Entertainment v culture: The implications of AI recommendation and generation

Imagine a time in the not-too-distant future when the majority of your media is both selected and generated for you personally by artificial intelligence (AI) systems.

Here’s one possible scenario:

After a long day at work, you arrive home and put your feet up. You ask your digital assistant to log you into your subscription streaming service. Rather than being offered just the usual carousel of movie and TV series recommendations, there’s a new option. It allows you to “design” your own film, which is then generated using AI technologies and streamed to you.

From pre-defined lists, you specify the genre (spy thriller, romantic comedy, musical, etc), director, duration, rating (parental lock is available if you’re under 18), some basic plot devices, the main characters (strong female lead, racially diverse group of superheroes with special powers, lonely male lead with vulnerable female co-star, etc), and who you would like to play them.

You can select from a range of major Hollywood stars – both dead and alive – who have exclusively licensed digital versions of themselves to the streaming company (this was one of the main reasons you signed up). Or if you prefer, a digital version of you can be the star, and your friends (who must also subscribe to this service and have agreed to have their digital likeness portrayed) can be actors in the film.

The merging of personalisation and generation represents the ultimate optimisation for media production. Everybody satisfied, all of the time.

Once you’ve specified these few high-level options, you click on “Create Movie”. The screen goes dark and a progress animation spins for a while. After about five minutes, a play icon appears, and you click to begin the stream.

The movie is good. Occasionally the actors speak some strange lines or do things that seem somewhat out of character, but overall it’s entertaining, and you think the subscription is worth the extra monthly cost. It’s fun to watch with friends around, where you’re all stars in the movie. Slasher-horror and sing-a-long musical genres are popular, as are romantic films with a first date where you star alongside your (hopefully) future partner.

Similarly, AI pornography streaming has become so popular that governments have tightly regulated both consumption time and moral rights about how your “digital other” – what used to be called a “deepfake” – can be used by these AI streaming providers.

Each time you generate an AI film, it’s different than all previous ones, even if you select the same options, but after a while you start to think that the plots often either seem formulaic or don’t really make complete sense – something that perhaps you overlooked at first due to the visual realism and the thrill of seeing yourself star in a high-paced action movie alongside your favourite actors. The AI is uncanny at being able to reproduce the look, mannerisms and voices of real humans to such an extent that it’s almost impossible to tell the difference between a real person and their digital other.

The more you use the system, the better it appears to get at generating content that you like. Some people have become quite addicted to seeing fantasy versions of themselves and people they know in these rich cinematic narratives, opting to live emotionally through their AI narrative rather than in real life. Others use them as outlets for revenge on former lovers, or to act out fantasies on others without their knowledge. Underground markets emerge that engage in illegal trade in digital others without the owner’s consent.

Maybe you begin to worry about how good this system is, and its impact on society. It always seems to give you what you want, yet you feel somehow unfulfilled. Even when you try setting scenarios that you think won’t work, the system somehow manages to give you something enjoyable. It’s almost like it knows you better than you know yourself.

You’d like to give up using it, but what are the alternatives, apart from being a total social luddite? The moral questions seem to have been side-stepped, it’s not real people in these films, so what does it matter how people use them? Somewhat ironically, these very issues came up in an AI movie you generated recently. It seems like AI-generated films are here to stay.

The technology exists – but what do we want from it?

While this is currently a future scenario, it’s technically possible today. AI systems have already written drama scripts, and deepfake methods can reproduce the appearances and voices of real people, given enough data. While several major technical challenges remain, it’s very likely that these technologies will improve over the coming years to the point where my scenario is indeed possible.

But a far greater challenge for us today is what we want from this technology, and how it will shape our future culture.

Mainstream creative practices are often referred to as “entertainment”, as this term sits comfortably within the neoliberal capitalist hierarchy. Entertainment is something you consume in between working to pay for it.

A broader consideration, however, sees a continuum of creative expression, from individuals to global media groups, experienced increasingly through digital media. As a whole, this continuum plays an important role in our collective culture. Here, I want to argue that we must be extremely cautious in how we adopt algorithmic decision-making and generative AI technologies into our culture.

Algorithmic decision-making already plays a major role in how we experience digital media. Music, books and cinema all get recommended to us by algorithms. Think of the Netflix “98% match” – meaning the system thinks there’s a 98% chance you’ll like this movie. In a self-fulfilling way, seeing this close match convinces you that you should watch.

But what’s wrong with that? After all, streaming services have tens, even hundreds of thousands of works – how could you possibly find what you’ll like without some kind of help? Why waste time looking at things you’re just not interested in?

In the pre-digital days, we relied on critics, friends, experts or advertisers to recommend what entertainment we should consume, and while that still happens, increasingly we defer to the instant gratification of being told by an algorithm what to read, watch and listen to.

In the pre-digital days, we relied on critics, friends, experts or advertisers to recommend what entertainment we should consume, and while that still happens, increasingly we defer to the instant gratification of being told by an algorithm what to read, watch and listen to.

A significant advancement will take place when the media you consume is generated by the technology that’s recommending and delivering it to you. Streaming services and media platforms exist to keep you watching, listening or reading.

Their optimisation is to maximise engagement by giving you what you want. The main limitation for this optimisation is the cost of creating the media content that you specifically want. Mass-appeal blockbusters have bigger budgets and more resources because of their mass appeal. But the more money you spend on blockbusters, the less you can spend on broadening your content – a classic problem in multi-objective optimisation.

There are two possibilities to further optimise – broaden the appeal, or broaden the content. Each requires significant resources, but both strategies are already used extensively. Indeed, the initial appeal of music streaming services was the number of artists and tracks instantly available over any individual physical record store.

You may have noticed how, in recent years, Netflix has switched its recommendations from maximising predicted matches to prioritising its own content. This again represents the optimisation cost of creating specific content (cinematic films or mini-series are very expensive and time-consuming to produce, and far more expensive to produce than to licence). But what if that content could be produced by AI, ultimately at far lower cost than any real production? And if the cost, in both money and time, was significantly lower, then you could personalise to the point of singularity – 7.6 billion individual blockbusters!

The merging of personalisation and generation represents the ultimate optimisation for media production. Everybody satisfied, all of the time.

But such a simplistic optimisation overlooks the broader implications – those that differentiate entertainment from culture. They inspire a cascade of questions, such as how is human culture changed if we lose the ability to decide what media we experience? Who ultimately determines the cultural messages manufactured by machines?

Seeking culture is a conscious act

Deferring to AI to give you what you like is a major step in further diminishing culture at a societal scale, and cultural difference globally. Like a wasted muscle that atrophies due to lack of use, seeking out and developing our culture is an intentional, conscious human act, not the passive acceptance of an algorithm’s generative capacity.

If you’re never exposed to things you don’t like, you lose the ability to actively discriminate and critique. You forgo the ability to seek out the new, the different, the unusual, the painful, the offensive, or the profane.

Even more insidious is the machines’ ability to recognise this, and to generate the content that diminishes as a side-effect of mass personalisation. So you would see the different, the offensive etc, but not because you sought it out. At that point, we’ve deferred human culture to machines, and lost a fundamental force in human development that we take for granted today.

This article was first published on Monash Lens. Read the original article