The future only remains uncertain if you’re too specific in asking something of it.

So if it’s inarguable certitudes we’re after, better to keep our query as broad as possible – even asking an explicit question might come across too nosey to solicit any forthcoming answers from proud Father Time.

And what we want to know about in the broadest sense is: the products of AI and the possibility of their seeming expressive. What of that, oh Future?

Somehow, the prediction comes easy. Without further ado then, lured from the leaky spots in the membrane of tomorrow – and not to be too combative, but anyone who disagrees is just plain wrong here – the final, conclusive, categorically definitive word on the future of generative programs in the creative space:

Either the products of AI will one day seem expressive, or they won’t.

Hold your protests – this is a more useful proposition than it might appear. From a roughly infinite number of possible destinies, we’ve extrapolated the only two relevant endpoints for AI’s creative evolution. At any given time in the future it must be the case that AI programs either are, or are not, capable of generating products which seem expressive. Doesn’t matter the program, doesn’t matter the process; either the output is that good, or it isn’t.

This finally gives us the foundation we need to synthesize all the heady babble (apologies for my own contributions on that front…) and begin working toward some conclusions as to whether human content is soon to be doomed to eternity’s scrap bin.

We first need to know what the criteria would be for AI generations seeming expressive. And what exactly we’re looking for can now be helpfully specified. Earlier, we agreed that the category of ‘expressions’ can be divided into:

Personal reactions to our own inner experiences.

And

Communications intended to affect the inner experiences of others.

On the grounds of the Hard Problem, we can toss the first type – it encompasses the kinds of private responses to an inner awareness that aren’t contingent on the observation or interpretation of others (think screaming into a pillow, humming to yourself, doodling in the sand as the tide advances…). That means there’d be no practical way to establish any good “seeming” criteria for this type of machine expression – given that we can’t access any inner experience to which machines might be subject, we’d first have to agree, without any evidence, on the likely nature of that experience, and then also form a consensus on how it would manifest itself. I guess that’s not impossible. But with both required steps still engulfed in the annoying murkiness of the Hard Problem, and fertile for all kinds of other disagreements, any resulting criteria would inevitably be weak and clunky.

The second type of expression is more relevant to our interests/anxieties anyway. And promises a more, if still not entirely, concrete and reliable way to evaluate machine outputs. Searching for type 1 expressions, we’d have trouble even ascertaining the nature of qualifying products – Inexplicable errors? Strange patterns in outputs? Unexpected behaviors? Outright declarations – “What follows is a private expression of my feelings; refrain from reading, meatbags!”? – almost anything might seem like an indication of machine consciousness, given almost anything could be assumed in the first place about what machine consciousness would be like. With type 2 expressions – communicative ones – we can evaluate something much more specific: the practical, factual effect of AI products on our own inner experiences.

The simplest criteria for an AI product seeming-expressive would then be centered on our reactions to it, on whether the product itself appears, to us, to be communicating something. So –

AI content could be considered expressive if by consuming it, we have the impression that we’re interacting with another mind.

Admittedly, this assessment is still a pretty vague one. It doesn’t propose any specific test, any set of conditions under which the content in question needs to be examined, or any explicit description of what it is to “have the impression of interacting with another mind”, other than invoking that this is what happens whenever we communicate with others through a given medium – language, music, artwork, performance, and so on. But it’s at least a reasonably direct and objective evaluation of seeming-expressive, which shifts our attention from speculative and unknowable interpretations of consciousness to the more empirical question of whether we perceive AI generated content to be functionally equivalent to the human stuff.

Seemingness is, again, up to our collective intuitions, so there was never going to be any way to avoid some ambiguity in putting it to use as a metric. The tacit assumption has been that the magic of democracy will sort out the details of what qualifies. That’s good enough for me. Anyway, it’s not too hard to envision roughly what it would take for a cultural consensus to admit that AI – behold! – has done it, become what creators have feared – stirring bestsellers, thought-provoking artwork in museums, poems and music that make stonefaced critics bawl like banshees mincing onion in a sandstorm. And all this, crucially, with a public sense of empathy, connection, contact, with the creative force behind the work; the same instinct that convinces the well-adjusted that their friends aren’t zombies, that convinces the religious to credit the natural world to a divine creator.

To reiterate then: AI will get that good, or it won’t. Q.E.D. No if’s-and’s-but’s.