Self-Driving Concept Car Being Unveiled At CES Triggers Anew The Feud Over Levels Of AI Autonomy

Self-Driving Concept Car Being Unveiled At CES Triggers Anew The Feud Over Levels Of AI Autonomy
New concept driverless car sparks autonomy debate. Getty

Quick, how many senses do humans have?

Most of us would rattle off the five senses of hearing, seeing, tasting, touch, and the sense of being able to smell.

Everybody knows that.

So, your answer presumably is five.

There are some though that claim we have a sixth sense, perhaps consisting of intuition or an ability to have premonitions.

Okay, the answer is apparently six.

Actually, maybe there are seven senses since some assert that we have a sense of space, meaning that you are able to feel where your body resides within your space or area of movement, a sense often referred to as proprioception.

Which is it, do we have five senses, or six senses, or seven senses, or are there even more senses and could we have nine or ten of them?

The debate continues to rage on about the number of senses that humans have and right now there’s no final answer, though the commonly accepted belief is that we have five senses.

There is a similar type of debate in the self-driving car industry regarding how many levels of autonomy can be best delineated.

Currently, the existing standard that is accepted by-and-large by self-driving car pros is that there are six levels of autonomy, which have been defined and promulgated by the Society of Automotive Engineers (SAE) and in conjunction with numerous other standards bodies.

· Level 0: No Automation

· Level 1: Driver Automation

· Level 2: Partial Automation

· Level 3: Conditional Automation

· Level 4: High Automation

· Level 5: Full Automation

Notice that the levels are numbered from 0 to 5 (be careful interpreting the wording that is associated with each level, since those short phrases do not necessarily well encapsulate the full embodiment of each level as strictly defined by the standard).

Some don’t like the aspect that the levels start with the number 0, since it becomes awkward to readily indicate how many levels there are. The topmost level is numbered 5, and thus some people mistakenly believe there are only five levels of autonomy.

If the numbering began with the number 1, you’d have levels 1, 2, 3, 4, 5, 6, and this would then be a more immediate tell that there are six levels of autonomy.

The counterargument is that the first of the several levels consists of no autonomy, therefore it deserves being considered simply as a level of zero.

If you were to give it the number of one as its level, the argument goes it would imply that there is some non-zero amount of autonomy included at this rock bottom level.

Plus, of course, computer people love to start things with the number 0, which though this doesn’t often sit well with the general populace, as they say when you have the power of the pen you are able to write the rules.

Another occasional flare-up is that the numbers should be abandoned and letters ought to be used instead, suggesting that the levels would be the letters of the alphabet, presumably the letters a, b, c, d, e, f, though this has not caught any attention and pretty much is dead in the water.

Anyway, for the moment, put aside the acrimonious mudslinging about whether the levels should continue to be counted from 0 to 5 or be something else such as 1 to 6 or possibly a set of letters in the alphabet.

The key here is that just about everybody agrees that there are six levels of autonomy and somewhat amicably go along with the SAE standard accordingly.

Aha, there is always room for contention.

Not everyone does believe that there are just six levels of autonomy for driverless cars.

Some believe that there are more levels, including some are pining away for 7 levels, some want 8 levels, and others have tried to get us to 9 or 10 levels.

If you are wondering whether anyone wants to reduce to say 5 levels, it’s a rare argument, though on a related tangent, due to the 0 to 5 numbering, there are those that say you shouldn’t count the 0 level and therefore they assert that there are really only 5 legitimate levels.

Say what?

Let’s move on.

When I indicate that some are arguing for making the standard be more than 6 levels, I’d like to exclude the obnoxious cranks that generate and fabricate fake news about self-driving cars and have no idea what they are talking about.

In other words, the media regrettably at times will state that there are some umpteen numbers of levels and do so with ignorance about the SAE standard. There are some that don’t know much about self-driving cars and will utter whatever seems to come to their minds.

We’ll put aside the suggestions about being more than 6 levels that are proffered by those that aren’t in the know, and for whom only say such things from the side of illiteracy or by the wanton act of wanting to seem controversial or outlandish.

Stick with the serious cases of wanting to expand the number of levels and please toss into the junk heap those worthless foolhardy ones.

Why bring this topic up at all?

Because Honda has just released a teaser for their upcoming reveal at the tech haven of CES (Consumer Electronics Show), taking place in Las Vegas this coming January, about a new self-driving concept car that they say showcases eight levels of autonomy.

Yes, go ahead and read that again, they are going with 8 levels of autonomy, rather than the SAE standard, which is widely accepted and usually agreed upon, consisting of 6 levels of autonomy.

The teaser and apparent approach of seeking to promote eight levels of autonomy has already rankled the self-driving car industry and sparked anew the debate about the number of levels of autonomy.

Honda also mentions that they are aiming at allowing a human driver to switch readily to driverless mode, and driverless mode to switch to a human driver, which has numerous downsides and remains a queasy idea within the self-driving tech industry (see my explanation at the link here).

You might be thinking that it is a tempest in a teapot and it doesn’t matter how many levels of autonomy one wishes to define.

Oh, but you’d be missing the bigger picture if you believed that.

There are quite serious and important reasons to not be willy-nilly about the numbers of levels of autonomy.

Let’s unpack the matter.

The Levels Of Self-Driving Cars

It is important to clarify what I mean when referring to self-driving cars.

True self-driving cars are ones that the AI drives the car entirely on its own and there isn’t any human assistance during the driving task.

These driverless vehicles are considered a Level 4 and Level 5, while a car that requires a human driver to co-share the driving effort is usually considered at a Level 2 or Level 3. The cars that co-share the driving task are described as being semi-autonomous, and typically contain a variety of automated add-on’s that are referred to as ADAS (Advanced Driver-Assistance Systems).

There is not yet a true self-driving car at Level 5, which we don’t yet even know if this will be possible to achieve, and nor how long it will take to get there.

Meanwhile, the Level 4 efforts are gradually trying to get some traction by undergoing very narrow and selective public roadway trials, though there is controversy over whether this testing should be allowed per se (we are all life-or-death guinea pigs in an experiment taking place on our highways and byways, some point out).

For Level 4 and Level 5 true self-driving vehicles, there won’t be a human driver involved in the driving task. All occupants will be passengers. The AI is doing the driving.

For semi-autonomous cars, it is important that people be forewarned about a disturbing aspect that’s been arising lately, namely that in spite of those human drivers that keep posting videos of themselves falling asleep at the wheel of a Level 2 or Level 3 car, we all need to avoid being misled into believing that the driver can take away their attention from the driving task while driving a semi-autonomous car.

You are the responsible party for the driving actions of the vehicle, regardless of how much automation might be tossed into a Level 2 or Level 3.

Self-Driving Cars And Levels Debates

In the case of the Honda teaser, they have not yet stated what the 8 levels of autonomy consist of.

Guess we’ll need to wait and see what they unveil at CES.

Is this then a clever marketing ploy to have floated the notion of eight levels and keep us all breathlessly awaiting the reveal?


Though there is a drawback to such a ploy.

For those that are already exasperated by the ongoing debate about the number of levels, they see this teaser as yet another attempt to make troubles where no such troubles ought to be.

It is, as the old saying goes, a needless poking of the eye of a bear.

Pros that have been down this trodden path too many times are apt to wearily ask: Do we really need to revisit the levels of driverless autonomy again and again and again?

Look, it’s six levels, they cry out, and stop rattling the cage.

Worse too, if the eight levels are potentially untoward, the concern is that we’ll once again have the overall media pick-up on the topic and create more confusion among the public and regulators.

It could be that the eight levels are not especially useful or even usable as a means of structuring the levels of driverless autonomy and sadly, regrettably, might foster confusion and angst.

Throughout history there have been many now-famous (or infamous) marketing schemes that ended-up turning themselves upside down and inadvertently started a storm rather than calming the seas.

The law of unintended consequences.

Time will tell in this case.

Beyond Six By Splitting Hairs

Well, let’s nonetheless consider the ways that the ongoing debate about six levels versus more levels tends to arise and why it does.

Where’s the beef, you might be asking.

One of the most common ways to go beyond the existing six levels of self-driving autonomy consists of seeking to blowout the existing levels.

Here’s what that means.

For Level 2, which many would suggest that Tesla’s with AutoPilot are currently at, some try to say that the Tesla is really at a level of 2.5, meaning that they acknowledge it is not yet at level 3, but they also argue that it is much more than merely level 2 in terms of being at the floor of level 2.

As such, this same kind of logic has been used for the other levels too.

Some believe that there should be a level 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5.

Note that this would add four additional levels (shown as 2.5, 3.5, 4.5, 5.5), assuming that you were willing to call the intermediary stages as levels (which some argue you should not do).

If you added those four to the existing six, you’d have a grand total of ten levels.

There is controversy about this, including the aspect that if you are going to have a level 5.5, shouldn’t you also have a level of 6 (adding even one more level to the pile and making it into 11 levels)?

This argument about having intermediary levels was seemingly already settled due to the aspect that the SAE standard clearly states that there is no such thing as an intermediary level.

The rules generally are:

  • If you have automation that fulfills a level, the car is then ranked or scored as being at that level.
  • If the automation does not fulfill a level, it is not considered as ranked or scored at that level.

Case closed.

Despite the stated rules, there are still some that argue we need to have intermediary states in the autonomy structure.

Those making such an argument then fall into one of two camps, namely you can refer to the intermediary states as sub-levels and not count them as honest-to-goodness levels, or you ought to count them as full-fledged levels (in which case, perhaps renumber all of the levels and make it into the numbers of 0 to 10, or 1 to 11, or however you want).

Some would say that doing so is merely rearranging the chairs on the deck of the ocean liner.

Sure, it might be helpful to have those sub-levels clearly identified, and it might then be easier for automakers and self-driving tech firms to assert more definitively what level of automation they are providing in their car.

But, it is an argument that some say opens a can of worms.

If we allow for those intermediary states, it could be a snowball that beings to roll down the hill.

After having available a 2.5, someone might come along and insist that we should have a 2.7 too.

Someone else clamors for a 2.9.

And so on.

Once you begin to splinter the levels, it could become a never-ending game.

Don’t do it, we are warned, since this splitting of hairs will become unwieldy and be impossible to communicate succinctly to the public and regulators, potentially making a mockery of the autonomy scheme.

Tantamount to excessive hair-splitting.

Adding Levels Rather Than Splitting Levels

For the moment, step away from the sub-level splitting aspects.

Is there anything that might be missing from the existing levels and for which a case could be made to expand the number of levels?

Some say yes, there are such cases.

First, keep in mind that the existing SAE standard indicate that the standard applies only to on-road driving.

That makes sense in that most of the time we are driving our cars on conventional streets and paved roads.

Maybe there should be a level 6, consisting of autonomy for off-road driving, and thus expand the number of levels to seven (ranging from 0 to 6).

Secondly, the existing SAE standard refers to human driving as based on being able to drive a car in user manageable ways.

Some have criticized the lack of clear-cut meaning to what human driving consists of, since how I drive and how someone else drives could be quite different in terms of driving skill levels. A racetrack or Indy driver presumably knows a lot more about driving a car than you or I do (for my exhortation for the importance of high-performance driving AI skills, see this link here).

Should driverless cars be held to being able to drive as could the best of human drivers, or be allowed to drive at the level of an average human driver (I’ve made the case that right now the aim is even lower, pretty much at the level of a teenage novice driver).

And, you could argue that we ultimately would want the AI to drive better than any human driver could, surpassing all human driving skills (something often referred to as a form of superhuman AI, though this is a phrase I discourage from being used, see the link here).

It seems like a laudable goal to have the AI be beyond human driving skill-sets.

We might then add a level 7, which consists of autonomy that can drive a car better than any human proficiency, including better than the best-of-the-best of race car drivers.

That’s now 8 levels of autonomy.

Sorry, no clue yet as to whether that’s the same as the Honda reveal planned for CES (probably not, the potential off-road and “superhuman” additions are askew of today’s conventions).

One complaint about both the off-road addition and the mighty-driver notion is that rather than expanding the levels, we might be wiser to infuse those elements into the existing levels.

In essence, no need to toss out the six levels, and instead aim to tweak and revise them, encompassing whatever additional sauces or flavoring we want to include.

You could readily argue that by infusing any new elements you will produce a more parsimonious result.

Maybe, or you might imprudently be trying to put eight pounds of rocks into a six-pound bag.


I’ve not yet fully explained why this all matters.

It’s a big deal because we need to have an agreed to and in-common parlance to describe what the levels of autonomy are.

Suppose an automaker proudly proclaims that they have a self-driving car that has achieved level X, but there is no definitive meaning for what level X is. Meanwhile, another automaker claims their self-driving car is at a level of X+1, implying that their driverless car is better than the other automaker’s self-driving car.

In the end, without a referee and some set of rules, it would be like playing a football game that hasn’t established how many points you get for a touchdown.

Nobody would readily know what the score is and how each of the teams is doing.

Right now, we have an existing standard that’s a generally accepted barometer, delineating six levels of autonomy for driverless cars, and it is a line-in-the-sand for automakers, self-driving tech firms, regulators, the public, and all other stakeholders that care about autonomous and semi-autonomous cars.

If anyone and especially someone in-the-know opts to craft altogether new or additional levels, on their own, and not with an intent on collaborating toward a new or revamped standard, doing so will certainly spur turbulence and a hefty kerfuffle.

Pointedly, it’s not that we should freeze in place the existing standard and never seek to change it, but instead, the notion that let’s go ahead and consider ways to improve and upgrade the standard, working in concert with the appropriate standards bodies and associations.

That sure seems like a means to have a level playing field, a multi-leveled playing field.


What do you think?

486 points
Upvote Downvote

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

UD explores energy efficiency through connected and automated vehicles

5 ways AI can make your car journeys less boring — and more safe

5 ways AI can make your car journeys less boring — and more safe