Partager :

Vidéo Jan 24, 2025

3 Questions with Tomi Heiskanen – The Evolution of AI Translation

Hello, and welcome to three questions with MotionPoint. I'm your host, Dominic Ditherbud. And today, I have the absolute pleasure of meeting with Fluentic's CTO, Tommy Heiskinen.

Tommy, how are you doing today?

I'm doing great. Thank you. How are you doing, Tom?

I'm doing good. Thanks for asking, Tommy. I'm I'm excited to get our conversation going. So let's just get into it. Why don't we?

Sure.

So question number one. What further advancements do you as Fluentic CTO expect or anticipate from AI in the next ten years?

Well, ten years is a really long prediction window given that OpenAI released chat g t p three only four years ago and has already taken a major share of the market, and they are valued just shy of one hundred billion USD.

In the past, there has been phases of rapid development and then some cool down periods, but the current explosive growth started from technology enabling inference of large language and really complex models in real time and some fundamental model architecture innovations. So I would anticipate that development continues during the next two next years. But, given the vast investment in the field, new frameworks and novel inventions will be published.

They will likely, will be major improvements to the current models, which have many fundamental flaws. For example, the other aggressive models are unable to utilize ground truths in their responses and tend to hallucinate.

In three to five years, AI translations are likely to become a commodity to the users, also in voice channels.

Yeah. So you're you're I agree with you. Ten years, maybe that's way too much of a stress for any of us to predict. Right? But, you know, especially for AI translation, I believe, like, the expectation, it was almost like an emulsifier to our to to to the buyers in in our industry to really increase trust in AI, where before, like, machine translation, which is, you know, tangent and it's related to it. But, it was it was a little bit, you know, people had more skepticism, right, towards it. And now I believe it's it's almost like their knee jerk reaction to solving for for these types of of translation needs that they that they have.

So what you’re saying that these hallucinations you believe are soon to be gone and that it’ll be much more commoditized?

It's likely that, the hallucinations will be a problem with the current architecture. But, as the technology evolves, it's it's likely that, there will be new technical technologies that solve the issue and and Mhmm. Come over that. But in every case, I think it's something that will, will will be utilized anyway because there are many use cases where the hallucinations are not causing damage.

So, for example, having a summarization of conversation, where you can always check from the original.

So in in that sense, it’s it’s possible that the AI translations will become a greater thing, in in upcoming years even though there are, some limitations with the current implementations.

Got it. And, you know, as these rapid changes happen, what do you suspect users and the consumer will be looking for in translation solutions as AI evolves? Have you thought about that?

Yes. So going forward with the translations, they will become more transparent part of the business processes that the users might not even acknowledge that there is a translation in the process.

Just like, it has happened with saving. Nowadays, pretty much, all software automatically saves your work in the background to the cloud. Earlier, you needed to manually insert a disk and so on. So, in future, users are are not actively requesting translations, but, but rather consume as part of the automated process.

I also predict that the AI voice translations will see major crowds starting next year.

And in some use cases where the translations has been a middle step before, for a human or information system to process the translated results, the translation step is no longer required as the entry results might be achieved directly with an AI solution, for example, in sentiment analysis or question answering tasks.

On the other hand, where the translation is still needed, the fundamental requirements for good translations are unlikely to change.

For example, the accuracy of the translation is still going to be an important factor. And in the short term, it might prevent the large adaptation of the LLM translation with the current architectural limitations.

That's interesting. I I agree with that. I mean, sometimes, for example, the saving example. n'est-ce pas ?

You you're working. You're going through through things. And now at this point, you just subconsciously expect that your work is getting saved, that it's working fine. n'est-ce pas ?

And that would that would that I believe that that's that's true. And it'll truly scale to almost, being unperceivable. Right? The fact that these are translations that are happening and not so much, like, you know, in language, the the content that's generated, originally, you know, content origination in that language.

That's a very good take. I hadn't thought about that for sure. Well, let's move on to the third question, which is, you know, as these advancements happen, right, and as as translation technology becomes more embedded in our day to day and becomes more imperceivable, as you mentioned before, how important is is privacy and security gonna be in translation technology, but as it pertains specifically to application translation. Right? And the reason we ask that is because in our experience, we've seen that the more, you know, the more secure if it's a login experience, you know, the more technical an ex online experience is, the more privacy and security is valid.

Do you believe that there’s gonna be any changes or or that AI is gonna impact the way people perceive privacy and security in translation technology?

Definitely to some extent. The EU has been long leading the privacy enforcement, and some companies have run into legal issues with it. For example, OpenAI with, Italy's authorities because of the privacy concerns.

But in general, the late late legislation regarding privacy and security is getting stricter every day. And and given the large scale data leaks and then horrifying effects on people, It's going to be a determining factor, on adaptation of the new technologies.

However, in many cases, translation is already done in a system where it’s a single point of failure in terms of security and privacy.

In our patented solution, the prefer personal identifiable information is redacted from other processing, and the customers largely expect such protection of by a PII nowadays.

Overall, the use of AI or large language models won’t make a big difference in terms of privacy and security, but potential usage of the customer data for the training of the AI is a major factor to consider for everyone, and accountability for both privacy and security is key factor in in that kind of translation service.

That's a great point. I appreciate that. Well, Tommy, anything else you would like to add to our conversation today?

We are living in interesting times.

Every day, there's a new model coming out from from some vendor, and and it's hard to keep up. But, there's interesting development and things get better and easier every day. So it's it's really nice.

Yeah, it's really exciting. I agree with that. It's hard to keep up, especially in our industry. I feel like every time I open, I don't know, Ninjie Slater.

There's a new article, a new press release of, you know, a new AI patented approach for from, you know, our peers in the industry, and that's that's brilliant. I do think at the end of the day that, you know, the rising tide lifts all ships. So, as long as, you know, we're we're focused and we we we take it, you know, day by day. It's a lot of efficiency and being thorough in the way that we implement and also serve our customers and and consumers with, you know, these new technologies with our own offerings that, you know, AI in translation is gonna be something that benefits everybody.

So that’s great. Well, that does it for three questions with MotionPoint.

Until next time, thank you very much for watching.