AI Expert Jacques Alomo: “You Can't Avoid it”

Jacques, what fascinates you about a seemingly "thinking" machine?

Jacques Alomo: In the first step, the machine is not really thinking yet. It's initially just probability calculation, and that on a very simple level. But it is already fascinating: OpenAI has shown in the past that more data performance leads to better capabilities – and especially to capabilities that did not exist before. In the end, however, one must not forget: Even an AI that feels like a real human, that can communicate very casually, calculates each individual token and puts words together to communicate in our language.

We are all already benefiting from AI in our everyday lives

A conversation with ChatGPT can quickly feel like a real conversation. What exactly distinguishes us humans from AI?

Jacques: AI models are trained with a large amount of data, and this data foundation enables them to engage in conversation. This is how the answers to our questions are generated. The fascinating part is that it's not really that different for us humans. Over the years, we have also been fed a large amount of data, which determines how we respond. Ultimately, my response in this conversation is based on how I have been trained or what experiences I have had as a human. So, perhaps we humans are not that different from a language model. However, when attempting to perform particularly creative work with the system, one can quickly reach its limits. Language models are powerful tools, but when it comes to finding creative, new, and surprising approaches, they still fall short.

Can AI learn to think and make decisions like a human?

Jacques: There is fundamentally nothing wrong with that. The speed at which we have reached today's level in recent years is a veritable quantum leap. I notice more and more how AI accelerates our daily lives. You don't even have to engage with it closely to benefit from it, for example, with a service hotline where you have to wait less because an AI has already collected all the necessary data. There are so many small things where AI can make a big impact. With a larger data base and more powerful processors, the results become significantly better.

What can the technology already do and where does it reach its limits?

Jacques: In knowledge work, such as working with text and data, a lot is now possible. AI has long been present in areas like object recognition in factories, where it's used to check whether produced parts meet quality standards or need to be sorted out. AI has already been successfully integrated into many production processes. The topic of automation is also very exciting, as it can simplify or automate data processing tasks, for instance, in email dispatch. Another aspect is the generation of new content, with applications in advertising, such as creating 3D models, images, or audio. However, video remains a challenging area where we're not yet as advanced. It's still quite easy to tell whether a video is real or AI-generated. SORA from OpenAI could change that.

How much AI is already being used in media productions such as advertising, journalism and documentaries?

Jacques: I could imagine that AI is still being used cautiously in films, documentaries and television. This could be because the legal situation still seems very unclear. But perhaps people also want to consciously keep their distance because they want to show real life in these formats. It's different in marketing. They say: ‘I'm proud to be an innovator’, and are taking a very strong lead with AI.

European AI Act: Initial framework conditions for the use of AI

At EU level, the European AI Act is an initial reaction to legally classify the use of AI-generated content. What do you think about this set of rules?

Jacques: We as the EU are the first to create a comprehensive framework for this. However, there are other players worldwide who give their companies all the freedoms and say: as long as we make good progress. I think regulations and clear frameworks are necessary. But I am concerned that too tight a framework could stifle innovation. The danger is that new companies will be founded in other countries simply because they can operate more freely and with less bureaucracy there. Part of the AI Act is that content generated by AI must be labeled. I think this makes sense, but at the same time, I believe this rule will become irrelevant because so much content will be AI-generated. In 90 percent of cases, this will be the case. Even in documentaries – you can't escape it.

How much danger does AI really pose?

Jacques: I would argue that within a year, it will be possible to build models to which I can name a company, and they will independently find out everything about that company. Who the CEOs are, which employees hold decision-making power. The model could clone the CEO's voice, develop a plan for a phishing attack, specifically call an employee disguised as the boss, and order a payment. At the same time, an email with important data is automatically sent. The employee has double confirmation of the order via phone and email, trusts the supposed boss, transfers the money – and suddenly two million euros are gone. I am sure that this type of fraud will increase. Not only in companies but also in private scenarios, where fraudsters try to imitate calls from family members with cloned voices to scam money.

You will show us your AI expertise on July 2 during a presentation at the Munich Film Festival. What can we expect?

Jacques: I will mainly be asking many, many questions, some of which don't yet have answers, simply because this field has developed so quickly and so forcefully. Questions that one might ask oneself when producing a documentary: How do I deal with AI-generated content? For example, in the documentary "What Jennifer Did," produced for Netflix, they used AI to generate photos of a deceased teenager to have more visual material. This naturally raises the question: Is that okay? Or differently: Can I generate a photo of some fictional person and use it in my advertising? What if this person actually exists somewhere in the world? And if this person doesn't want that, but I don't have usage rights to the face because the face was AI-generated? What about cloning the voices of deceased individuals? What about deepfakes? Can I now just create a deepfake of you? I want to open our minds in this direction and make us aware of how many issues there are that are still unresolved.

Will we all soon have our own AI assistant?

Let's take a look into the future. How much AI will be part of our everyday lives in five to ten years' time?

Jacques: That is incredibly difficult to assess. You can either be completely wrong or completely right. Basically, I can very well imagine that in ten years' time, if we want it, we will all have an AI assistant who knows us and our characteristics and who knows how we behave. And who is our personal contact. It can buy things for us, write emails, manage our entire digital and real lives. I would argue that we in Germany are generally a little more skeptical about this type of innovation, but the practicality will clearly outweigh the drawbacks - as has been the case with many other innovations in the past. What is a little more in the direction of science fiction - but you are allowed to think freely - is the question: could we have robots that support us in our everyday lives in ten years' time? Research is currently making great strides in this area too. In the end, the question is: would we bring an android robot into our homes after 20 years of science fiction dystopias? (laughs)

And what would your answer be?

Jacques: I don't even have an Alexa at the moment. Instead, I use GPT4 several times a day. And I'm very excited to see what the future holds in terms of interaction with AI models. The topic of security in your own four walls will become very relevant and I ask myself when I would bring a robot into my home. When would the practicality and added value be great enough? I can't answer that for myself yet.