Firstly I should point out that I am not an expert in machine learning, but I think I know enough to give you a simple explanation. Because at its core, Machine Learning is simple.

It is important however, to point out exactly what Machine Learning is and what it isn’t. Machine Learning isn’t Artificial Intelligence (AI). Intelligence infers learning and applying that to new and completely different and creative situations, Machine Learning only has limited capacity in different situations.

I have two young children and when they were very young (<2 years old) I found it fascinating how they would learn by pointing at objects and people, as parents we would respond “tree”, “dog”, “bird”. After time they would repeat back verbalising as they pointed at objects, slowly refining what they pointed at and making less mistakes. This is Machine Learning. In time, humans progress past this type of learning and become more intelligent.

Feeding examples of images, words or data and labelling those pieces of data into a neural network is what Machine Learning is today. If you give the model enough data and accurate labelling, it can provide predictions based on the patterns that the neural network detects with new data, even if it has never seen that exact data before.

Machine Learning has produced many breakthroughs that we take for granted today, for example Speech Recognition, Translation, Text Recognition (OCR) and more recently object detection and soon fully Autonomous cars.

The biggest downside of Machine Learning is the huge samples of data that are needed and the human input required to label that data accurately. As mentioned before, humans are the greatest Machine Learners on the planet, so who better to teach the machines.

This can take a lot of time and can be expensive. But innovative ways have been created to make labelling less laborious. The best systems for data labelling are the ones that get humans to label text or images without ever knowing it. Remember the reCaptcha (the squiggly letters)? The problem that was being solved was to stop automated bots from accessing websites, but the scaled solution was teaching a machine to learn letters that had previously been missed by its OCR models. Humans were inadvertently teaching a machine! And in the process digitising books and archives.

early machine learning input

More recent examples are spotting images of road signs, stop lights and pedestrian crossings. You are actually teaching self-driving cars objects that are found along the pathways of an Autonomous Car.

Raw computing power and the ever-increasing storage of data is the primary reason Machine Learning is starting to become widespread within many technology solutions today. Whereas in the past you would have to write a program to recognise many different nuances, scenarios and data relationships, not to mention the foresight to recognise those patterns in advance. Today you can simply feed in well labelled data and ‘teach’ your machine to do a task.

Amazing future solutions are imagined to be, spotting cancers in scans, better weather forecasting, more efficient case law and detecting fraud in banking.

At drivible our intentions are a little less ambitious, but just as revolutionary for car dealers. We have begun to use Machine Learning within our program to more efficiently process test drives and soon will use our expertise to organise dealer’s data and sales processes.

The future is very exciting.    

It appears that no matter the challenges involved in getting a car to completely drive itself, they are nothing compared to the ingenuity and self-belief of our tech savours (Silicon Valley engineers). While the progress of humankind is likely to solve the many challenges of level-5 autonomy (driving with no human input), it is likely that many of the unforeseen challenges and consequences will take longer to solve than the challenges solved thus far. Here are some of the challenge we don’t think are going to be fixed anytime soon…

Congestion

There is a lot of speculation that congestion in our cities will reduce because people will go from owning a car to using a car sharing service, this will free up parking garages as people won’t need their car during the day or when they are at work. While this may reduce the overall number of cars, it is unlikely to reduce congestion and will probably make it worse.

There is a very real possibility that congestion may increase as cars that would normally be left dormant will now be on the road looking for a rider. Or for privately owned autonomous cars, they will be travelling home empty to wait at home until the afternoon pickup. 

Charging

It is speculated that most autonomous and electric cars will be more suited and more economic for congested cities rather than rural areas or urban sprawl and this makes sense. However, cities are the most difficult areas to keep electric cars charged and ready as many people don’t have off-street parking, many people don’t own their roof, making solar generation and battery storage unlikely for most. Requiring large parking and charging garages will likely be necessary, contradicting predictions that parking garages will become obsolete.   

Systemic failure

The hardest consequence to predict is a possible systemic failure. While human drivers are fallible, we are all bad drivers in our own individual way meaning that we are unlikely to all make the exact same mistake on the road. Individualistic and chaotic systems have served humans and nature well, they create paradoxically stable systems overall. However, when you remove millions of individual decisions being made on the road and replace it with one decision logarithm the consequences could be diabolical.     

A nightmare scenario could be thousands of cars plunging off a ravine like lemmings because of some bug or miscalculation that only occurs because of the autonomous cars interaction with other autonomous cars. i.e. an individual autonomous car wouldn’t make the same error in isolation, but it’s interaction with others causes some unforeseen feedback loop.  

This is a science fiction nightmare and it’s scary to think about.

Blackouts & Disasters

I was fortunate to be driving home from work in what was considered to be an apocalyptic blackout in Adelaide on the 28th of September 2016. It was caused by a once in a 50 years storm and a main transmission connector line being flattened by high winds that left the majority of the state (about 1.7 million residents) without power. There are many reasons for the cause of the debacle, which I don’t want to go into here.

The weather that day was horrible, with rain pelting the windscreen, wind blowing debris onto the road and all major traffic lights on my commute totally blank on my 30-minute commute. Because of the sheer number traffic lights down, the police had no way of policing and regulating traffic through the major intersections that I needed to pass through that night.

So, what happened? Chaos? Crashes? Violence?

Not at all, in fact I don’t even recall hearing a beeping horn. Drivers just nudged their way through the intersections giving way where they seemed appropriate, taking turns to pass through the intersections that normally handle up to 50,000 vehicle per day. While we always assume that technology can do a better job than a human, faced with a complicated and nuanced scenario, humans generally do a pretty good job.

I can’t help but wonder, what will happen in the future with autonomous cars trying to negotiate an intersection that has no rules? Sure, some clever spark will write a logarithm to deal with downed lights, but no one disaster scenario is likely to be the same. Or, what would happen if all the communications went down for a period? I assume the autonomous cars would carefully stop in traffic while the riders just wait for communications to be restored. I can’t think of anything more frustrating!

Even more worrying, what happens when people need to evacuate a hurricane or bushfire? I can’t help but think of the confusion and helplessness that people might experience when a fleet of robo-taxi’s are tasked with evacuating large populations all at the same time and all in the same direction.

Freedom

The transition to autonomous cars has been compared to the disruption that automobiles had on the horse and cart, but I believe this will be a much bigger impact to society. When we ditched the horse and embraced the steering wheel, at least we gained more control over where we wanted to go, when and how, and as a bonus, we reduced the likelihood of being kicked in the face by a horse. A computer may be better at driving than the average human, but people will be receding control and it always feels more comfortable to be driving verses travelling in the passenger seat.  

In addition to the perceived loss of control, there is a very real threat that humans may need to be segregated from robot cars just like factory workers are with their fellow robot workers. There are already predictions that human drivers may be banned from entering certain areas of future cities and special walkways for pedestrians will need to be erected as to make the life of robot cars easier and more predictable.

This isn’t the way technology is supposed to be. We shouldn’t have to yield our freedoms to make technology work better, it should adapt to us. 

With the rise of automation, many in the industry believe we are only years away from getting fleets of robo-taxis that will ferry us away to our meetings, parties, workplaces, schools and dinners. The reality is, the technology, infrastructure and regulation all need to change first and this is unlikely to be a mere “few years away”.

But when all the factors line up and cars do have the ability to drive themselves, will everyone just book rides through ride-share apps? There is no doubt that some people who live in inner-city areas will be perfectly happy with ride-share as their only transport method, but large segments of society will still wish to own (or subscribe) to a specific car.

For many people cars aren’t just for getting from A to B, cars are used for storing sunglasses, nappy bags, prams, used coke cans, footballs, shoes, phone charges and mints. They can feel like our second home. If you have kids, are you really going to re-install child seats every time you get into a ride-share?

So the methods by which people have access to cars is going to become more fragmented, some people will 100% ride-share but large numbers of society will need 100% access to a car, 100% of the time.

Perhaps some families will own or subscribe to one car and then use a ride-share service for any other trips, but the complete death of car ownership is greatly exaggerated.