Can Computers Write Creative Programs That Solve Problems?


In the context that I think you mean, creativity refers to the application of one pattern of relations to a different circumstance thereby solving a previouslly unsolved problem, and a problem whose solution is not already present in the domain of solutions expressed by the program code.  This set of associations is what produces ‘aha’ moments in humans: when ‘clouds’ of related ideas are connected.

In this sense, I think the consensus is, that there is no reason we can’t build computers that can do this.  The problem is currently size, expense, and the structure of symbols’memories’ that we put into computers.  So, like interstellar flight, we are really just trying to find an affordable way to do it.  We CAN send someone to mars, or something to another star. It’s just absurdly expensive compared to what we THINK we can do with some innovation. So we’re waiting until its cheaper.

I don’t have a lot of time right now to be thorough and the information is available elsewhere.  But the simple version is, that if you have a biological organism and start with basic stimuli (a subset of light, sound, vibration, touch, time and memory) that the structure of the physical universe, evolution and experience form a fairly accurate but simplified set of categories in our memories and therefore minds.  We build a set of symbols (patterns) from seemingly disparate stimuli, the same way we ‘see movement’ in the static figures of a flip-book.

Computers we use today do not start with this atomic level of representation, they start with symbols. And we are just beginning to understand how to symbolically represent  physical reality in commensurable terms, and any computational system requires commensurable terms. Humans have senses,  instincts and preferences which largely form our commensurable terms; all language being an allegory to experience, and all systems of measurement producing allegories to experience.

For example, money renders all things commensurable by price.  But without prices and money you couldn’t form a division of labor – a market. and we’d still be hunter gatherers or small family farmers.   Likewise  we can’t quite yet design software that symbolically represents reality. (ALthough this project has been underway for more than a century in philosophy, its largely been fruitless.)

Since language (the written word) is an allegory to experience, that language should (in theory) represent symbols that are commensurable (subject to comparison and evaluation) even if only on ordinal (ordered), not cardinal (numbered) grounds – because humans operate ordinally not cardinally.

The closest we have to that body of symbolic information that is broad enough in scope to represent enough of the physical world that the errors produced by sybmolic assocation are Turing-testable, is the Google search index.  (And google is fully aware of that). 

But the computational power to use that data given that its index is not commensurable with other domains, (we think) is approximately equal to the total computing power present on the planet today. And even then, we suppose the mechanical process ‘thinking’ would be very slow.

(I worked with a group of very bright people on the possibility of raising venture money for solving this problem, given that we are pretty sure how to a) create the programming tool set, b) use existing hardware technology, and c) represent the data in sets of mathematical manifolds, but it is far too early and far too costly to produce this scope of work. And to the venture community it is indistinguishable from snake oil.  So I’m not unfamiliar with the problem set, or the possible technical solutions. And I was willing to put my own money in. So I”m pretty confident.)

Most solutions today are attempts to model the human brain with digital systems. The general idea is that it’s cheaper to do this with existing hardware than it is with to build dedicated hardware for the purpose.  And even with that technique, most recent estimates I’ve seen are in the billion dollar range.

But it’s not that it’s not possible for computers to be ‘creative’.  Its that the minimum threshold for ‘creative association’ is a higher than the intelligence of a domestic dog, and we are still programming in symbols, not patterns, because those symbols incorporate our existing knowledge. And we’re doing that, it looks like, because it’s metaphorically the equivalent of a trip to the stars, and no one is ready to pay for that yet. 

I think that in this short space, that’s the most accurate statement we can render.  It is a matter of money, not logical possibility either by the symbolic route, or the neural route, or the dedicated hardware neural route.


Leave a Reply