Can you sing an AI generated song in worship?

Yesterday, Tim Challies linked to an article titled ‘Worship AI – What an artificial intelligence lyric generator teaches us about image bearing’. You can read the article here. In his summary comments, Tim posed an interesting question:

While the technology is still a long way off, it seems to me we could see a day when AI created a good song and we’d have to decide whether or not to sing it. Would you want to sing a song that had no brain and no heart behind it?

I found that a fascinating question and thought I’d have a stab at an answer. So, here are four reasons why I probably would sing a good song that had been churned out by AI.

Truth remains true wherever it comes from

At heart, I would land hard on this point: truth remains true wherever it comes from. When Balaam prophesied against Balak despite being paid by the man to pronounce curses on Israel, there wasn’t much heart behind what he was doing. But what he said nonetheless remained true. When people tell Paul that there are others preaching Christ out of rivalry and selfish ambition, there doesn’t seem to be much godly motivation going on there. But Paul rejoices because Christ is being proclaimed. No doubt you can think of other Biblical examples.

But the point here is a simple one. The lack of true worship and the potentially ungodly nature of the one speaking doesn’t stop the sentiment being true. If the words being stated are true, if they speak to the reality of who God is, what Christ has done and those truths work for the upbuilding of his people, it seems to me entirely legitimate to sing them.

We already sing words from questionable sources

There are churches who sing songs from people who belong to errant traditions and by those who believe what is manifestly untrue. In some cases, they may not have been believers at all. Whether it is the traditional church happily singing hymns by John Henry Newman and Frederick Faber or the more modern church continuing to sing songs by Vicky Beeching, it is not unheard of for us to sing songs from questionable sources. Even reformed folks have no problem belting out Charles Wesley hymns despite their being full of Arminian and Weslyan holiness theology with its shades of sinless perfectionism. These things strike me as more problematic in that what they actually say and in what the writers certainly meant. But it barely raises a whisper for most evangelicals. Most default back to the previous point. If what was written is true and is capable of building up the body, then it is good for singing.

This is not so much an argument in favour of singing such things. It is more an argument for consistency. If we are OK singing songs by Anglo-Catholics who deny the gospel as we understand it, it is hard for us to be credibly upset at the words produced by an algorithm that means nothing by those words at all.

Words capable of meaning what we mean

We might well insist that authorial intent matters. Despite the standard argument against problematic Weslyan hymns being ‘oh, but that’s not what I mean by that’, we can’t deny that what Wesley meant when he wrote the hymn was something altogether different and used words that do convey his intended meaning. Can we realistically sing ‘changed from glory into glory til in Heaven we take our place’ knowing – according to his theology – our glorified state (as those words fairly clearly say) occurs before we actually get to glory! Simply asserting, ‘oh, that’s not what I mean when I sing it’ doesn’t really help because Wesley stated that he would spread his theology through his hymnody, which is precisely what he did.

Authorial intent changes what words are capable of meaning – they mean, at least to some degree, what the author intended them to mean. Unless they are Humpty-Dumpty, using words that don’t actually convey their meaning, we can’t do the same in reverse and simply impose our own meaning on words that clearly say something else altogether.

But words written by an inanimate computer seems to me a different matter. We can’t read too much into authorial intent because there is no intent. Maybe, should we be inclined, we could argue that the intent of the person who created the algorithm may be the issue at stake. Were they a believer or not and were they intended to generate worship songs or not? But, personally, I think that too far removed. Instead, I would take the non-existent sentient intent to mean that there is no malevolent intent. We can, therefore, simply take the words for what they are. Are they capable of meaning, credibly, what we want them to mean in the form they are presented to us? If they can be, then I think we are at liberty to use them. For the reasons I gave in the first point and the one that follows.

Worship that can be redeemed

There is a case to be made for taking words written in a vacuum and redeeming them. As the previous point, I would have my concerns with singing songs written by people who actively meant something very different to what I would want those words to mean. I struggle with Lead Kindly Light because no matter how much I tell myself that I am singing about Christ, I know full well that Newman was writing about the Catholic Church. I don’t think we can completely invert authorial intent.

However, I think we are in a different situation when it is words written in a vacuum. As above, I think the form of words we are presented with can be taken for what they are. They either state truth or they do not. They are capable of being understood in a godly way or they aren’t. They can be assessed on their content value alone.

However, one might pushback that such song writing is an act of worship. An algorithm is not worshipping God in producing these sorts of songs. Do we really want to be singing what isn’t produced for the glory of God?

A few things bear saying. First, again, I would point back to my argument from consistency. If we are happy to sing songs from those who have since shown themselves to be unbelievers, and claim to have redeemed them, we have no real ground not to sing those produced by an algorithm if we are being consistent.

But, let’s say you are consistent on that point and don’t sing such songs by unbelievers. What are we to say then? I would say, secondly, that the argument then comes back to the question of authorial intent. The algorithm has no intent when it writes the song. It is artificially producing words. The question of it, therefore, meaning something malign is moot.

This brings me to the third thing. Whilst the algorithm producing those words may not have been engaged in an act of worship in churning them out, that is not to say they cannot be redeemed as such by God’s people as they do engage in worship. Just as lots of things do not have truly Christian roots, but are nonetheless redeemed by the church, so words that credibly mean what they say, written by a computer, may not be an act of worship in their production may well become an act of worship in their usage.

Consider, for example, our church buildings or any of the furniture therein. Most people have no built their buildings or furnished them inside exclusively with stuff produced by believers. The act of building the building, or making the furniture, was not an act of worship by those making them. But the act of God’s people meeting together and using those things for that purpose makes those things part of our worship. They weren’t necessarily created for that purpose, but they are used by God’s people such that they do nonetheless serve that purpose.

It strikes me randomly generated words from an algorithm would fit that same principle. The words may not have been intended as an act of worship. There may have been no specific desire (because computers have no desires) to worship God in their creation. But the church may nonetheless take those words and, if true in the form they are presented and capable of building up the church, turn them into an act of worship in their usage.

So, there you are. Four reasons why I probably would sing an AI generated song.