2021 predictions for AI in Creativity & Marketing — and what to do about them
This is a summary of the presentation I gave at the Christmas edition of I’ll Be Back South West, a monthly forum for people who work in the AI/creativity/advertising/marketing space.
Many thanks to Kerry Harrison and Richard Norton for inviting me to speak. A video of the talk is also available here.
TL;DR: Erosion of trust, an AI skills shortage, machine learning taking over VFX, investment in data infrastructure, and the increasing merging of the physical and virtual.
First off, I should come clean and admit that I think predictions are (mostly) bullshit. You only have to look back at previous predictions you can see how wrong they can be. For example, in 1964, the Rand Corporation gathered 82 experts to make predictions — predictions which included, robots as household servants by 1980, two-way communication with extraterrestrials by 2000 and the breeding of apes and other animals for menial work by 2020.
Unfortunately my monkey butler has yet to materialise.
Short-term predictions can just as easily flop. Last year, Goldman Sachs, The Economist and Merrill Lynch all predicted strong US jobs growth. Brent crude was predicted to reach $65 a barrel (it got to $40) and the US were forecast to top the medals table at the Japan 2020 Olympics.
All these predictions were thrown off due to the accidental creation of a virulent nano-particle in a city the other side of the world.
But despite those strong caveats, I have had a go at making some predictions.
First off, it’s worth saying that there are some things which look like strong trends but which I have ruled out of this piece for a variety of reasons:
Omnichannel 2.0
Emerging physical retail and ecommerce. Like a lot of areas that have remained resilient to change, retailers will no longer have an option not to bring together their brick-and-mortar businesses with their ecommerce businesses. On Shopify for example, retailers who offer local pick-up/delivery boost spending by 25%.
But this probably isn’t relevant to the world of AI/creativity.
Smaller but smarter agencies
Forrester have forecast that 23% of all agency tasks in the US will be automated by 2032. But I don’t think this will properly get going for a couple of years — so not a 2021 prediction.
Fashion x Gaming
Balenciaga are presenting their A/W 2021 collection in a video game. Marc Jacobs and Valentino created Animal Crossing looks back in May. But I think this is mostly driven by lack of ability to have shows IRL, so probably not something that will continue post Covid (new strains permitting).
Third-party cookie death
Google plans to kill third-party cookies in Chrome by 2022 and this will have an impact, mostly on programmatic marketing. On the basis that programmatic is a really dull topic — just a personal prejudice really — I have excluded it from my list.
Right, that out of the way, here are my five predictions for 2021 and their implications — what we need to do/think about, given these predictions.
The thing that will define next year, is this year.
Fingers crossed, the end of Covid some time around the spring/summer will mean things getting back to normal. We will be going out again, socialising, eating out etc. This is not the crisis of 2008 — things will bounce back quite quickly. There are even some indications of pent up demand for those lucky enough to have spare money.
The first prediction though is something that has been ongoing for a number of years…
1. Trust will continue to decrease
The Ipsos MORI Veracity Index asks members of the public which professions they trust to tell the truth.
Trust in ‘the ordinary man/woman’ in the street has gone down by 11 points in five years.
Trust in ‘ad execs’, whatever that means, has gone down by 24% just in the last year (and stands at only 10% for those with a degree-level education)
Delving even deeper, into the world of AI-powered tech, there is clearly a a dichotomy going on between people blaming the ills of the world on algorithms and the real benefits of using them.
The mutant algorithm at the heart of the A-level fiasco in England, versus the ‘saviour’ algorithm which has cracked the problem of protein folding.
Oli Feldwick from The&Partnership has written a great paper on the problem facing the marketing world when it comes to AI, which boils down to four potential problems:
- Unfair advantage: Appropriate use of AI in marketing skews the game in an unethical way
- Unconscious bias: that seemingly working systems are based on unethical data, bias or premises
- Unintended consequences: Well-intentioned systems lead to unintended and undesirable consequences
- Unethical usage: downright unethical or reckless use of AI by bad actors
Four solutions according to Feldwick:
1 — Keep humans in the loop. This should be an “oversight and augment” role. Humans should be acting as the super-manager of the system, not just handing over control to the machine
2 — Diversity of data, teams and thought. Don’t rely on single data points (eg ‘clickthroughs’), make sure you’re not creating a Silicon Valley monoculture, ensure real diversity of thought/opinions
3 — Safeguards: explainability and a ‘kill-switch’. Make sure you really understand what your algorithm is doing and what could go wrong. He suggests a “pre-mortem” exercise, where you imagine all the ways that things could go catastrophically wrong, building in a kill-switch that disables features if they are being used in problematic ways
4 — Design ethicists and algorithm design experts. Have the people who can look inside the black box, monitoring usage and understand what’s going on
Which leads us neatly to my next prediction…
2. Finding the right skills will continue to be a problem
There is already a shortage of data-scientists. According to data from Quant Hub, job postings for data scientists outstrip searches by 3 to 1. It’s the second hardest role to fill next to cybersecurity (although, now that the ballerinas are retraining as cyber experts, that might not be such a problem next year)
This shortage particularly affects the creative industries, which are not the natural home for data science.
As Mike Burgess from trend-consultants, WGSN puts it, “It is a challenge finding people who understand the [fashion] domain as well as data science… data scientists who know what Bayesian Inference is, and also know what puff sleeves are.”
The antidote to this problem is a three-pronged approach:
WE: need to continue to educate ourselves, brush up on our skills, stay up to date.
YOU: start schemes within our businesses to educate others — everything from lunch-and-learns through to training sessions or even full-on sabbaticals
THEY: lobby industry bodies to start training schemes and champion this area. Currently, none of ISBA, The Advertising Association, The IPA, WARC, iab or the ASA have training courses in this area — they should.
3. The AI VFX take-over will continue
Something that’s been rumbling for a while now, but I think 2021 will be a landmark year for it.
There are increasingly things that machine learning can do to manipulate images which are quicker and cheaper than using traditional techniques. And these can be applied to almost every aspect of VFX and film production;
- Face manipulation and re-enactment
- Green screen/rotoscoping
- Noise reduction
- Motion and volumetric capture
- Grading
- Lighting, shading & texture generation
According to Rob Bredow, from Industrial Light and Magic, the company behind the Star Wars special effects:
“it won’t be too long before we’ll be able to use something like that at [feature film] quality. I don’t think we’re 10 years away from that. We may only be one or two.”
Three interesting examples of this are:
Face manipulation
Using machine learning and a $500 computer, user ‘ Deepfakes’ was able to do what Warner Bros expensive CGI experts couldn’t — get rid of Henry Cavill’s moustache.
Ageing
For charity, Malaria No More, Digital Domain aged David Beckham by applying machine learning to clips of both Beckham and an older stand-in delivering a speech. There was no 3D model or scan of Beckham, no face dots to paint out in post production, no extra ‘witness cams’ or witness cam crews.
Lighting
As anyone who has shot exteriors will tell you, natural lighting is a nightmare to manage.
Luckily, research scientist Ming-Yu Liu has used unsupervised image-to-image translation to render a daytime scene into night. It’s a bit rough and ready, but it clearly shows an interesting future for this technology
My two take-aways from this prediction are:
- If you work with VFX, ask yourself, ‘could this be a job for ML? Is there an algorithm for this?’
- As consumers, this does potentially open another avenue manipulation or disinformation so it’s another thing to be aware of in the ongoing truth wars
4. Data infrastructure will become more important
From a creative point of view this may seem a little dull, but it’s extremely important and so worth including.
Today’s data workflow is a mess, data is everywhere, spreadsheets all over the place. No-one knows what’s accurate or up-to-date. I heard a recent story where critical data was recorded on a whiteboard before being transferred to excel and emailed to someone. This is no way to run a business.
But by moving to cloud data warehousing, that can all change.
This is going to be a big agenda point for 2021 because the move to cloud data is going to touch pretty much every company — and fewer than 25% even of large enterprises have made the move yet.
The fact that this is so critical has already been reflected in something big that happened in 2020: snowflake. Not the white Christmas variety or the internet insult, but the biggest the largest software IPO ever.
Snowflake went public in September. They were valued at $33bn when they IPOed, nearly 3x their last private value of $12.4bn.
As I write, their stock is up 30% since launch.
If this talk of numbers and infrastructure is putting you to sleep, that’s fine, but do yourself a favour and watch this 5 min video which is a great primer from Andreessen Horowitz
As boring as it might seem, your clients will need to know about this topic.
5. “Phirtual” (there, I said it)
The merging of virtual and physical is not new, but it is definitely is on the up, driven by things like the implementation of Lidar on iPhone and new volumetric capture techniques.
In the workout space, three projects have caught my eye
Nintendo’s Ring Fit Adventure
In which you use a Pilates ring and leg strap to control your avatar on screen. (I’m at level 67, thanks for asking).
Holofit’s Holodia VR
A fully immersive VR running/rowing/stepping environment.
BBH Singapore’s Running Stories
Which builds an audio narrative based on where you’re running.
And also, new ways are being found to allow the physical to affect the virtual, such as this way of using a real light to cast virtual shadows using a £9 IKEA lamp:
Clearly, during lockdown, there has been a massive move to live streaming events with physical and virtual emerging. Ultra VIP seeing events in person, but more and more livestream remotely.
This is the Dua Lipa Studio 2054 livestream. 300,000 people paid $15 to watch it ($4.5m — kerching!)
But perhaps the most impressive physical/virtual mash-up I’ve seen to date is is the group stage show open for Worlds 2020, the League of Legends World Championship.
The implications for us of this prediction are twofold:
- There will be increased demand/investment in this area, so this is a good time to get on board
- The bar has been raised, the quality and ambition for this stuff is enormous — so, think big!
That’s it. Let’s come back next year and see how many of these have come true.
In the meantime, my plea to everyone who’s read this far is: let’s make 2021 the year of the monkey butler. We can do it!
Thank you and Happy Christmas
If this was interesting, please leave a comment or And please comment or share this article if you have enjoyed it as it helps other people find it. If you’d like to be on my (100% spam-free) mailing list to be updated when a new article is published, you can sign up here. get in touch
Originally published at https://www.linkedin.com.
