Humans vs AI as a team building exercise

"We've been flung into the future” I tell the group of people who've gathered, “a future where AI is coming for us and it's a matter of life or death.” With this, I offer a choice; to fight for humanity, for freedom and for dreams or to join our AI overlords and fight alongside them against the last scourge of humanity. It’s a pretty extreme choice, but when we talk about AI it’s often in extremes. AI is either hailed as our saviour or the demise of civilisation, a beacon of productivity or the killer of industries, a game changer taking things to the next level or the death of originality and the peddler of blandness. In a world where developing AI strategies for organisations is increasingly crucial, the only clear conclusion is that we need to try and normalise AI; and what better way to do that than making a mockery of these extremes.

This is Humans vs AI; it’s part competition and part experiment that a few of us within Telstra's ARIA team have organised. With the goal to have fun testing both AI and human limitations and start to normalise the use of AI. Upon reading about Nick Cave’s violent reaction to song lyrics written as Nick by ChatGPT, I was inspired to run an event where humans competed against generative AI to write song lyrics. There have been a lot of opinions lately from professionals in a range of fields, predominantly creative, media and education, on AI’s place in their line of work (just scroll through Linkedin's content). While this sometimes adds to the conversation of extremes, it most commonly centres around the argument that AI on it’s own cannot do as good of a job as a human professional. Which is valid given AI has a range of limitations, failing to grasp context or apply a personal touch like professionals in those fields can; it also can’t write a Nick Cave song as good as Nick Cave, but you know what… neither can I. So when it comes to normalising AI, what can we learn if we take a bunch of people and put them against AI in a scenario where they're not professionals?

To form the brief for the lyric writing competition, we took some suggestions on distinctive artists and held a vote with our peers to determine the style ahead of the actual competition on 6 May. With Beastie Boys taking out the vote for the style of the song lyrics, we injected some spontaneity on the day to make it more of a challenge. Right before kick-off, we drew a bunch items out of a hat which determined what the song was about (self-determination and moments), words that the song needed to include (work, coffee, Cinderella, houseplant & USA) as well as some must haves (an R&B section, a reference to a political figure and a reality TV show title). Apart from meeting the brief, the main rule was that teams needed to stick to their weapon of choice. AI teams needed to provide outputs generated directly by AI verbatim and not revised or edited by humans, while humans could not use any predictive or word suggesting technology. After that, all that was left to do in the remaining 45 minutes or so was form teams on either side and get writing.

We ended up with two AI teams and one human team, I turned the finished products into actual tracks and we invited our peers to help judge them. The irony of using AI voices over the top of a human created beat is not only amusing to myself, but also helped to add to the fun with the voice clones of Adam Driver, Danny Devito and Lady Gaga being able to make an appearance. While it was the human written entry that won from the 40 respondents who voted, there was a lot to unpack. Listen to the finished products of all 3 entries at the bottom of the page and judge for yourself. The voting was very close between the top two voted entries, the human-written track (“Prompt Queen”) and one of the AI-written tracks (“The Name of the Game”), but the other AI-written track was clearly not as popular, indicating that things are not as clear cut as we expected when it comes to AI capability and the insights we gained:

Using AI is all about context

• If you need to deliver something quickly then AI’s use case is sound. The teams using AI had a finished product before the human team had even written their first full line of lyrics. Even if the initial output is far from perfect, the advantages of using a large language model as a guide can be an incredible asset to productivity.
• Related to this is overcoming lack of time as a research constraint. One of the AI teams who participated were not familiar with the Beastie Boys music, yet with the help of AI were able to produce some lyrics that roughly represented the group’s style within the tight time frame of the competition. I don’t know any other scenario where this is possible with the same level of result.
• However, if it’s collaboration and culture you want to nurture then AI is not the answer. It was really difficult for the AI teams to collaborate or workshop whereas the humans were a model of collaboration, idea generation and synthesis. The overwhelming feedback from the human team was that the whole exercise was “so much fun!”, not something actively shared by the AI teams.

Using AI can be art

• It’s a hotly debated topic, but one of the AI teams took a maestro approach which highlighted that use of generative AI can be a form of craft. The group that came up with “The Name of The Game” spent most of their time iterating on their initial output to optimise, craft and refine it like a sculptor toning the edges of their clay into a refined piece of art. While the approach is different, it’s not drastically different to the humans who iterated on the words, phrasing and arrangement of their lyrics to uplift the quality of their finished product. I don’t know if there was backlash against digital photography and the editing process that came with it, but AI is just a tool to enhance and assist creativity and using it in the right way can be art.

AI doesn’t necessarily compromise on emotion

• What even is music if it doesn’t make you feel something? I had initially expected that the main advantage that humans had was the heart, the emotion and the passion and that this would be evident in the lyrical output. However, the AI team that kept iterating their craft to produce ‘The Name of the Game’ was ranked highest of the three tracks in ability to conjure emotion in the listener. With the power words yield, it’s hardly surprising.

Articulating human “value” can be difficult

• The voting and ranking of tracks against criteria of cohesion, memorability and emotion left the humans’ “Prompt Queen” and “The Name of The Game” neck and neck. It was only in the popular vote at the end the “Prompt Queen” took home the trophy with a staggering 55% of the votes. When I asked why this was when “The Name of The Game” had held an equal lead when assessing specific attributes, the answer wasn’t entirely clear, for some it was “the hook” but for others it was difficult to articulate why “Prompt Queen” was preferred. This is an important consideration when assessing the value of humans against AI; it’s not always about a checklist or anything that can be defined on paper, maybe there is magic there that we cannot always explain with traditional KPIs.

It all comes down to leadership

• Whether or not it was working with humans or providing prompts to AI, the main thing that mattered was having a clear vision and being able to work towards it. Identifying success and driving people towards it directly affects an output in any scenario and this is no less true providing prompts to AI than it is for inspiring humans. Leadership is one skill that will likely be more difficult than most to replicate, so if you want to invest in resource that is your best bet.

While there are a number of learnings and insights into using generative AI, nothing from our little challenge can give any real indication of what this might mean for industry. If Sting’s recent comments come to fruition then potentially humans vs AI will become a more serious competition for survival. When you look a little closer at many of the arguments against generative AI, it’s not actually AI itself that’s perceived as a threat; it is the homogenisation of art and industry. The problem with that is homogenisation has probably been happening for a long time now. In music for example, social media and the economics of streaming have been driving songs to be increasingly shorter and a third of all 196 million tracks available on digital services were released in the last couple of years, for which tens of millions of those tracks rarely achieve even one listen. The growth of generative AI may see these things exacerbated but it can’t shoulder all of the blame. When it comes to everyday life and work however, the best way forward is to lean in and try to take advantage of the speed, quality and learning opportunities that generative AI presents and steer clear of it when collaboration and culture are needed. AI will only become more prominent in our lives in a boundless array of directions but until then, AI is probably not going to kill creativity, in fact; if used right, it may help incite it.?

Listen to Prompt Queen here

Listen to The Name of The Game here

Listen to Unstoppable here