This year Te Pōkai Tara Universities New Zealand released a Pan-university Statement on Open Access. The statement included a goal to:

Increase open access across our university repositories from 48% of published research (current) to 70% by 2025. This is possible if all eight universities deposit all eligible manuscripts into their repositories.

On Friday 8 September I facilitated a discussion among colleagues from other university libraries, many of whom manage our repositories. We had a couple of people from Crown Research Institutes too. I asked the group: what exactly does this target mean? How do we assess it?

As a first step, we worked together using the wonderful OpenAlex tool to create some league tables as a means of friendly competition.

First of all, if we look at all works in OpenAlex (about 20 years’ worth), what proportion of them are open?

All the proportions here are low, which is to be expected as generally speaking the amont of open access publications has increased markedly in the last 5-10 years.

If we look at the rates over the most-recent three years, we get a much different picture:

Interestingly, the CRIs are taking the lead here. The universities have a fairly tight grouping around and just over half of our work being open.

What about the most recent whole year, 2022?

The CRIs remain out in front. For the universities, 2022 is the first year we would see some impact from Read and Publish deals and we do see the 2022 openness is higher than the three-year figures from the previous table. (Digging deeper into the OpenAlex data we can see the number of “Hybrid” publications has increased markedly in 2022 and 2023, a topic for another post perhaps.)

Finally, what about a running total for 2023, as things stand in September?

The trend is very clearly upwards, with the CRIs nearing 70% and the highest univerities topping 60% now.

All these tables are to suggest that there are different ways of assessing the proportion of open — and that when we check makes a difference. It is clear that the proportion is increasing year-by-year.

After creating these “league tables”, the group dicussed a range of things, including:

• There is no “right” way to assess our proportion of open. So it may be that we could use a range of measures, as per the different tables here.
• Embargo periods: one year is the most common, and after two years 90% of journal articles will have passed their embargo period (with the remaining 10% almost exclusively with one publisher). It might be reasonable to assume any accepted manuscripts of closed articles more than two years old could be deposited in an institutional repository without having to determine embargo periods for each and every artilce (excluding that one publisher).
• The data we are talking about here only includes publications with a DOI. For some institutions a lot of research does not have a DOI. This is a limitation of this kind of assessment. But it’s clear we should place this caveat on all these measurements.

When you’re making an open book or resource commonly you will want to rely on things that other people have made. However, it’s quite possible you might struggle to find openly-licensed text or images for something and turn to Genrative AI to make your own version. Here is a quick guide to copyright, Gen AI and making your own open book — not legal advice, yada yada yada.

The first thing to say is that this is a rapidly evolving area and there are some uncertainties. As with everything, when in doubt consult your local copyright expert.

Who owns stuff I made using generative AI? 

With Gen AI services, always read the terms of use. Different services have different terms and this is what determines who owns output from Gen AI.

To use Open AI’s terms of use as an example (as at 25 July 2023), in short, they claim no ownership over anything you provide as input or receive as output (see 3(a)). So, that would mean you might* own an image made by Dall-E or text generated by Chat GPT and can then use it in an open text and, as copyright holder, apply a Creative Commons licence to it.

Wait, what did you mean by “might*”?

All Open AI is saying is they don’t claim any rights in what you put in or what their tools put out. In other words, there might be existing rights or new rights created but they’re not party to those. So, when I say you might own the copyright in an output, I have these things in mind:

1. You shouldn’t input anything that you don’t own in the first place, as that might be an infringing act itself. (Also likely against the terms of use).
2. Some would assert Gen AI is illegally using copyright material for training itself, which might make anything it produces infringing. (This will be tested in the courts, where Gen AI firms will use fair use as their defence. This might succeed under US law.)
3. Owning copyright in output would depend on the degree to which it was original. Copyright requires some form of originality.
4. Output could, by chance, be so similar to something that already exists that you could breach the copyright in that original (their Terms 3 (b)).

It’s also worth noting that different countries have different laws about copyright and things that are computer-generated but that is another area of uncertainty. In New Zealand the Copyright Act specifies that an “author” includes someone who made the arrangements for something to be generated by computer (s 5 (2) (a)).

So, can I use it in my open textbook?

Personally, I would use Gen AI to make something to save me time provided that thing was sufficiently “common knowledge” or not especially original that neither input or output were objectively the same as something that already exists. This, in fact, is pretty much how things work with copyright outside of Gen AI considerations.

I would also recommend (as do Open AI) being transparent about when you’ve used Gen AI to make something.

As an example, if I wanted a diagram representing a standard molecular process, I might be able to prompt an output that used multiple sources to produce something original; however, if I simply wanted to replace a particular process described in only one or two places, and the output closely resembled those, I would be very cautious. The “gut feeling” test is: would someone look at this and be able to say “that is a copy of X by author Y!”