The Deep Fake Weekly Update
Do you remember any of these people?
Actually, you can’t. They are fake. Fake as in they are 100% AI generated images. Look at the detail:
The stubble, the blemishes, the lighting and shadows. Scarily realistic. This is not expert level photoshop. Developed by Tero Karras and his colleagues, these images were generated by a Generative Adversarial Network (GAN). This is a form of machine learning where two neural networks are pitted against each other to see which can most fool a third network who acts as the “judge”. (That is a dramatic oversimplification that most high school computer science students can probably explain with much more precision. Best to read this article.)
The value of this technology is that you can create something like a training video in many languages in a far more cost-effective manner. The humourous side is that you can make a TikTok video of fake Tom Cruise doing a magic trick. The dark side is that you can pretty much create a video of anyone in any situation with a level of detail that is virtually impossible to tell if it is real. Next generation fake news – as if we needed that.
Of course, the original deep fake was released in 1989, when Meg Ryan met Billy Crystal.
((N.B. be very careful about the words you use to search for this image on the Internet. As Scarlett Johansson says, “The internet is a vast wormhole of darkness that eats itself. There are far more disturbing things on the dark web than this, sadly.”))
Deepfake videos are trendy, but the power and accessibility of machine learning is not fully appreciated. For many years the domain of very large and sophisticated businesses, smaller organizations often find the most benefit of AI for their customers/members, specifically in the areas of customer service, customer lifetime value and customer targeting–critical marketing functions of any business.
UK-based ecommerce clothing retailer ASOS has released a new app where you can upload a photo of your favourite celebrity wearing something you like, and they will scan their huge database of clothes to find something similar but more affordable. ASOS’ app results in 48% more product reviews, makes visitors 75% more likely to make a return visit, and results in orders worth 9% more. Well considered, AI can help you be more effective, more efficient and more timely.
There is a caution. AIs can only be as good as their training material so they have a tendency to exacerbate and perpetuate existing inequalities. For example, an AI system for estimating the risk of criminals reoffending was twice as likely to incorrectly flag black defendants as high risk. Amazon had to scrap a hiring algorithm because the algorithm didn’t want to hire women.
Algorithms can also be used to deflect blame. Stanford University Hospital set up an algorithm determining who got the first COVID vaccines to prioritize seniority. When front line staff protested, they essentially said “oops, the algorithm did it” before they issued the standard variety twitter apology.
One of the uses of AI you right be personally familiar with is online dating. This was invented by Joan Ball in the 1960’s and used punch cards to screen out malapropos* potential suitors. While it would take a few years for AI to take over the analysis, the principles were the same.
In a world where small businesses have closed the competitive gap with large businesses in many ways, AI will be a crippling disadvantage if we don’t make an investment in practical experience today. With technology that advances at the pace of Moore’s constant, many of the early-stage challenges will be resolved and we could be left behind before we know it.
On an unrelated note, one of our team members who prefers to remain nameless (for obvious reasons) has suffered a work-related repetitive stress injury… resting his chin on his hands too much in Zoom meetings. Seriously. Although it’s hard to take this injury seriously at all. Probably a pretty weak deep fake meme.
Stay strong, stay safe and try not to stare into your screen like an Auguste Rodin sculpture named “Philip.”
* Malapropos is our WOWU (Word Of The Weekly Update). Basically, it’s a fancy word for icky. And if someone uses it with you, remember that Joan Ball has much higher standards. I’ll have what she’s having.