A.I. that Thinks Creatively, Part 3

From Narrow To General AI
10 min readApr 5, 2020

Part 1, Part 2, Part 4, Part 5

What drives you? Where do your drives come from?

In the previous article you saw how human creativity is a result of your ability to recognize and solve problems. Yet no amount of knowledge can help you divine a solution if you don’t care about the problem to begin with. You have to perceive a situation as a ‘problem’ before you are willing to put effort into solving it.

For a person, or an A.I., to create works of art or solve difficult challenges, it must have the right motivations in place. To be artistic an A.I. must want to entertain, inspire, or enlighten others. A writer must want to communicate something important, and when she doesn’t feel she has, she must be driven to become a better writer. Indeed, a more diligent writer may be disturbed by even minor imperfections in his writing, and would seek the means to correct them.

Experiencing something as a problem motivates you to innovate, to think of new solutions where you had none before. The depth of your ingenuity is directly related to how many problems your idea or approach solves.

The right motives, combined with the right knowledge, are the necessary catalysts for creativity. This article and the next one you’ll learn how motivations arise in you, and how they can be replicated in A.I.

The Source of Motivations

Let’s start with the basics.

Some motivations, like pain, hunger, or exhaustion are built-in motivations. When you were born these were already in your body. You never had to learn them. You will feel them under specific conditions, whether you’d like to or not.

Each is a problem, and your goal is to find solutions. When you feel pain, one solution may be to move away from the cause of pain, take an aspirin, apply a band-aid, or keep your wounded hand still. Often your solutions are even more creative. Every one of your motivations is a problem to be solved, a stress to be relieved.

You also have inner drives like shame, trepidation, boredom, injustice, and loneliness. These motivate you to seek their opposites, namely confidence, security, adventure, equality, and social connection. These are only some of the hundreds, perhaps thousands of internal motivations you have.

Where do these come from? A newborn baby is not ashamed at being naked, as an adult might be. Nor is a baby worried about doing her taxes, or anxious about death. We acquire each of these anxieties and aversions throughout our lives. But why acquire them at all?

Public Speaking

Imagine that you’re a teenager who doesn’t like to be embarrassed in front of friends and strangers. Now imagine that every time you’ve given a speech in front of an audience, it’s ended in embarrassment. Gradually, or quickly, you will start to dislike public speaking itself. Whereas before you only disliked and avoided being embarrassed, now you also dislike and avoid public speaking.

Later on in life you land a job which, much to your dismay, forces you to speak in front of unsympathetic audiences. You now begin to dislike your job, and dread going to work every day.

One at a time, you transfer your aversions from the things you dislike to the things that put you in the situation where you have to experience them. Aversions are contagious. As a newborn baby, you dislike pain. One or two brushes with fire, and you dislike getting close to fire as well.

Here’s another typical chain of aversions. This one motivates you to try to understand how the world works. The feeling of pain causes you to dislike objects that cause pain. In order to avoid these, you try to impose your needs on the world, and get upset when you can’t control the situation. Since you can’t control what you don’t understand, the desire for control causes you to dislike unfamiliar things; which can only be resolved by familiarity, that is, understanding. Here, the chain of dislikes moves from pain, to lack of control, to lack of understanding.

Motives have two sides: aversions and goals. This article focuses on aversions, things you dislike and try to avoid. The next article will discuss goals, things you move towards.

Every chain of aversions leads back to the basic motives we are born with, like pain and hunger.

This transfer doesn’t happen all the time, though. For example, if I don’t like public speaking, and I’m asked to give a toast at a party, I can usually decline. My discomfort is avoidable if I take the right action. I don’t develop an aversion of going to parties, since I can deal with these problems as they arise. It seems there are very specific conditions in which you transfer an aversion from one situation to another. Let’s dive into what those specific conditions are.

The Mental Shift

In each of the above cases, you didn’t choose to dislike the new situation. Your mind creates new aversions without you even realizing it. Let’s look at an example which illustrates the automatic nature of this process.

If you drive a car, you may have found yourself in the embarrassing position of accidentally blocking an intersection, if you’ve crossed the intersection without waiting for the other side to be clear. Once the cross-traffic light turns green, you find yourself confronted with a wave of angry, honking drivers, without having anywhere to move. After one or two such experiences you may have developed an aversion to crossing an intersection without waiting for the other side to clear.

If you were developing a self-driving car, you’d probably like it to have a similar aversion. Let’s think about how it might have happened for you, then see how a car could develop the same aversion. I’ll describe three possible outcomes, and show how only the last one creates a new aversion. To best approximate a self-driving car, assume that you are a relatively naive driver, as yet unaware of social expectations, and about to learn them the hard way. You do, on the other hand, have an aversion to lots of cars honking angrily at you as well as being in an intersection when the light is red. It is assumed you acquired these elsewhere, through the same process outlined below.

So let’s start with scenario 1.

Scenario 1: Everything Goes Well

Imagine that you are late for an appointment, so you find yourself rushing, trying to avoid red lights, and impatient to get across town. At some point you find yourself in the middle of an intersection and realize the cars in front of you aren’t moving anywhere. As a naive driver, you only feel minor impatience at the slow state of the traffic. You have no idea yet that you’ve made a mistake.

If the cars do get moving again, then no harm, no foul. At this point your goal is to get to your destination as fast as possible. Only a highly conscientious driver would consider these events to be anything more than a nuisance. You might even pat yourself on the back for getting through the intersection by a thin margin. Since you never broke the law, nor did the cross traffic ever get angry with you, you didn’t develop a new aversion.

Scenario 2: You Solve the Problem

Again, you are late for a meeting, and find yourself stuck in the middle of an intersection. This time, the traffic light changes to yellow then red, and the traffic in your lane still hasn’t moved. The cross-traffic starts to honk at you. This causes you some stress. As both a human, and a self-driving car, you would already know that this situation is something to avoid, even if the cars were not honking. You are in an intersection when the light is red: that is already a cause for stress.

You try to solve this problem and remove the stress. You look around for an opening, an escape. Thankfully, the lane next to yours is free. You quickly duck into that lane, and clear the intersection. The stress is gone, the problem is solved.

Besides escaping this stressful situation, you’ve also learned a new trick. Next time you are stuck in an intersection with an opening in the next lane, you would repeat this solution. You learn how to respond based on what previously worked. If instead of switching lanes, you had reversed out of the intersection, then that would be the first thing you’d try next time. You can read about this learning process in detail here, here and here.

No new aversion is made. Why? Because you found a solution. And since you found a solution, there’s no reason to be scared (yet) of a repeat of that situation. Nor is it reasonable for a self-driving car to avoid all situations where it, at one time, experienced some stress, especially if it was, and is, able to relieve it.

Scenario 3: You Get Frustrated

You are rushing to an appointment, and find yourself stuck in the middle of an intersection. The traffic light goes red, and you can’t move. The cross traffic starts to honk at you, and you are in the middle of an intersection; these are all reasons to feel stress.

As you look around, all lanes are blocked. While the cars honk furiously at you, you try to shift your car to clear the road. You look at the cars in front of you, you try to maneuver between them. But there is no opening in the wall of traffic. The cars behind you block any attempt to reverse.

After a few seconds you find you’ve made no progress; you’re still facing the same wall of cars you were trying to get around before. You’re still in the cross-traffic’s way, so the stress is still there. You feel frustrated at your own failures, at your inability to move even one meter out of their way.

At this moment, and without you realizing it, something shifts in your mind. It happens under the hood and you are unaware of it until after the fact. You are now averse to being stuck in this intersection facing a wall of unmoving cars. You ‘hate’ it. The frustration you felt is a sign that this change has occurred. Unlike in scenario 2, where you learned to respond to the situation by going around the cars, here there is no solution. The situation itself is now something you want to avoid.

Look again at the key step that made the difference. The situation you tried to respond to was the sight of a wall of cars. When you found yourself looking at the same wall of cars once again, and the prior stress had yet to be relieved, you realized you were out of your depth.

Feeling frustrated is a clue that a new aversion was created. Other similar feelings that signal a sea-change are the so-called “big feelings”: embarrassment, anger, despair, regret, etc. Each indicates that you are now averse to something you weren’t before.

One last note on this topic: your personality is determined by your likes and dislikes. These are your values. Therefore this change also means that your personality has shifted, ever so slightly.

Creating Dislikes in A.I.

Why do this at all? What benefit is there, either to an A.I. or to a human being, in learning to avoid a hopeless or frustrating situation?

We all have limits. There are many problems that you and I may not be ready or equipped to face. Avoiding these shows discretion, caution, and wisdom. A baby (or a robot) that, despite repeatedly being burned, doesn’t learn to dislike and avoid fire, is not being brave. It is being foolish. Fortunately for you, your mind automatically learns to avoid situations which lead to insoluble problems, for example, putting your unprotected hand in a flame.

Their insolubility is what separates them from the regular, everyday problems. As in the case with the car above, as long as you have hope that the problem may be solved your persistence might pay off. But if it is clear you have no hope, that you will definitely be stuck in a stressful situation with no way out, or that no amount of willpower or cleverness can protect any human hand from fire, your brain would be wise to take it as a sign that this path leads only to failure, and that it should try to find another route to its goal.

If an A.I. is to be wise, it must learn to do the same.

Part 1: Imagination

Part 2: Solving Useful Problems

Part 3: Motivation

Part 4: Defining Skills

Part 5: Thoughts: Knowledge and Intents

Are you also interested in applying Artificial Intelligence to human creativity, human understanding, even human values? Do you feel that our current goals with A.I. are limited? I’m looking to connect with others who have a similarly ambitious vision of the future of A.I., whose goal is to tap the full creative potential of human intelligence through software.



From Narrow To General AI

The road from Narrow AI to AGI presents both technical and philosophical challenges. This blog explores novel approaches and addresses longstanding questions.