Could your AI software generate problems?

Some of our largest and most challenging existential questions right now involve artificial intelligence. But there are also some more straightforward ones. These have less to do with machines superseding humans and more to do with getting stuff done – such as “could ChatGPT help us with our newsletter” and “could we be using AI to get our marketing costs down?”

Let's look at some of the intellectual property challenges and uncertainties that might come bundled with your new generative AI software.

First things first. Generative AI is a type of AI system that can be used to generate new content, including text, images, audio, etc. The systems are trained on a selection of data and can then use that data to generate new content which is different from the training data.

This offers a world of possibilities to businesses, but as ever in new territory, there are some unknowns and uncertainties.

In New Zealand, copyright protection for computer-generated works hinges on human creative input. However, the extent of this input remains ambiguous. Content created entirely autonomously is unlikely to be protected, but the threshold for what constitutes sufficient human involvement to qualify for copyright protection is unclear. Is it as simple as plugging in a request for a basic landscape with a house, a few clouds and a road? It’s unlikely.

Even if there has been sufficient human input, if the output is not original, or only marginally original, then there will be limited or no copyright available in the work.

The US Copyright Office has so far taken a conservative approach to computer-generated work. In a recent decision, an author could not claim copyright in computer-generated art for a comic book, despite the hundreds of descriptive prompts provided to the generator to get a rendition of the work. Instead, the Copyright Office said the generator “generates images in an unpredictable way”.

In Japan, the government has recently decided to expand fair use to include copyrighted works for training certain AI, such as ChatGPT, even for commercial purposes. Whether other countries will adopt a similar approach remains to be seen.

We have yet to see how this will play out in New Zealand, particularly given our Copyright Act recognises computer-generated works. In the meantime, keeping an accurate record of inputs, including prompts, re-prompts, and personal contributions will ensure you have a clear record of the ‘human input’ provided, should a copyright question arise.

Businesses should conduct due diligence on their chosen AI service provider before using any generation service. Like most software apps, free or cheaply available platforms are unlikely to offer much – if any – protection against privacy or intellectual property concerns. 

It is important to understand the terms and conditions of the service provider. Key things to look out for are:

Training Data Legally: Has care been taken to ensure the AI has been trained using content that is in the public domain (i.e. not copyright material) or otherwise obtained legally? If not, using content generated by the platform could infringe copyright, especially if the algorithm is not well trained on a diverse library of content.

Data Usage: Will user inputs or outputs be stored or used to train the algorithm, or to generate content for other users of the platform? If yes, appropriate safeguards should be implemented to prevent users from inputting confidential information which may then be published or disseminated far more widely than intended. 

Content Usage Restrictions: Does the platform have restrictions on how the generated content can be used? If so, it is important to understand the limits to this scope and make sure your use is within these bounds. For example, some platforms require users to identify the content as being artificially generated. Also be on the lookout for any warranties you may be providing about the ownership of generated works you license others to use or publish.

Clear and regularly updated company policies regarding the use of generative AI and thorough reviews of the terms and conditions will be needed.

In one sense, the answer to both the largest existential questions about AI and the more mundane ones are the same: immense potential, but proceed with care.

And, as ever in legal matters, never let your excitement about a shortcut cloud your good judgement.