuneus is often considered the part of the brain that provides our sense of free will and what is coined our agency.
One key aspect of the study was that there didn’t seem to be one specific area of the brain that could be considered the culprit for the syndrome.
Comparing Alien Limb To Computer Systems
As a seasoned AI developer and software engineer, I’ve had situations involving computer systems that in some analogous way appeared to have been overtaken by an alien limb syndrome.
I remember one time that I was involved in creating a rather complex piece of software that had lots of components. One of those components went awry and intermittently created problems, though it was not readily apparent when and why the software was posing such difficulties.
In a manner of speaking, we had an alien limb syndrome.
One of the “limbs” of the core system had gone alien on us. The core system wasn’t doing it. The limb was acting on its own.
AI Autonomous Cars And Alien Limb Syndrome
What does this have to do with AI self-driving driverless autonomous cars?
At the Cybernetic AI Self-Driving Car Institute, we are developing AI software for self-driving cars. One aspect that AI developers generally should be doing is building their systems to catch and prevent an alien limb syndrome from overtaking the rest of their AI system.
This is especially crucial in a real-time system and really especially so in a real-time system that controls a self-driving car — there can be serious life-or-death consequences for an “alien limb” acting up in an AI self-driving car.
Sensors Go Awry As Alien Limbs
Let’s start with the potential of the sensors to become one or more alien limbs.
Suppose one of the cameras on an AI self-driving car starts to go rogue.
Rather than providing images at a particular pace as established by the AI system, the camera instead begins to generate tons of images.
If the rest of the AI system is not prepared in-advance to handle this kind of alien limb activity, it could cause quite a problem.
The AI effort to make sense of the images as to whether there is a car ahead or a pedestrian in the way might be marred by getting the images on such an unexpected basis. If the AI is fooled into believing what the camera is providing, it could lead to an internal cascading set of errors and confusion.
For example, during sensor fusion, if we assume that say the radar and LIDAR are still functioning properly, there will now be a potential contention between what the camera indicates and what those other sensors indicate. Which of the sensors is to be believed by the AI? If the AI falsely assumes that the camera is correct, it could attempt to override the radar and LIDAR, or might give the radar and LIDAR less weight in trying to ascertain the surroundings.
Virtual Model And Alien Limb
Suppose that the miscues then were passed along to the updates of the virtual world model.
Maybe the virtual world model places a marker that there is a pedestrian on the sidewalk when the actual fact is the pedestrian is standing in the street.
When the AI action planning kicks in, it will inspect the virtual world model and falsely get an indication that there isn’t a pedestrian in the way. The AI action planning might opt to issue car commands that tell the self-driving car to continue forward at the ongoing speed, even though the self-driving car is now moving closer and closer to hitting the pedestrian that’s in the street.
Tricky Aspects Of Alien Limb Of A Self-Driving Car
In other words, the error detection for most sensors would be that the camera is not working at all, maybe due to having encountered an outright hardware failure or maybe it got smacked by a piece of debris that flew up from the street and cracked or broke the camera.
In the alien limb syndrome notion, I’m saying that let’s assume the camera is otherwise working just fine and it is now working on its own and not necessarily at the command of the rest of the AI system.
In that manner, the error detection by the sensor itself might not even realize that the sensor has gone rogue.
The usual error detection involves that the sensor has blurry images or no images, while I’m suggesting in the alien limb manner the images are overall fine. This is the same as when a human with alien limb syndrome suddenly has their arm and hand act up, namely that the arm and hand are working as an arm and a hand and there is nothing wrong with those appendages per se (i.e., the arm still extends, the hand still grasps), and they function as an arm and a hand are expected.
It is imperative that the AI be prepared for realizing that a sensor, any of the sensors, might suddenly go rogue.
It might involve other functionality of the sensor too.
For example, suppose the camera can be automatically adjusted to focus on nearby objects or far away objects. Let’s suppose that the AI has recently set the camera on the far away focus. During the rogue action, the camera might on its own switch into the nearby focus and provide those images. The AI was expecting far away images and meanwhile it suddenly gets nearby images. Would the AI be able to detect this? That’s the million-dollar question, as they say.
The other complication is that the rogue act might be fleeting rather than consistent.
Detecting An Alien Limb Scenario In Real-Time
Detecting the alien limb is crucial.
Having a means to deal with the detected rogue acts is equally crucial.
If a flood of images are pouring in, the AI would need to ascertain which images to keep and which to potentially discard. You might be thinking that shouldn’t it analyze all of the images received? Well, keep in mind that as a real-time system it takes time for the system to analyze each image, and it could be that the system will fall behind if it merely opts to analyze every image being flooded into the system.
Imagine that the flood caused a backlog of the image analyzer.
Meanwhile, the self-driving car is still moving ahead.
The delay in the sensor analysis of the backlog might mean that a few crucial seconds are lost that might have made the difference in terms of the AI action planner realizing that the self-driving car is going ram into a pedestrian or another car.
Beyond the sensors, an alien limb can strike other aspects of the AI self-driving car system.
Perhaps the sensor fusion goes rogue.
Maybe the virtual world model goes rogue and starts populating the model with all sorts of markers that aren’t based on what the sensors and sensor fusion have reported.
The AI action planner itself might go rogue.
Generally, the deeper within the AI system that the alien limb strikes, the harder it will be to ascertain and deal with.
Preparing For Alien Limb Instances
I’ve mentioned that the AI system needs to be prepared in-advance for alien limb syndromes that might arise.
Some AI developers might balk at this notion that they need to develop the AI system to cope with rogue behavior and offer instead that the AI system should by itself be able to deal with alien limbs.
I’d vote that since we’re dealing with an AI self-driving car, and since there is a solid chance that the AI self-driving car could cause undue damage or injury, it makes a lot more sense to prepare the AI system beforehand, rather than hope or assume that the AI will somehow miraculously figure out what to do.
There is a slim chance that some kind of Machine Learning (ML) or Deep Learning (DL) capability of the AI system for the self-driving car might be able to identify that something is amiss, and maybe gradually figure out that it is rogue behavior. This might though take many iterations and it is usually the case with today’s ML and DL that tons of examples are needed to find patterns.
I don’t think we want the alien limb acts to mount up and instead want to catch them as soon as they arise, thus, waiting for the off-chance that the ML or DL might catch on is not a good strategy for safety purposes.
Further Twists On The Alien Limb
There are some added twists to consider.
One is a false positive effect.
The AI might falsely accuse a capability to be suffering from alien limb, and yet perhaps the capability is actually functioning properly and appropriately. The danger is that if the AI has opted to now perhaps disregard the limb or otherwise treat it as suspect, whether it is a sensor or some other component, the AI is doing so falsely.
Thus, the AI might be incorrectly now be overriding or wrestling with a component that actually is able to work fine.
The other side of this coin is a false negative indication.
The AI might somehow inspect or assess a capability and determine that it is not suffering from an alien limb syndrome, and yet the capability actually is. In other words, we cannot assume that the AI system is going to necessarily always correctly discern when an alien syndrome is occurring.
There are some AI developers that are oblivious to the alien limb syndrome when it comes to their AI systems.
It is their assumption that the components of their AI system are going to work correctly and that within them is some kind of self-error checking. Therefore, the rest of the AI system does not need to be concerned about the component because the component itself will let the rest of the AI system know when it is not working properly.
For an AI self-driving car, any of the “limbs” of the AI system can wreak havoc if it opts to activate whenever it opts to do so.
Let’s not have an AI self-driving car that suddenly opts to swerve the car unexpectedly or slam on the brakes, doing so in the manner that a human might uncontrollably button or unbutton a sweater, all of which arises due to an alien limb syndrome.
Properly developed AI systems for self-driving cars need to be prepared for detecting and acting upon an alien limb and do so quickly and prior to allowing an alien limb to cause an untoward action.
For free podcast of this story, visit: http://ai-selfdriving-cars.libsyn.com/website
The podcasts are also available on Spotify, iTunes, iHeartRadio, etc.
More info about AI self-driving cars, see: www.ai-selfdriving-cars.guru
To follow Lance Eliot on Twitter: https://twitter.com/@LanceEliot
For his Forbes.com blog, see: https://forbes.com/sites/lanceeliot/
For his Medium blog, see: https://email@example.com
For Dr. Eliot’s books, see: https://www.amazon.com/author/lanceeliot
Copyright © 2019 Dr. Lance B. Eliot