Judeumeh's Blog

In my first post on this topic, I described how the most striking aspect of the sudden rise in popularity of generative AI models may be the accessibility and ease of use of such applications as ChatGPT, which seem to be approaching a tipping point towards mass adoption. Could this be a much anticipated AI Summer, as opposed to AI Winters of the past? What are the key implications of these developments?

Note: This is the second in a series of posts intended to explore the capabilities, implications and potential ways to adapt and take advantage of opportunities offered by Generative AI. The second carriage explores certain key implications for jobs, IP and ethics, as follows:

Connecting the experience – It feels like a fire of expectation is washing over the AI landscape as frenzied experimentation with generative models, and the applications they support, approaches fever-pitch. There are so many guides, top tips and tricks out there that it is virtually impossible to mistake the air of excitement. Even the much wizened and cynical, er mature, technologists among us, are almost keeling over in contemplation of the possibilities that these technologies promise to bring. So, let the hacking commence, as people start to chain-connect various experiences, e.g.: from text to speech or text to images and back again in meaningful, impactful ways that exceed the component capability. For example, a Nigerian filmmaker & AI artist described the heart-wrenchingly original Inspiration for his use of Stable Diffusion and Midjourney, plus photoshop wizardry to create a most outstanding collection of seniors in a fashion showcase setting!  

All change with new roles and job titles. Think about it. If generative AI can do most of what some people do today, to a high enough quality and accuracy, then no jobs are truly safe are they? Well, I wouldn’t rush to any conclusions just yet. The current state of play is such that AI generated content are not necessarily mission ready, or consistently reliable or accurate enough to hand over the reins just yet. If anything, we still need humans-in-the-loop to ensure the best possible outcomes. However the nature of human work is bound to change over time as AI systems mature, and we’ll likely see a shift towards more human optimal activities within a system (e.g. common sense fact checking and governance based on human ethics and values). New roles will likely evolve, e.g. Prompt Engineering is now a thing – Yes, prompt engineering, an emerging discipline aimed at asking the right questions and creating the optimal prompt for a given model and desired output. According to Amatriain.net blog, this requires a certain level of domain knowledge, and a programmatic approach with an iterative process to achieve outcomes such as fact checking, auto correction, divergent opinions and statefulness-enabled complex conversations. The current crop of generative AI systems may struggle to generate coherent, long-form content, e.g.: books or feature length videos, but that hasn’t deterred some equally brave and/or foolhardy adventurers from attempting herculean feats of prompt engineering. You heard it right, It’s all happening and no holds barred experimentation is the order of the day as the jobs of tomorrow are on the line.

Implications for IP and Ethics – regardless of the excitement and achievements of generative AI, this part won’t be complete without a look at various limitations, constraints and implications, particularly around ethics and intellectual property:

• Starting with the aforementioned generative porn, which has high stakes in both camps, there continues to be a robust debate on the ethics of generative AI content and its use. Key ethical concerns include the intentional or inadvertent generation of abusive, offensive and discriminatory content.
• Also, given that generative AI is often trained on public data, and can mimic the styles of established creators, it stands to reason that it may likely infringe the moral rights of said creators in subsequent AI-generated interpretations of their work or style. 
• In addition, the ability to generate reputation-damaging deepfakes of private individuals or celebrities brings its own set of ethical challenges around privacy, defamation and libel.
• The IP implications are no less problematic – e.g. the question of whether AI should be granted IP rights was a hotly debated topic at last year’s World IP Forum event in Bangkok. Now, even owners or users of content-generating AI may themselves run the risk of having their IP applications refused or revoked if cases like the AI generated comic book are anything to go by.
• In a similar vein, a class action lawsuit filed against the creators of GitHub’s AI powered Copilot claiming software piracy at unprecedented scale, is just one more signal for major turbulence ahead. 
• Google’s MusicML, a text-to-music generative AI application remained unreleased apparently because its generated output sometimes included minuscule direct copies of IP-protected music from its training dataset! We all know what means in the litigious music industry. In fact, this is a consistent issue with many content-generating AI systems which typically source their training data from the Internet without the knowledge or consent of IP rights owners.

I wonder if there should be a new class of IP rights to grant compensation for rights-holders whose contents are included in AI training datasets? Perhaps some sort of micro-IP compensatory mechanism may be in order, but that could prove a nightmare to administer, or would it? Hmmm…

The third and final carriage in my train-of-thoughts focuses on the likely outcomes and next steps (aka Now what?), as we prepare for a dramatically enhanced, AI-augmented way of life.