top of page

Is it possible for machines to ask great questions and to have sparks of inspiration?

I think sparks of inspiration come from questioning what others may have not thought to question, and then gathering into your mind all at once all the right puzzle pieces so that you can see the picture when they come together. That’s an epiphany.

Most of the people we call geniuses just have an ability to question things a bit differently, hold a little bit more in their minds all at once, and have a passion for learning, so that they have puzzle pieces when they need them. Now think of the mind of a machine. It can hold an almost limitless number of puzzle pieces in attention simultaneously, but can it ask a unique question? And can it see a solution when all the puzzle pieces come together? If we build machines that can do these things, what wonders lay in store for us?

So how do we come up with great questions? I think it tends to happen when you’re trying to explain how the world works, which is basically a repetitive process that involves asking what causes a phenomenon of interest, and what causes that, and what causes that, etc. Little kids do this annoyingly well. It’s the infamous never ending series of “why“?

How “great“ the question is depends on how much the phenomenon of interest impacts your life and the things you hold most dear. So, there needs to be an understanding of human values embedded in any machine that would seek to ask a great question.

Then, suppose you found some hole or inconsistency in the knowledge network, you recognized that the neighboring phenomena are very important to people, and you had a large collection of observations about the world, how would you go about fitting pieces into that hole, and how would you recognize when the missing piece found its place? I suppose this is may be the same question as, how would you recognize inconsistencies in the knowledge network?

The process of finding inconsistencies, or consistencies for that matter, comes down to applying some basic if – then heuristics or rules of logic. So a system that can execute rules of logic would be necessary to both ask questions and to answer them. But where do we get our rules of logic? And how do we know that they are any good?

Most of our rules of logic come from observing patterns in the world; relationships or associations between phenomena that seem to occur consistently across instances and sometimes even across types of instances. We call these patterns “truths.“ We can also get these truths from listening to the observations of others. Machines are excellent at systematically observing patterns and copying others’ observations into their own memory, so this part of the architecture might actually not be so hard after all.

But wait, it is not enough just to observe a pattern. Developing a rule of logic or a heuristic requires the ability to recognize when a pattern applies not only to a certain type of thing, but actually the larger class to watch that thing belongs. But as long as individual concepts and more abstract classes are both part of the growing knowledge network, then perhaps a single pattern finding process would still allow for more abstract observations and conclusions.

So the next question is, how do “things“ get identified as “things“ in the knowledge network? How do you go from recognizing a single instance to recognizing that a whole set of things can be classified by a more abstract concept or group? I suppose this would be a relatively simple process of grouping things together according to shared characteristics. It’s unsupervised learning.

Machines can do this well for any characteristics they can observe and record. Since machines can observe and record many more dimensions than humans can access with our five senses, they actually have a much greater capacity to recognize uniqueness and similarities in the world, to group things together, and therefore to reason abstractly.

I guess all of this leads me to conclude that it is possible for machines to ask great questions and to have sparks of inspiration. We just need to put all these pieces of the architecture together.

Recent Posts

See All

Today I heard a beautiful story told by Bre Gastaldi on the Once Upon a Gene podcast, and it reminded me of something wonderful that happened when I was in 5th grade. One of my neighbors had cerebral

bottom of page