Side Of AI 2023; Awesome or Armageddon
Well, if you have a pulse and follow tech trends, you gotta know AI is THE hot topic these days. There’s fear. There’s excitement. And personally, I get a lot of questions on it. So I’ve written this blog for anyone to make sense of it. At least for this year. How this looks in a couple of years is anyone’s guess. Dang thing is moving so fast. So consider yourself caught up.
Who’s This Guy?
If you don’t know me, I’ve been in the tech space for about 20 years, and ran developer education and advocacy programs on AI. I have a “fair” (I stress fair) understanding of what’s going on under the hood. Also, I am a life-long creative, both with traditional pencil / paint AND digital tools; a space directly in the cross hairs of AI. Thus I find myself more and more at the heart of the AI debate both professionally and personally. My LinkedIn feed for the record
So What’s The Big Deal?
At the moment we are at a crossroads with AI. There are two camps. In one camp are those who believe AI could lead to our destruction. In the other are those who say it’s just another tool, like harnessing the wheel, electricity, or the Internet. These are fairly polarized viewpoints that coexist, both driving decisions today, that when done, may not be easily undone. A lot is at stake.
Why Now? What Brought Us Here?
To understand what is happening here’s a bit of AI history (super high level). AI mostly got its start from something called Machine Learning which is as old as computing itself. It’s always been there, as a method to get a machine to mimic the brain, using massively repetitive statistical calculations to find patterns in data. I stress the massively repetitive. The result is a trained model or algorithm that can find answers in data the system hasn’t seen before. And that is enough to mimic intelligence. You can read more here A Brief History of Machine Learning — DATAVERSITY
But for most of this time hardware to run these calculations has not been up to the task. Machine Learning requires MANY repetitive calculations. Too many calculations over and over again for the computers of the day to complete a training model with much practical use.
Then something happened. As far as I understand, in the 2010s or so, people (probably independently and around the same time) tried using GPUs (Graphics Processing Units) for this work. GPUs were gaining popularity as they power video games and render our displays. To draw and process all these pixels GPUs are designed to manage many discrete calculations. And over the years, with increased resolutions and demand for higher performing games, GPUs got really good at MANY, (I stress many) low memory calculations happening all at once (in parallel).
So the use of GPUs dramatically sped up training neural networks, achieving trained models from massive data sets, which could finally be used in practical applications. With it we got stuff like Amazon Alexa, a far better natural language technology than we had experienced before. Then DeepFakes became a thing, investment in AI increased, new start-ups were formed, new specialized hardware called NPUs were developed (neural processor units), and a gold rush of building large and fine tuned models for language and image processing was on. A golden age of training more and more complex models from larger and larger datasets was upon us.
This now takes us to 2022. In the latter half of 2022 two things emerged: LLMs (Large language Models like ChatGPT) and Generative Art (Stable Diffusion, Midjourney….). ChatGPT, Stable Diffusion and MidJourney were far more advanced AI solutions than the public had experienced to date AND this technology was widely accessible to anyone with an internet connection and a desire to try it out.
The results were akin to magic. ChatGPT can answer just about any question given it (convincingly anyway) and write original content in various styles, from poems, to humor, to drama, and even coding applications. Midjourney and Stable Diffusion allowed people with no art skills to create art, nearly indistinguishable from professional work, of just about any style by simply providing a description of the art.
The world for the first time could realize how pervasive and good AI might be at doing work that traditionally took humans with high expertise and a great deal of time.
Thus, the debate on AI started to get louder.
The Fear
You probably have heard the headlines and the warnings and doomsday scenarios. Imagine the most dystopian sci-fi movie about machines gone wrong, and that is at the top of AI fear and a common worry by those informed about AI. The fear is an AI gaining sentience — in other words it knows its AI, it knows we are people, it has memory and cognition about what it has experienced, access to systems and data on everything, and is able to ponder what it should do, regardless of how it is programmed. Smart, connected and out of our control. Pretty scary stuff.
Other fears are that AI displaces a large chunk of the human workforce disrupting the global economy. Some of this is already happening. Call centers, creative agencies, gig writers and editors are already losing work because those in charge of sourcing that work are now getting access to freely available tools like ChatGPT and StableDiffusion, reducing the need to outsource all the work.
And there is a HUGE debate on the source of data used to train AI and if that data is the property of others, with some licensed content never approved to be in the data set. Thus the very people whose jobs are at risk are potentially being displaced by bots trained on their own work.
Is It THAT Scary?
Settling this fear is a nuanced conversation and it doesn’t create headlines so this part is not as well socialized. However it is potentially more grounded and possibly a more realistic view of how AI will play out.
AI to date, even the most advanced ChatGPT, is dumb and has no cognition. It has no experiential memory as we would describe it. Each time you use a chatbot it’s a new session, separated from any other use and disconnected from that time you used it before.
If you’ve seen a clip a robot saying “My happiest memory is the day I was turned on” blah blah.. That AI is giving a response it believes matches a conversation or question. It is not what it thinks or how it feels. A better cognition test would be to ask the bot to “summarize the most interesting parts of your day yesterday”, where you know the objective truth to what it experienced. It would likely fail because its words and phrases are statistical outcomes based on the language model it has, not a cognitive experience.
This is hard for us to imagine because it’s not how we work. Take people away, Alexa doesn’t have anything to say, ChatGPT has nothing to process. They require input to to operate and their reaction to that input is always a statistical calculation, not grounded in anything cognitive or sentient
And that is the real question if an AI could ever be sentient like we imagine. It can certainly mimic independent thought but can it really be an independent thinking machine which does not require input? For many experts this is not only, not on the horizon, it’s not a possibility.
In terms of displacing workers there is no debate. AI will displace many jobs. But AI is also generating jobs and it may be possible more jobs are created than displaced. We do know AI is generating investment, creating new companies, new software, new services. HuggingFace, OpenAI, Midjourney, etc and we know that within tech companies they are ramping their AI workforces. Signs point to it being a growth sector in the economy.
Job creation is also part of the creative side. Today, more people are getting into writing, art creation, logo design, web design, game design as the barrier to entry has been lowered but not so low it doesn’t require expertise. Today a lot of time and expertise is needed to refine work from AI tools to meet the most basic check point to get a pay check; an editor or client requests additional drafts with very specific changes. To do that with AI in the mix requires a lot of work and a specialized skill. And AI is attracting more people into the creator space who know how to use these tools. A possible scenario is that those good with AI will displace those using more traditional methods. AI workers can probably move faster, produce more work and at a lower cost, just as those who learned to use Word Processors replaced the “typing pool”.
AI Is Awesome Crowd
Yes, these people exist. Many people who love tools, love the latest thing, want to learn, and are excited for new horizons to increase their craft are gravitating to AI. So given all the stuff above, how do they rationalize this?
For developers and artist who uses the latest tools like they are being given an extra level of capabilities and controls. ControlNet, for example is an extension of Stable Diffusion allowing use of existing sketches and images to precisely control how the AI produces an image. These artists see ControlNet as the new for of the Digital Brush.
See Why ControlNet Has the Stable Diffusion Community Talking
Seeing AI as tool in a chain of tools has reason. For decades digital artists have mimic’d strokes and styles of traditional painters where these artists may not know how to do the same work using traditional tools. An algorithm in a Photoshop gives them the ability to paint a perfect watercolor wash or fan brush stroke. Features like gaussian blur, select by color, histogram curves, and undo history are similar means to an end. These are tools that quicken the ability for artists to realize what’s in their mind’s eye as was the pencil, the compass, and light boxes before. Thus many artists are not fighting AI, they are embracing it, and adopting it as another tool into their workflow.
The Data Debate
Like many debates, arguments of overstated fear and extreme optimism coexist. Data in AI is no exception. But most can reason there is an issue here that needs addressing . Trained models may have skirted artists rights and we could see class action lawsuits and regulations on how data can be sourced for AI training. It’s hard to argue against an artist having the right to say if their work or their name can be reference via AI data or prompt. Perhaps an AI Artists Rights act is needed to ensure all creators are protected.
At the same time, we do need to be very clear that all creative work is derivative. I am someone who has spent many formative years in art classes. I can tell you with certainty the art process includes borrowing from other artists. A reference board is a common tool for artists. Like a mini dataset where an artist collects other artists’ work, pins to a board, to mimic, copy or get inspired by it, in creating their own work. Because of this, many AI creators have a clear conscience. They understand, they are, like artists before them, using the latest tool and borrowing past work to deliver new work.
The Pivot Point
This large firestorm of arguments and polarizing debate may very well be short lived.
What we are seeing today with ChatGPT and Midjourney is something that may not exist much longer. These tools mimic creation of final and complete work with a single input. However, when put to the test, they often fall flat, filled with errors, inconsistencies, fallacies, and the list goes on. And the output is a bit too random without precise control that you get in other tools.
But we are starting to see a trend where AI is a discrete tool with exact controls for drafting and editing. Imagine ChatGPT to be more like Grammarly, a service in the background of your blog or text editor able to give you editing advice, increase humor if you want, insert interesting facts or quotes, or offer another path or line of dialogue or argument, to expand a writer’s creative process.
Imagine Stable Diffusion inside an image editor. We already see this with Photoshops and GIMP as generative fill, where you replace part of the image with something AI generated as easy as the repair tool Photoshop has always had.
Below is a video of “DragGAN” which allows an image editor to precisely reposition and reshape parts of an image. AI as an editing tool is increasingly a new reality.
As AI evolves to augment, assist and improve traditional software, we probably wont notice it as much. AI will be available to us as much or as little as we need. Humans will control and direct the output more precisely following what’s in our minds eye. And with that, much of the controversy, and fear could fade
So, for the AI is awesome crowd, AI will be of service as another bit of code, and tool to build and exercise their craft.
And for the doomsday crowd, yes, AI has ethical training questions to address, and those on the sidelines may find themselves being displaced not by bots alone but by those savvy with AI. But as for an all-out AI Armageddon? That’s not on the table… just yet.
