Posted: May 28, 2023 | Categories: Artificial Intelligence
I originally started this blog back in 2018 but never finished it.
When I started this blog, I assumed voice interfaces were the key to everything I wanted to write about. It turns out that voice is key to the world without apps, but AI’s taking a bigger role in what happens when apps go away.
In 2018, I purchased a new Audi Q7. I splurged and purchased a model with the Vision Package which gives me a really cool 360 degree view of ground outside of the car when I’m backing up or parking in a parking spot. This feature isn’t AI in any way that I can tell, it’s just some fancy math to stitch together views from multiple cameras simultaneously.
Another Vision Package feature I love is the car’s ability to read speed limit signs. As I drive around, whenever the car ‘sees’ a speed limit sign, it reads it and displays the results on the dashboard. I can even set an alarm so it warns me by blinking the speed limit notification on the dash if I exceed the speed limit.
There’s a spot in my neighborhood where tree branches partially cover the speed limit sign and rather than guess about the speed limit, the car dutifully warns me, loudly and annoyingly, that it can’t read the sign.
The car isn’t smart, it doesn’t truly understand speed limit signs. It has a built-in Machine Learning (ML) model it uses to tell if it ‘sees’ a speed limit sign, and, when it sees one, it uses another ML model to transcribe the numbers on the sign into a number the car uses to craft the speed limit graphic on the dash determine if it needs to flash the speed limit exceeded warning there as well.
Flash forward 5 years and the car still gives me the exact same warning every single time I pass by that particular speed limit sign in my neighborhood. This happens, and I’m only guessing here, at least 100 times a year, probably more.
Now, I can probably solve this problem by driving over there one day with lopping shears in the back of my car and asking the homeowner if I can trim the branches, and perhaps I will someday. But fixing the visibility of the sign is not the issue here.
So we’re talking about Artificial Intelligence here, right? Where’s the intelligence?
The car has a GPS system as well, with an accurate set of maps and real time updates that show traffic levels on larger streets. Pretty cool stuff. With that in mind, I made the connection that since the car has GPS and the car knows it can’t read the street sign, it should be able to make the connection that every time it drives by the sign, it’s not going to be able to read the sign.
Granted, someone could swing by and trim the branches, so it’s not guaranteed that it can’t read the sign every time, but work with me here.
Don’t you think that over time, like after the 100th time the car warned me unnecessarily that it can’t read the sign, that the car would stop warning me? No, it doesn’t. It is a residential neighborhood, so the chances are that the speed limit will be 25 MPH no matter what anyway.
The car doesn’t make that connection because the two systems (GPS and Sign reader) aren’t interconnected. When the car sees it can’t read the street sign, it has no idea where it is at the time this happens. Well, that’s not true, the car knows where it is, but the speed limit sign reading model doesn’t. If these systems are so intelligent (the I in Artificial Intelligence) why didn’t Audi tie the two together so the system can learn about this and not give me the same lame warning for 5 years and more?
Now, you may be thinking that the car doesn’t know this because it doesn’t know it’s me driving so it wants to make sure that anyone driving the car at this particular place knows that the sign can’t be read. Audi connects the key to the user settings, so as soon as I get in the car with my key, the car adjusts the settings to match. So, the car does know it’s me. On top of that, the car pairs with my phone every time I’m in the car, so when it’s my key and my phone, it doubly knows its me driving.
Yes, that’s a faulty argument, but my point is that the car has a lot of really interesting technology in it that are all islands, disconnected from each other. The speed limit sign reading capability is amazing, but the car only uses it to display either a solid or blinking speed limit image on the dashboard.
I write this as the world is absolutely freaking out about Artificial Intelligence (AI) and the impact it has or will have on humanity. However, none of this is AI, it’s just a bunch of crafty ML that appears to be intelligent but isn’t. Until someone builds systems where hundreds, thousands or even millions of models work together to solve problems, we’re not even close to true intelligence.
I’ll write more about this soon; I have some thoughts about ChatGPT and the like that I want to share with you, hopefully in the next few days.