Deep Personalization for AI Systems, Pumps-Up AI Self-Driving Cars

Dr. Lance B. Eliot, AI Insider

Image for post
Image for post
Deep personalization gets AI to be all about “you”

Does your car know you? In some relatively simple ways, it might.

The other day I rented a car during a trip up to Palo Alto to speak at an industry conference and upon getting into the rental it took me about a solid five to ten minutes to adjust the car to my preferences. The driver’s seat was too close to the steering wheel and so I moved it back a few inches. Side mirrors of the car were at an angle and a position that made them unusable for me, so I re-angled them. It was going to be a somewhat lengthy drive in the rental car and so I opted to set the radio station settings for ease of choosing the ones that I like. For the in-car temperature control, I adjusted the settings to fit to my liking.

In some respects, I was able to “personalize” the car to my own preferences, such as the seat settings, the temperature settings, the GPS destinations, etc.

Another aspect that you typically need to get used to when using a rental car is the nature of the brakes, the accelerator, and the steering wheel. Since cars typically differ in terms of how sensitive the pedals and steering wheel are, whenever you get into a different car than your own, you often need to figure out what the sensitivity level is like in this “stranger” car that you are going to use.

What does this have to do with AI self-driving cars?

At the Cybernetic AI Self-Driving Car Institute, we are developing AI software for self-driving cars. One aspect for AI self-driving cars is the potential for them to provide deep personalization.

Personalization of Your AI Self-Driving Car to You

Personalization refers to how well your AI self-driving car knows you and your preferences, along with the AI trying to fulfill those preferences as best it can.

There is a range of personalization. In some cases, the personalization might be relatively shallow and not provide much of any personalizing to you. In other circumstances, the personalization can be “deep” in that the AI has been able to get to “know” you via in-depth pattern matching, and can potentially anticipate your preferences and abide by those preferences when feasible.

Many of the auto makers and tech firms that are developing AI self-driving cars are considering this notion of personalization to be an edge problem right now. An edge problem is a corner case or considered at the edge of the core aspects that you are trying to solve.

I’d like to first clarify and introduce the notion that there are varying levels of AI self-driving cars. The topmost level is considered Level 5. A Level 5 self-driving car is one that is being driven by the AI and there is no human driver involved.

For self-driving cars less than a Level 5, there must be a human driver present in the car. The human driver is currently considered the responsible party for the acts of the car. The AI and the human driver are co-sharing the driving task.

Another key aspect of AI self-driving cars is that they will be driving on our roadways in the midst of human driven cars too.

Let’s next consider the various levels of personalization.

Four Levels of Personalization

I categorize the potential levels of personalization of AI self-driving cars into these classes:

  • No personalization
  • Shallow personalization
  • Substantive personalization
  • Deep personalization

The lowest level, consisting of no personalization, gets assigned when there is no particular attempt at personalization and that by any reasonable judgment we would likely agree that there is nothing really built into the AI self-driving car system for personalization purposes.

The shallow form of personalization consists of attempts at personalization that seem rather token and flimsy.

Substantive personalization would be personalization that is genuine in nature and spirit, providing a somewhat convincing set of personalization aspects.

The topmost category is the deep personalization mode. For deep personalization, the AI needs to have gone all out to provide a wide and in-depth set of well-coordinated and aligned personalization capabilities

At this time, most of the auto makers and tech firms are by default aiming at no personalization (or, at best, shallow personalization), meaning that they are not devoting key resources or attention toward personalizing their AI self-driving car technology due to it being considered an edge problem. They have in mind to eventually get around to adding personalization, but it is not in the cards for years to come.

Anthropomorphizing Automated Systems Can Lead to False Understandings

There are some potential adverse consequences related to personalization.

One being that if the human occupant believes that the AI knows more than it does, the human might make undue assumptions about what the AI is capable of doing, and the human might get themselves and the AI self-driving car into some untoward spots. Anthropomorphizing automated systems can lead to false understandings by humans about what the technology can and cannot do.

This then starts us down the path of understanding true personalization in terms of having the AI be able to identify patterns of behavior that can then be “learned” about over time by the AI and then reflected in what the AI does related to the actions of the self-driving car.

One feature that some cars have today consists of allowing you to set the driver’s seat in terms of its position forward, its tilt, its heating or cooling pad capability, and so on. Once you’ve set things to how you like the seat, you can have the seat configuration “memorized” (actually, just stored in memory of the computer), and later on if someone messes with your seat settings you can have it return to your personalized settings.

The AI of the self-driving car can do that same kind of “memorization” and yet go even further by being able to consider the context of the circumstances. Context can be crucial to have personalization that makes sense versus personalization that is “dumb” and simply taking place on a rote basis.

Anonymize Data Into the Cloud for the Benefit of Fleet Learning

Presumably, personalized data could be used for aspects such as fine tuning the AI system based on the kinds of behavior that people are exhibiting inside the AI self-driving cars of that auto maker or tech firm. In essence, they could potentially anonymize the data and try to find patterns collectively across all those that are using their brand of AI self-driving cars. This is known as fleet learning.

On a more adverse viewpoint, an auto maker or tech firm could potentially mine the data and then opt to sell the data to third parties or use it to beam advertisements to her.

The deep personalization would not solely be focused on the “owner” of an AI self-driving car.

When you consider that AI self-driving cars will most likely be used as ridesharing vehicles, there is going to be ample opportunity to track the behavior of humans in terms of their use of AI self-driving cars.

Creepy Personalization in the Eye of the Beholder

When I’ve spoken about the use of deep personalization at AI self-driving car industry events, some inquisitive person invariably asks the rather pointed question of whether this might be creepy. Yes, I certainly agree there is a kind of “creepiness” factor to this. Though, you can say the same about any kind of personalization service (especially the “deeper” or more highly personalized it gets, rather than shallowly personalized versions which are easier to shrug off).

Creepiness can be in the eye of the beholder, which perhaps this next personal anecdote might so reveal.

Years ago, I was doing some consulting work for an east coast client and the client arranged for my hotel stay nearby their headquarters. I arrived at the hotel around dinner time and got out of a cab that had brought me from the airport to the curb directly in front of the hotel. The bellman grabbed my bags from the trunk of the cab and told me to proceed to the front desk, not needing to wait for him.

When I got up to the front desk, the check-in clerk greeted me by using my name. I was puzzled that the clerk would know my name, since I hadn’t yet stated what my name was. I asked how she knew my name. She explained that the bellman had looked at my luggage tags and then had radioed to her that I was coming up to the front desk to check-in.

Clever or creepy?

I then went to my room with my bags. I left them sitting in the room, still closed up, and was in a hurry to meet with the client for dinner. I figured that after dinner and when I got back to the hotel room, I’d then have time to open my bags and remove the items that I wanted to store in the room. Meanwhile, I had been reading a book and placed it onto the bed, face down, at the place where I was reading, so that I’d know what page I was on when I returned to the room.

After a very pleasant dinner with some fine wine, I managed to ultimately get back to my hotel room. When I came into the room, the lights were dimmed, and I could see that the bed had been turned down. I realized that the nighttime service that you see at some hotels had taken place. I’m not normally very keen on someone coming into my room, but anyway it is customary in many hotels and didn’t seem especially unusual.

I then noticed that the drawers in the room were slightly ajar and the clothes closet was slightly open. This got my curiosity going. When I opened one of the ajar drawers, I saw that my various underclothing and socks had been neatly placed into the hotel room drawers. When I looked in the closet, my shirts and coats had been neatly placed on hangers.

They had opened my bags and opted to put my belongings throughout the room for me.

As a topper, they had put some bedtime slippers and a robe on the bed, which I’ve seen before, but the kicker was that my book was sitting on the nightstand next to the bed. At first, I was a bit irked because I had purposely placed the book open faced on the bed so that I would not lose the spot in the book that I was last reading. When I went to get my book, I realized they had put a bookmark at that point in the book, and it was even one of those bookmarks that comes with its own little nighttime reading light.

Clever or creepy?

In my own case, I thought it was rather creepy. They had not asked me whether I wanted to have this done. Even if it was considered their standard operating procedure (SOP), it seemed to me that they should have first established that I wanted this done. They could then have placed something into my hotel records that indicated I either wanted this to be done or not, and on subsequent visits have used that “memorization” to guide their actions.

When I told others about what had happened, some of my friends thought it was a tremendous service and I was being thin skinned about it. Others agreed with me that it was rather extraordinary and actually quite odd. One even said to me that he would have started looking around to see if the room had any cameras or audio listening devices. He figured that if they went this far on my bags, who knows what other surprises they had in store. It was a good point and I admit that I was somewhat on edge the rest of the time there (you might find of interest that on later visits to this client, I opted to book at a different hotel, thus, their form of personalization had the opposite effect of what they presumably intended).

Conclusion

Deep personalization for AI self-driving cars carries with it the rather incredible possibility of leveraging the AI of the self-driving car to provide a fully personalized experience for the occupants. It is even quite conceivable that the brands of AI self-driving cars will perhaps be differentiated from each other by which ones do no personalization versus those that do, and ultimately a differentiation between those that do some amount of personalization versus those that do so deeply.

To achieve personalization, deep or otherwise, the AI has to have this capability explicitly built into it. There is no magic wand that somehow allows the AI to just miraculously do personalization.

For those auto makers and tech firms that aren’t yet focusing on the personalization capability, at least they should be providing a spot to plug-in such a component, being ready to add the feature when it is built and tested. This capability could then be loaded into the AI self-driving car via the OTA and thus not necessarily need to have the AI self-driving car go to a dealer or auto shop to gain the personalization capabilities once they are ready for adoption.

The thorny topic of how to best provide the deep personalization is something that will need to be gradually figured out, gauging the public’s reaction and also potentially whether any regulations arise around the facets of it. Let’s aim to have the deep personalization be something in hot demand and that will inspire people to accept and use AI self-driving cars. We don’t want to have creepiness that causes people to be worried about a Big Brother kind of AI watching their every move.

Deep personalization for the good of mankind needs to be the mantra of AI developers for self-driving cars.

For free podcast of this story, visit:

The podcasts are also available on Spotify, iTunes, iHeartRadio, etc.

More info about AI self-driving cars, see:

To follow Lance Eliot on Twitter: @LanceEliot

Copyright 2018 Dr. Lance Eliot

Written by

Dr. Lance B. Eliot is a renowned global expert on AI, Stanford Fellow at Stanford University, was a professor at USC, headed an AI Lab, top exec at a major VC.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store