Welcome to the era of interactive AI

At the end of last year, I was one of the DeepL leaders who shared our predictions about the developments in AI technology that we could look forward to in 2025. One of the most important of those developments will be the experience of humans collaborating with AI. I predicted that the interaction between the two will get much more dynamic, with AI understanding users better, and proactively offering suggestions. We’ll see AI that’s designed to work with users, rather than just demonstrate how clever it is. I believe that will turn out to be a very significant shift, indeed.

The news that we announced last week — our new feature, Clarify, a new way to interact with DeepL’s Language AI — illustrates that this prediction is already starting to materialize. 

From delegation to collaboration

So far, for most people, engagement with AI has been a one-way process. In fact, that’s where a lot of the obvious value of AI comes from. When you delegate a task to the technology, AI goes away into its metaphorical black box, works away independently, and comes back with the result you need. You’re not required to get involved, which results in big time savings and increased productivity. 

DeepL’s Language AI delivers a lot of value this way. In Forrester’s Total Economic Impact analysis for DeepL, the platform saved employees 90% of their time from general, translation-related tasks.

However, saving time by taking tasks off our hands isn’t the end of the story when it comes to AI generating value. It can help a lot by completing tasks without our involvement, but it can do even more when it intelligently and proactively asks us to get involved. When AI can identify the moments when it needs human input to produce a better result — and then proactively reach out to said human, to ask for that input — the scope and quality of what it can deliver hugely increases. The experience for human beings working with AI becomes much more valuable and positive as well.

Contextual understanding that helps AI interact intelligently

With Clarify, this happens when DeepL’s Language AI detects potential ambiguities or miscommunications that can occur when translating from one language to another. Those include everything from abbreviations to idioms, to how gender works in different languages, to how people write dates. Thanks to training with specialized proprietary data and insights from language experts, Language AI has a contextual understanding that highlights when it needs more information to choose the right translation. It reaches out with relevant questions to confirm what people mean and how they want to express themselves. This prevents misunderstandings and gives people even greater confidence that their translation gets across exactly what they want.

Clarify is a great feature, but it’s just the start of what we can do with interactive AI. It signals three key ways that interactivity can redefine how we work with this technology.

AI that understands what it doesn’t understand

Firstly, training AI to interact intelligently with us will produce better, more accurate and more human-like results.

As our CEO Jarek Kutylowski explained, when interviewed for a recent Forbes feature on Artificial General Intelligence, we’re still quite a long way off from any AI understanding the world in the same way that a human does. If we pretend otherwise, and insist that human understanding can be replicated by technology alone, we’ll inevitably miss contextual signals and nuances that get in the way of tasks like AI translation. This produces worse results. It hurts quality and accuracy and commoditizes what people produce and how they express themselves, with everyone using AI in the same way, to produce the same sub-par outputs.

Training an AI to recognize when it needs human input is a challenge, but it produces far more efficient and effective results. An AI that thinks with humans quickly exceeds the limitations of the technology. It produces results that are much more human, much more accurate, distinctive and personalized. When it comes to AI, we’ll be able to do more together.

From passively consuming to actively thinking with AI

Secondly, as was recently argued in an interesting Economist feature on using AI for research, very different things happen in your brain depending on whether you’re working with an AI, or just reading what an AI produces.

When we’re involved in generating ideas, we’re engaged in active thought rather than just passive consumption – and this often produces better and more original outputs. This is true if you’re writing a paper on economics, and it’s equally true if you’re translating something into another language. 

Translation should be an inherently collaborative and bi-directional process. It involves understanding how something is likely to be received and interpreted in another language, and then thinking imaginatively about how best to express yourself within the culture that language embodies. Interactive AI can enhance these crucial aspects of the process. Thinking through what you’re saying, and why, helps to express ideas better. Thinking about them from the perspective of a different language might also help to refine and improve the ideas themselves.

Building confidence through collaboration

Thirdly, collaborating with AI in an interactive way is a fundamentally more positive experience for human beings themselves. We get affirmation from actively contributing to the end result. We understand what’s being produced and why, and our confidence in using AI increases significantly as a result.

Working with Clarify is a good example, because the questions that it asks, and the explanations that it gives, help people to take an active role in choosing the translation that works best, and equip them with the contextual insight they need to make those decisions. 

Clarify directs users’ attention to the elements of a translation that can benefit most from their unique understanding. The result is a translation that’s been reviewed and edited by a human, but in a fraction of the time that this would normally require. You don’t just get a more accurate and more personalized translation. You know that you’re getting a more accurate and more personalized translation. That’s the crucial step in properly empowering people through AI.

We’re taking our first exciting steps with interactive AI. It’s great to see the difference it’s already making to the translation experience, but I believe that’s just the start. Interacting with AI models in new ways can encourage us to do more positive things with our minds, not fewer. And that’s something worth looking forward to.