Issue link: https://digital.copcomm.com/i/1513434
22 cgw o c t o b e r • n o v e m b e r • d e c e m b e r 2 0 2 3 OUTLOOK AI Hype vs. Reality BY JIM TIERNEY DIGITAL ANARCHY: PRESIDENT T here are, broadly speaking, two types of AI: older school, neu- ral network AI, and the newer large language model (LLM) AI that things like ChatGPT use. Older AI is the underpinnings of things like face detection or speech-to-text, while LLM is be- hind most 'generative' AI. One of the very common topics at NAB NY this year was, "How long before AI takes my job?" Like, damn near half the conversations I had were some variation of that. The answer is: a really long time—if ever. It's rather unlikely we'll ever have the "Create Award Winning Doc" button. The caveat is that it's going to 'change' your job soon. In small ways at rst, and bigger ways later. So just as when video editing moved from oine edit bays to NLEs, there are going to be new skills to learn. But for those of us that work with AI and develop those tools, we understand very clearly how limited it is, and how prone it is to go o the rails. Take face detection. It's a use for AI that has been around a (relatively) long time. It's a mature technology and it works well if you're facing the camera. Turn your head, and it fails oen. We use it in our retouching plug-in Beauty Box, and a large part of building that into the product was building tools around it to deal with how oen it fails. Like a lot of AI algorithms, ultimately it's a net benet to use it, but a lot of work goes into providing tools in the product to let the user x issues when it fails. Looking at generative AI: It's very hard to keep it on point, and for video, it's hard to see it advancing far enough any time soon to gen- erate good imagery for professional use (for TikTok, well...). That said, will it help you storyboard or think through potential rough cuts? Most likely. It's a tool. Potentially, a very good, time-saving tool. But a tool nonetheless. And to use that tool, you'll need to become familiar with models, reners, control nets, prompt engineering, etc., etc. There's some learning that has to happen. But it's not going to re- place you. I look at these mature AI technologies and how inconsis- tent they can be, and I just don't see AI replacing video editors or cinematographers or whatever. The hype around AI has vastly outrun the reality. I think folks are seeing a few examples where the tech worked and extrapolating it to the moon. What they don't show you is how oen it fails. So take a deep breath, budget in some time for learning/experimenting, and keep on keeping on.