AI: an Industrial Revolution?January 10, 2023
ChatGPT, MidJourney, etc, have been out for a while (in AI terms)
Plenty of bytes have been dedicated on Twitter to how it will now replace designers, developers etc. (and it's time to buy their course on becoming a prompt designer!). And I am writing this to see how badly it dates in 6 months, a year, or five years.
Replaced is the wrong word. These models are another gateway to a whole new suite of tools we have to create new experiences. I am optimistic about what this technology may lead to.
It's equivalent to another movable type or desktop publishing moment. The technology was exciting, but what people did with it and how creation and consumption changed is interesting.
I can make grander assumptions. Is this a dawn of a new Industrial Revolution?
Will we see increased leisure time, enabling a new swathe of ideas blocked by our current approach to work? Or will it continue to consolidate wealth and power?
- Further Democratisation of the practice.
- More time for leisure and creativity
Democratisation of knowledge
Providing teaching and learning resources to people who can't take courses for whatever reason, e.g. time, finances, local availability.
Alternatively, a chance to keep up-to-date and up-skilled in your current vocation.
While Google provides access to a vast amount of our knowledge, AI can be at our side to help us understand it.
Redistrubition of time
Allowing individuals to explore more options, routes, and ideas than they could on their own in any given project timeline.
Automating repetitive tasks or reducing the time taken is a massive boon to creative work. Tools like Github CoPilot, once they have learnt my style, have been incredibly useful in reducing the time to compose design tokens.
On the other hand, its issues have to be resolved and considered.
As a role's responsibilities and expectations change, along with what tools are expected to use, current practitioners will be alienated. Design has had enough bumps already with the swathes of different titles and changes in tools. Whether it's Photoshop to Figma, or Quark to InDesign, learning a tool that is similar but slightly different is a costly activity.
Accuracy and being plain wrong
Any tool that provides explicit solutions (code as an example) needs some form of accuracy indication and/or citation. Following the breadcrumb of ideas through references is an excellent way of building a bigger worldview.
ChatGPT is prone to presenting incorrect solutions with such confidence (arguably, this is a perfect simulation of human behaviour)
It is only as good as its input.
This is twofold. Does it have enough material to provide a holistic, balanced output?
With more people creating with it, will the input (the web) become flooded with even more of the same content? Further, extenuating the echo chambers?
I'd like to think we are some way away from serendipity, that seemingly randomness of how two different thoughts can form an interesting and novel idea. But there may be a ShowerThoughtGPT in the works right now.
Control, power, compensation
Will a few small organisations control the services? Leading to issues of accountability and transparency of what powers the models. We don't need another Facebook, Google, Amazon.
Copyright and permission to use human-created content as a basis is a problem. Organisations building their models via their services must be upfront about this and provide routes out and compensation to creators.
What I am hoping for
I am after a few things. As an independent designer/developer, I often work on projects without peers. Having the option to simulate collaboration and learning is really beneficial to my practice.
- The continued reduction in time to complete project setup
- On-going learning, exposing or helping me understand new methods.
- Feeding a tool sketches, references, notes and getting stimulus back.
"Hovering art director: mode, asking for variations of layouts and allowing me to remove and push what I think is right. Some of this is already happening in the Figma space, and I expect it will ramp up with demos and working products over the coming months.
The above two would be fantastic from my time but are the most troublesome without the service compensating where the model got its ideas from. GitHub is already facing legal action over CoPilot
However, this raises a question about what difference would this be to say the mood-boarding/referencing process that design projects go through?
In the future, will this change the economics of design? Could an individual or small studio compete with a large agency? It isn't just you anymore; you have a whole team of creators!
With that logic, will it mean that large agencies can take on even more work, pushing out small studios and independents? Jobs that may have been passed on could be automated.
All of this is a what-if. Currently, it is difficult to use the service due to demand; building your own is prohibitively expensive (OpenAI has had billions of funding). And it may all be a hot minute, filling the noise web3/crypto has left unfilled with its crash. Right now, the ideas are speculative; implementing them will take a lot of thought and, ironically, people making them.