A very important part of recent AI technology development is the ” creator ” class of AIs. Generally rooted in an algorithmic technology known as Generative Adversarial Networks (GANs) and related algorithms, these AIs can learn patterns from existing data and use these patterns to not make predictions, but to create new artifacts.
Why is this class of AI worth special mention? In the broader area of AI Ethics, there are many aspects of AI technology that humans struggle with from an ethical perspective. When people think about AI Ethics, the first thought that comes to mind is bias, driven by the (valid) concern that AIs can learn historical patterns that we would rather not repeat, and then perpetuate these biases. Other ethical considerations include the environmental cost of AI computation, the risks of an AI arms race, etc.
So why are creator AIs so special? I would argue they present all the ethical risks considered above, and a few more. AIs that create are entering a space we normally consider reserved for humans – innovation. This particular AI type forces us to consider other thorny issues: can AIs be granted patents?, where will AI creations stop? How would we control them? Should we?
To help illustrate the unique problems and opportunities created by creator AIs, I list three outcomes already present from this technology. I call them the Good, the Bad, and the Ugly of creator AI.
Creator AIs are already making positive waves in everything from music to medicine. In art, AIs have been able to create beautiful artworks, sometimes from inspiration from other art and sometimes simply from a text description. AIs have also generated text-based creations from poetry to literature – with varying degrees of quality. Medicine may be the area where AI-generated content can have the most profound impact – by creating synthetic data that can help train disease detectors without concerns for patient privacy. This particular application of creator AIs can make the adoption of other AI technologies in medicine easier and safer. Creator AIs can also create new medicines, opening up a space for solution discovery that can extend the capabilities of human experts.
The same technologies that can create realistic synthetic medical data can create other realistic data about humans, but this time for nefarious purposes. DeepFakes, for example, Face-Swaps, can create realistic-looking fake “evidence” that can be used to misinform, blackmail, or otherwise damage individuals, business, and even governments. There is at present a technology race between tech that creates these artifacts and tech that can detect them.
What can be worse than misinformation, you ask? Creator AIs that excel at creating solutions to human-defined problems will not just create new medicines. They can also create new weapons. A recent article showed how an AI can create new biological weapons in a fairly short time frame. From the article – researchers demonstrated “ in just six hours of operating their model produced more than 40,000 potential deadly molecules, including some closely related to nerve agent VX”.
As we ponder the ethics of AI, we may need to take special consideration of creator AIs. In the future, our best solutions, and our worst problems, may come from this same place.