Smarter recommendation systems that really know what I might need

aau/MüllerHow do you teach a computer to think? Not just calculating, remembering or combining information – but real, creative, human-like thinking? This is precisely the question that drives Ali Kookani, a doctoral student in the FWF-funded Cluster of Excellence ‘Bilateral AI’. His research focuses on smarter recommendation systems that are significantly better than previous ones at generating suggestions for which film I might want to watch next or which product I might want to buy.

Ali Kookani wants to make artificial intelligence not only more powerful, but also more human – in the best sense of the word. ‘Today’s language models already seem creative to many people,’ he says. ‘But in reality, they only repeat what already exists on the internet with statistical probability. They mix information, but they don’t have any ideas of their own.’ What is touted today as “reasoning” – for example, when a system presents a ‘chain of thoughts’ for complex tasks – is in reality not a genuine conclusion, but rather an algorithmically simplified puzzle.
Ali Kookani’s research focuses on recommendation systems. Customers encounter this technology in many ways – on streaming platforms such as Netflix, in online shops such as Amazon, or on social media. But as sophisticated as these systems appear, they quickly reach their limits. ‘I sometimes get good suggestions, but sometimes I get completely inappropriate ones,’ says Ali Kookani. ‘For example, when an online retailer continues to recommend other laptops to me even after I’ve bought one – instead of accessories such as a matching bag or mouse.’ The reason for difficulty in recommendations is obvious: people are unpredictable. They change their minds, develop new interests, break with old patterns. This is a major hurdle for machines that rely on past behaviour.
Ali Kookani’s dissertation project is embedded in Dietmar Jannach’s working group at the Institute for Artificial Intelligence and Cybersecurity, where there is already extensive expertise in recommender systems. The institute is a project partner of the Cluster of Excellence ‘Bilateral AI’ funded by the Austrian Science Fund FWF. The aim of Bilateral AI is to bring together the two strands of research in AI development: symbolic AI is based on logic, while sub-symbolic AI involves training artificial neural networks with huge amounts of data. With Bilateral AI, the hope is now to significantly expand the performance of AI systems by combining the strengths of both methods.
Ali Kookani found his way into AI research via a roundabout route. Kookani comes from the Iranian capital Tehran, a vibrant metropolis with over ten million inhabitants. When asked whether he has always been enthusiastic about mathematics, Ali smiles: ‘I was good at it, yes – but not a genius like Stephen Hawking.’
He studied electrical engineering as his bachelor but during his studies, he discovered his enthusiasm for programming – and thus also for artificial intelligence. This was followed by a master’s degree at the University of Tehran, the foremost university in the country with a focus on deep learning. Through a contact with Dietmar Jannach, he became aware of a doctoral position at the University of Klagenfurt – and started his new job in March 2025.
The move from the hustle and bustle of Tehran to a small Austrian town was a big but welcome step for him. ‘I used to spend more than two hours a day on the subway,’ he says. ‘Now I cycle to university in fifteen minutes.’ He can now invest the time he has gained in his research. He emphasises that there is still a lot to be done in the further development of artificial intelligence: “We are all familiar with the scenarios of robots getting out of control from science fiction films. I see little danger in this because we are still a long way from “understanding” AI that can think like humans for many useful technologies.‘ Ethical guidelines are built into the models Ali Kookani works with. ’We want to create models that make decisions with genuine creativity and intelligence – within the framework of responsibility, fairness and transparency.”

.flex_column.av-h1qpuap-80840bc4412f76d88705f663dabdfa23{
border-radius:0px 0px 0px 0px;
padding:25px 25px 25px 25px;
}

#top .av-special-heading.av-mcd07068-ccf3c5366a6ec1e9afca90df1caf6b53{
margin:0px 0px 0px 0px;
padding-bottom:0;
}
body .av-special-heading.av-mcd07068-ccf3c5366a6ec1e9afca90df1caf6b53 .av-special-heading-tag .heading-char{
font-size:25px;
}
.av-special-heading.av-mcd07068-ccf3c5366a6ec1e9afca90df1caf6b53 .av-subheading{
font-size:15px;
}

A few words with … Ali Kookani

When was the last time you discussed your research with someone outside the scientific community?
I believe, it was two months ago that my mother asked on the phone, can you explain to me a little bit of what you do. I was trying to simplify why currently AI doesn’t reason and how we try to empower it with reasoning.
What is the first thing you do in the office each morning?
I check my mail-box which is always full of emails from the university to see if there is something I should take care of.
Who do you regard as the greatest scientist in history, and why?
Though for many it might be Albert Einstein or Isaac Newton, I am more interested in Roger Penrose since he is contemporary scientist with many pioneering theories and a Nobel prize in his pocket.
What makes you furious?
Unlabeled datasets… and people who name their files “final_final_REALversion2.docx”.
And what calms you down?
A well-tuned model, a clean dataset, and the sound of GPUs purring in harmony.
Do you take proper holidays? Without thinking about your work?
I simulate holidays. Mentally, I go offline—but only after running one last experiment. Or five.
What are you afraid of?
Being trained on the internet again. Also, quantum computers lurking in the future.
What are you looking forward to?
I’m looking forward to the moment when AI becomes a true partner in human creativity and problem-solving — not just a tool, but a collaborator. Also, the day when an AI can understand a toddler, write a grant proposal, and fold laundry… all before lunch.
Der Beitrag Smarter recommendation systems that really know what I might need erschien zuerst auf University of Klagenfurt.