Solving Useful Problems

A.I. that Thinks Creatively, Part 2

From Narrow To General AI
8 min readApr 5, 2020

Part 1, Part 3, Part 4, Part 5

The roots of consciousness are mysterious. As far as I know, my conscious thoughts are my entire existence. My thoughts are what make me who I am. They make me happy or miserable, productive or inept. Yet I have very little insight or control into what they are, or how to change them.

“A thought occurs when it wants to, not when I want it to.” — Nietzsche

Some of them seem to come from nowhere. This makes explaining my thoughts and actions to others very difficult. It makes explaining them to myself difficult too.

I know that one thought can lead to another, and that something in the real world can trigger a related thought. So I can say that thoughts are often caused by triggers or cues. They are rarely random. If they were random, my thinking would be only chaos. It wouldn’t even be “thinking”.

In addition, the words I hear in my head are usually in English, and the scenes I see are loosely connected to my life’s experiences. This means part of the answer to this question must be that thoughts come from my experiences.

Given I know all this, why is it still so hard to figure out where exactly my thoughts come from?

To answer this question, let’s start by looking at how a person might learn something new.

Learning a Skill

With any skill you learn in life, there is some motivation that is driving you to learn it. You can only gain skill in drawing if your drawings are currently sub-standard and you want to make them more beautiful. You become a better writer because you can’t reach and impact your readers as well as you’d like to. To get better at something means you didn’t like how you were before, and that is why you learned the skill.

As a child, when you first learn to tie your shoelaces, you are gaining a new skill. There are many possible motives. Perhaps you’re anxious or feel shame at not being able to do something so basic. You may resent relying on others to help you. You may be frustrated at your repeated fumbling and failure. Or you may simply want to make your parents proud.

Before you’ve learned to tie your shoelaces, the sight of untied shoelaces causes you some stress. It indicates a problem you have yet to solve.

Thinking more positively, you have a goal in mind. You anticipate a sense of pride and independence you will feel once you no longer have to ask others to help you, or once you can show people how you did this by yourself. The actions that you end up learning, i.e. tying your shoelaces, is how you bridge the gap between your problem and your goal.

What a great moment it was, when you learned the correct sequence of actions to tie your shoelaces! You no longer felt the anxiety of impending failure or shame. You felt a rush of relief…

…and it soon became an automatic habit. Once you’ve overcome the original problem presented by untied shoelaces, you no longer feel much pride in the accomplishment. Your ‘high’ only lasts as long as there is a real challenge. But once you learn some reliable solution, your anxiety and joy both disappear together. You carry out the actions almost robotically.

We know this from experience. No single source of joy lasts forever. Even the greatest accomplishments, repeated enough times, become mundane. The feeling of joy only comes from surmounting difficult or treacherous situations.

Skills solve problems

Every day you face hundreds of problems, large and small, like dealing with a rude co-worker, cleaning a sink, or doing your income taxes. How many of yesterday’s problems and anxieties can you remember today? What about last October’s? Because you most often forget the problem once you have learned how to deal with it, it can be difficult to connect every skill with its instigating problem.

There was a struggle involved in even learning to speak English. Regardless of if you remember it or not, every ability you’ve ever learned comes from overcoming a challenge. Before you learn a skill, there is something you wish you could do but you can’t, like tying your shoelaces. At the time, then, you must have had a problem.

The purpose of learning a skill is to sidestep the problem altogether. Once you are reliably able to do this, the problem will, in general, no longer be present, and neither will the accompanying joy at solving it.

Fortunately, there is a long-term benefit that you gain from solving a problem. It’s a new learning, a new skill. A new learning can be a new set of actions, or as you’ll see in a moment, a new thought.

Thoughts Solve Problems Too

In another article I discussed how thoughts are self-generated experiences. This article on the other hand is about why you have certain thoughts at certain times, and how they can be useful to you.

Just like tying your shoelaces, learning a new thought solves a problem. Both actions and thoughts are responses you learn so you can deal with problems. They are both triggered by a cue. And they are both “useful” in that they reliably sidestep the original problem that caused you to learn them. But with thoughts, the problems you solve are not in the world outside, but rather in your imagination.

This can be as simple as learning someone’s name.

Imagine you’ve seen a coworker a few dozen times before you begin to interact with her. At one point you suddenly realize that you don’t know her name. You need to get her attention, but this causes you anxiety, because getting her name wrong would likely be embarrassing. You imagine yourself saying the wrong name, and you visualize the resulting fallout. You may have experienced such unpleasant consequences before, and those memories now flood your mind with anxiety.

By now you’ve established the cue: you need to call out her name. And you’ve established the problem: saying the wrong name and embarrassing yourself.

By luck, someone else calls out her name and she turns around. You breathe a sigh of relief. You now have a solution. You heard a word, (“Queenie”), and your target confirmed it was her name by responding. And so a link is made; but this time the link is not an action, but a memory.

Just as with actions, the sights and sounds you hear before the problem goes away are automatically linked to the cue, and get triggered the next time the cue occurs. Your brain automatically assumes that whatever it saw or heard the moment before the problem went away was what caused it to go away .When you remember it later, it is a thought.

Going forward, anytime you need to call her name, the memory of her name will immediately pop into your head as a thought. That is the new learning you created. Your mind goes immediately from the cue to the solution. You no longer imagine yourself being embarrassed by saying the wrong name, because avoiding that situation was the reason you learned her name in the first place. You have sidestepped the problem.

Here’s the kicker: as soon as thinking of her name becomes automatic, you may forget all about the incident when you couldn’t remember her name. You may forget that that was the reason you learned it in the first place.

Similarly, you may forget the stress of not being able to tie your shoelaces. You may have no recollection of where you were at the time, how you learned it, or how many times you failed. All you will remember is the solution you learned. By far the majority of your thoughts and actions keep no record of when or where you learned them.

You probably know over 5000 English words. Look at the words in this sentence. Do you remember how old you were when you learned each of them? Do you remember where you were, what the circumstances were, what book you were reading, what TV show you were watching, or who you were talking to? The problem-history of your thoughts fades very quickly. In retrospect, you only remember the solutions that you learned. No wonder people generally have a rose-tinted and nostalgic view of their past. They rarely remember the problems they experienced while learning.

This is why thoughts seem to come out of nowhere. The moment during which we learned them is usually long gone from our memories. They are now automatically triggered by their cues.

Though it seems difficult to grasp, each piece of knowledge, each thought, at one time had a specific moment of learning like the ones above. A specific problem presented itself, then a solution.

The problem need not be so drastic as being embarrassed. Any problem that you want to solve will do. Drawing a blank on the right word when you’re trying to express an idea is a problem. If, while struggling to find the word, it is presented to you, or jumps out at you from your jumble of thoughts, it gets linked to the idea, and from that point on automatically presents itself when you think of the idea.

Getting A.I. to Solve Problems Creatively

In the previous article in this series, I discussed how the imagination, and thoughts in general could be replicated in an A.I. But without some reasoning to guide your thoughts in a useful direction, they are not much more than garbled hallucinations. This article shows us how an A.I.’s thoughts can actually be useful.

With nothing else to go on, you can assume that anything that the A.I. does, or sees, or hears before a problem goes away in some way caused the problem to go away. This rule is certainly not perfect, but at the start it is a good first guess.

Part 1: Imagination

Part 2: Solving Useful Problems

Part 3: Motivation

Part 4: Defining Skills

Part 5: Thoughts: Knowledge and Intents

Are you also working on applying human creativity, human understanding, even human values to Artificial Intelligence? I’m looking to connect with others who have a similarly ambitious vision of the future of A.I., who want to tap the full creative potential of human intelligence, in 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.