Something for the weekend, sir?

Newsletter cribnotes

Robot Nirvana

Why the emergence of artificial intelligence that can write its own Nirvana tunes tells us quite the opposite of what some people think it does.

Over The Bridge, a mental health awareness organisation, has created The Lost Tapes Of The 27 Club, a compilation of songs generated by AI. The four songs mimic the tonal qualities of Amy Winehouse, Jimi Hendrix, Kurt Cobain and Jim Morrison, imagining what music by the artists were like had they not passed away.

Emphasis added. No, creatives, you are not out of a job.

In 1993 there had not been a Pink Floyd album of any kind for six years, and there hadn’t been a good one for fourteen. The Machine, judging the mood of the record-buying public, which tends to buy Pink Floyd records regardless of what they are like, considered it was time for a new one. But key band members were not talking to each other — other than through the medium of commercial litigation — and others were drifting away.

Now a new Pink Floyd album in 1994 could have taken one of the following forms:

• An actual, new Pink Floyd Album, where Roger Waters returned to join the, wrote most of the material and bossed, grumped and control-freaked the remaining members of the band as, allegedly, he did in the 1970s;
• The Pink Floyd album that was eventually released, The Division Bell, with Waters replaced by a session musician, and David Gilmour’s partner writing all the lyrics; or
• An album created by musically gifted Pink Floyd fans who are passionate about the band, deeply understand it, and have views on its thematic and musical direction, and who can create what they would expect a Pink Floyd Album to sound like.

In 1994 we were still deep in A.I. winter, so a Pink Floyd Album made by A.I., deploying machine learning, neural networks and natural language processing to emulate the band’s recorded output wasn’t really a viable output. But realistically, after the novelty had worn off, you would expect to to be somewhere low in that pecking order of options.

As it turned out the remaining members, minus curmudgeonly creative force Roger Waters, pulled off a fair approximation of what machine learning can do now. Beautifully recorded and redolent of the band’s signature crystalline guitar solos, swampy organs and moody synth pads, The Division Bell went straight to number 1 in the UK and the US, eventually selling something like 10,000,000 worldwide. But the critical reaction was “mixed”; one wittily dubbing it “Wish You Were An Animal On The Dark Side Of The Wall”. Despite phenomenal sales, posterity has favoured Roger Waters’s judgment that The Division Bell was “just rubbish ... nonsense from beginning to end.”

Roger Waters still tours The Wall. No-one tours The Division Bell. In a way it is the direct equivalent of Nirvana by AI: a luscious, flawlessly executed recombination of a bunch of old Pink Floyd records, put together without the “watchmaker” of Pink Floyd himself.

The importance of authenticity. Why is it not the same when it isn't David gilmour playing that guitar solo?

The importance of effort. We should not underestimate how we we value the effort required to produce intellectual property. Many years ago go robotics engineers designed a contraction that could play the flight of the bumblebee on classical guitar. Undoubtedly the machine was extremely complex, the programming highly ingenious and it executed the police flawlessly at tempo, undoubtedly more perfectly then the finest classical guitarist could. But would you pay money to sit in a concert hall and watch a robot playing classical guitar? Once the technical problem has been solved and can be inexpensively replicated the value of the performance tends to 0. Even though we can can program robots to flawlessly play, at no cost, we will still pay good money to watch a human virtuoso doing the same thing less well than the machine.

The segues into a conversation about the meaning of value. The same way that meaning does not exist in the words on a page, value does not exist in in the technical performance of a skill, but lies somewhere between the performer and and it's audience. Similarly, science is not simply demarcation of the correct answers to questions, but is demarcation of the correct questions requiring answers. This is a dynamic. It is complex in the technical sense, the ground rules are approximate and shift without warning based on the attitudes of the conversants.

Leaving aside all the overpowering psychological reasons not to value an AI version of Pink Floyd, there is the bluntly practical one. They can only ever be a flawless moment it can recombine existing elements into a new you form. But it cannot create genuinely new you output because it is not the artist. Whatever the machine comes up with it will not be what nirvana's next album was going to be. Of course, we cannot know that, but consider an AI algorithm directed at The Beatles first four albums. Is there any chance it could have devised music resembling that on revolver or rubber soul let alone tough White album or sergeant pepper's? An AI analysing Pablo honey and The bends will not produce amnesiac or kid A.

Allegory, fairy stories and the hubris of taking things literally

We have been been warning ourselves since the dawn of civilization about the folly of using magic to take shortcuts. If we take Arthur C. Clarke at his word that any sufficiently advanced technology is indistinguishable from magic then are we forgetting our oldest lessons?

Critical theory, post-modernism, and the death of objective truth

They say every successful conspiracy theory contains a grain of truth. They have to be something for even credulous people to glom onto.

Critical theory’s grain of truth, ironically, is that there is no truth. This is its debt to postmodernism, and it is a proposition that contemporary rationalists find hard to accept.

The irony deepens, for defenders of the enlightenment bring critical Theory to book for its ignorance of obvious truths, while critical theory itself has bootstrapped itself into assembling a new set of of objective truths, which happened to be different to the conventional enlightenment ones.

The deep problem that critical theory has, all agree (from Christopher Hitchens, Richard Dawkins, Helen Pluckrose, Douglas Murray and recently Matthew Syed) is that something things — physical sciences are a favourite example — just are true. No amount of identifying with an alternative theory of gravity will stop you from hitting the ground if you throw yourself out of a window.

On the other hand Jacob Howland made the interesting assertion recently that so completely has critical theory escape its postmodern origins, that it has become captured by, of all people the high modernists who inhabit an intellectual world that seeks to solve all problems by top-down taxonomies and computation.

An illiberal alliance of technological corporatism and progressivism is rapidly turning universities into a “talent pipeline” for the digital age. When fully constructed, this pipeline will deliver a large and steady flow of human capital, packaged in certifiable skill sets and monetized in social-impact or “pay-for-success” bonds. But the strongly particular or eccentric shapes of mind, character, and taste that make human beings, as John Stuart Mill says, “a noble and beautiful object of contemplation” would clog the talent pipeline.

Critical theory has escaped its usual confines in the liberal arts faculties of universities and is now inhabiting the management and human resource departments of corporations, and who are using their rationalist framework to advance what is a fairly radical political agenda. Critical theory is not an alternative narrative by which we can puncture the arrogant assumptions of the capitalist class: it has displaced them altogether and is making its own arrogant assumptions in their place.

That's not altogether a bad thing — although the practical effects of the updated dogma seem more pronounced the further from the executive suite you go — but it seems to me to substitute one set of bad ideas with another.

The idea of transcendent truth — a truth that holds regardless of language, culture or power structure in which it is articulated — is not false (that would be a paradox right?) So much as incoherent. It is incoherent because, as Richard Rorty pointed out, truth is a property of a sentence about the world, not the world itself. Truth depends on language.

And languages are intrinsically ambiguous. This is the tragedy and the triumph of the human condition.

The statement there is no truth is not an article of postmodern faith, by the way: you can trace it back as far as David Hume, Adam Smith, Charles Darwin Friedrich Nietzsche, Karl Popper, Thomas Kuhn and Richard Rorty. I know, I know: all old, dead, white, men. And Nancy Cartwright.

If you accept the proposition that truth is a function of a sentence and therefore the language of that sentence comma for there to be a transcendent truth the language in which it was uttered would necessarily need to be complete, comprehensive, and itself true. The nearest linguistic structures that we have to to complete languages are those of mathematics and perhaps science. Yet we know that mathematics is a necessarily incomplete language something Colin from that we know that any natural language is necessarily incomplete semicolon and in the case of science we know with certainty that science is not what a complete and comprehensive statement of the laws of the physical universe. We haven't solved the universe yet. There are large fundamental unknowns; dark matter; dark energy; the incommensurability of quantum mechanics and and special relativity. Even if the concept of transcendent truth were coherence we have nothing like enough information to access it. In the same way that the fielder does not have enough physical information to calculate the trajectory of a cricket ball, and therefore pragmatically approximates it, so we do not have anything like enough information to confidently predict the scientific performance of the universe and therefore we pragmatically approximate it.

Pragmatic approximations comma being provisional, contingent, and subject to revision at any time I’m are are more tolerant, plural and liberal than concrete scientific calculations.

The lack of a a coherent concept of transcendent truth is a a roadmap to tolerance, pluralism, and liberalism. It obliges us to treat as contingent anything we know comma to expect things to change and to be prepared for new and more effective ways of looking at the world. All it requires is that we substitute a certainty about how we view the world and ash that we see it as true with a pragmatism about how we view the world, seeing it as effective.