AI in Paligo

With all the hype around AI in the world recently, you may have wondered what Paligo is going to do in this area.

Are you eagerly anticipating AI features coming to Paligo or maybe feeling a bit of unease?

Well, that’s what we wanted to know.

So, to begin with, we have certainly been very busy in this area for quite some time now. However, we didn’t want to just jump on the bandwagon and release AI features without seriously considering all aspects.

The first thing we wanted to do was to try to find out what our customers were thinking about AI and the prospect of having it in Paligo. So at the end of summer last year, we conducted a survey with a variety of users in our customer base.

We asked customers from a variety of sectors, in different industries like life sciences, robotics, insurance, manufacturing, software, and more. We also included a wide range of customer company sizes from SMBs to Fortune 100 enterprises.

The survey was carried out with complete confidentiality, so we’re not able to share the results in detail. But they were truly fascinating.

The survey clearly showed that we were right to be excited about AI and what productivity improvements it could add to Paligo going forward, but they also showed that we were equally right in proceeding with caution.

We were, on the one hand, right in being excited about AI….but also about proceeding with caution.

Without going into details, I think some high-level data is interesting to share.

• 75% of the respondents were interested in having AI features in Paligo.
• 12.5% were not interested at all.
• 12.5% had no opinion either way.
• 60% of those that did want AI features still had concerns. The most common concerns (not surprisingly) revolved around security and accuracy.
• About 75% of all users in the survey wanted to participate in providing input and feedback on the development of AI features. This was true regardless of whether they wanted AI or not.

These metrics have been very important in our subsequent work on developing AI.

So the survey showed (as we had suspected) that despite some worry and concern, most users were still in fact interested in having AI capabilities in Paligo. And consequently we have continued to put a dedicated focus on this in our development for some time now.

But we have also tried to use the input from our users to find the right way to address some concerns. We have looked at several ways to do that. In our view, AI is evolving so quickly that we don’t want to lock ourselves into a particular implementation too early. So we have chosen to explore a number of different paths and will continue to do so for some time. One of the paths is the most well-known one, i.e. OpenAI. This is the one being released in closed beta right now. But our philosophy is to be LLM agnostic, and we are exploring others in parallel.

The major concerns were security and accuracy. Security was a given, and of course that needs to be considered carefully.

Of course, accuracy is important, especially considering the examples of hallucination we’ve already seen in abundance. And our core user base, made up of technical writers and content experts, will certainly not tolerate poor accuracy and hallucination.

Technical writers and content experts will certainly not tolerate poor accuracy and hallucination

So we are taking these matters seriously and addressing them in several ways:

• No training of the LLM with customer data: The OpenAI API implementation of the current beta is not using customer data to train the LLM. Other options we’re exploring are considering data privacy matters in similar and other ways.
• AI-adapted approval workflows: In Paligo we will aim to have AI content fit into approval workflows where AI content will be clearly marked up as AI-generated, and require specific approval by a human author. This actually increases security around AI as opposed to users potentially using external tools.
• AI audit trails: The AI features in Paligo will provide a full audit trail of content that originated from the AI assistant, so you can always go back and see this in the revision history.
• And more (to be worked out in consultation with our customers)

This actually increases security around AI as opposed to users potentially using external tools like ChatGPT

Our approach to implementing AI is not only about addressing security and accuracy issues, even if they are important. It’s also to keep in mind that AI-generated content may in some ways be a bit contradictory for a CCMS to begin with.

In fact, the point of a CCMS is, of course, to minimize the generation of content and instead reuse it. When developing AI features, this needs to be considered at all times, so the AI supplements the core features of content reuse rather than indiscriminately over-producing content.

We believe AI features can be a great addition to increase user productivity. And we believe the best way to make it really useful is to implement it in conjunction with you, our customers. You know what would be helpful to you, how you want to get the most out of your CCMS and single-sourcing, and how, as a technical writer and a content expert, you think AI can increase your productivity.

Enabling the AI features is entirely up to our customers. This applies not only during the beta phase, but also when we arrive at the production release. If you are excited about AI features, you can enable it, but if you don’t want AI in Paligo you can leave it completely disabled.

So in closing, we’re looking forward to having you sign up to join the beta program and help us make AI functionality that can really help you as a content expert.