Artificial intelligence can be fair

“Machines will dominate the world only if we humans allow it – if we train them to break the rules,” says Ioannis Emiris, president of the board of directors and director general of the Athena Research Center. As the only research center in Greece specializing in digital sciences and technologies, Athena plays a pivotal role in the country’s leap toward artificial intelligence (AI).
“Machines,” he explains during a quiet lunch in the northern suburb of Kifissia, not far from his base, “are indeed becoming smarter than us; they can solve problems they have never encountered before. However, it is by no means inevitable that they will turn against humanity if we implement proper safeguards.”
Artificial intelligence, he notes, is evolving uncontrollably – and rapidly. “However, it will become more specialized across fields. Interest in AI will eventually cool down, and it will become commonplace – ubiquitous, essential, but ordinary.” Emiris speaks with simplicity and calm authority, easing the anxieties of the uninformed. He exudes the composure of someone who has introduced generations of students in the USA, Japan, France, Switzerland and Greece to the complex world of artificial neural networks. His contributions include groundbreaking work in computational geometry, algorithms that solve mathematical problems quickly, and cutting-edge research into 3D artificial intelligence.
When pressed about fears surrounding AI, Emiris addresses the challenges of bias. Algorithms, he explains, are trained using data that may reflect societal prejudices or flawed human attitudes, which can lead to harmful decisions. Can we create algorithms free from such issues?
“We can,” he affirms. “I am currently coordinating a European project called AutoFair, which focuses specifically on this: developing methods for automated machine learning to ensure greater fairness in decision-making systems. Of course, this is highly complex, because it’s not enough to simply hide attributes like gender, ethnicity, or income. You need an algorithmic solution from the outset – that is, you must create better algorithms by addressing foundational scientific and research questions. To achieve fair AI, we need strong basic research. Without it, AI risks exacerbating inequality and limiting access.”
He continues: “Basic research also provides a competitive edge. For instance, it’s what made Google the powerhouse it is today. Stanford PhD students developed an algorithm in a course under the mentorship of Indian professor Rajeev Motwani, a pioneer in randomized algorithms. Do you know his story?” he asks.
When I admit I don’t, he recounts: “Motwani, who was also my professor at Stanford, became a shareholder in Google thanks to his mentorship. He amassed great wealth, built a swimming pool at his home, but tragically, he didn’t know how to swim and drowned. He was only 47.”
His studies
What shaped Emiris personally? “My family and my studies in the US. The surname Emiris originates from the island of Kasos. My grandfather’s family, who had emigrated to Alexandria, carried the diasporic spirit in their DNA. My father, who grew up in Egypt and studied engineering in France – he saw a mountain for the first time at 20 in Grenoble and in Greece at 25 – wanted me to pursue my education in the US. So, as soon as I finished high school, I left for Princeton.”
There, he earned his first degree in computer science, followed by a master’s and a PhD at Berkeley, before returning to Europe – specifically France. “Why?” I ask.
“We were a Francophile family, and I also wanted to return to Europe. I had idealized it as a place of student life, social engagement, and politics. Life at American universities is intense, heavily focused on time management, even if it retains a certain sense of innocence. So, I found myself in the South of France, in a society that was closed, traditional, and politically conservative, much like Greek culture. Of course, the region itself was exceptional, and the work environment at INRIA Sophia Antipolis – known as the French Silicon Valley on the Cote d’Azur – was remarkable.”
After that came his tenure at the National Technical University of Athens (NTUA) in the Department of Informatics and Telecommunications, along with his leadership role at Athena. The list of the center’s success stories is long. What are the most notable?
“The acquisition of the spinoff Innoetics by Samsung,” Emiris begins. “The pioneering use of open data, particularly in sensitive medical applications, which is a European first. The clusters and incubators of Corallia, which have supported over 800 innovation companies and startups. Some of these companies, like Helic and Inaccess, have soared to success, eventually being acquired by major US tech firms like Ansys and Power Factors. Today, Corallia is also supporting 20 space technology companies.
“Another success story is Meltemi, the Greek equivalent of ChatGPT. It’s in high demand. However, ChatGPT in general must be specialized to achieve over 90% accuracy in critical fields such as justice, healthcare, public administration, education, and industry. It also needs to become more transparent and reliable.”
Emiris highlights the transformative changes at Athena, particularly with the establishment of the Archimedes AI unit. Archimedes focuses on artificial intelligence, data science, and algorithms and was founded with the support of the prime minister and the Greece 2021 Committee. Its creation involved significant contributions from luminaries like Constantinos Daskalakis, Christos H. Papadimitriou, and Timos Sellis. Another milestone was the founding of HERON, the first Robotics Center of Excellence (CoE), which is dedicated to transferring research results into the economy.
Your most dynamic spinoff? “AviSense.ai. It’s powered by researchers from Athena and the University of Patras. The company focuses on autonomous driving technology, developing vehicles capable of ‘seeing’ behind physical obstacles, offering multiple communication channels, and planning the safest possible routes using diverse types of visual information.”
Robot missions
Have you built robots? “Yes. One of our robots operates in vineyards, monitoring plants for diseases. It doesn’t resemble a humanoid – it looks more like a spider. It’s also designed to harvest the fruit. Another robot cleans ship cargo holds, meticulously removing pollutants from compartment walls. For example, after coal has been transported, the robot ensures the space is clean enough to carry wheat. Robots are invaluable for challenging tasks. They can enter fires, descend into wells or oil tanks filled with fumes, and even perform missions in space – the ultimate frontier for innovation.”
What does a good algorithm require? “Computational resources. A strong research paper – the kind that gets accepted at a top-tier international conference – costs around €10,000 to produce. A PhD student needs at least three such papers to complete their doctorate. To conduct experiments and train machines, vast amounts of data are necessary. Machines learn from examples, which are expensive, often insufficient, and require significant preprocessing to become useful.
“We need open data, open science, and open-minded research. This year, the Nobel Prizes in Physics and Chemistry were awarded entirely to experts in artificial intelligence: one for using physics tools in designing artificial neural networks, and the other for computational protein design. This not only underscores the dominance of AI technology in our lives but also highlights that science, at its core, is universal and interconnected,” says Emiris, who trains exceptionally skilled students – some of whom launch companies while still in their fourth year of studies.
Do your students use ChatGPT?
“They use it as part of their research. A paper written entirely by ChatGPT won’t be good – it lacks originality and proper citations. If you ask a student to identify and document the sources ChatGPT used, they’ll end up doing the work they should have done from the start.
“That said, the new version, ChatGPT 4.0, is impressive. It’s more accurate and up-to-date, combining older texts with new online references. But the overall question remains: Who controls these systems? Through a tool you believe is objective, someone could spread fabricated information. We need to educate society about how machines think and the risks they pose.”
Urgent need for ‘green’ AI
“Artificial intelligence consumes immense energy resources. Recently, we learned that Unit 1 of the Three Mile Island Nuclear Station in Pennsylvania, which shut down in 2019 (Unit 2, as you might recall, was the site of the most serious nuclear accident in US history in 1979), is being reopened to meet Microsoft’s growing energy demands for AI. Nuclear energy is being revived to power machine learning systems.
“However, it’s critical that we focus on green artificial intelligence: algorithms that consume less energy, systems that train more efficiently, and intelligent machines that learn with less data. These systems are already capable of generating text, creating images, mimicking voices like Putin’s, producing films, and will soon play a pivotal role in developing medicines.
“AI must be sustainable, transparent, and socially inclusive. This is a monumental challenge – and a legacy we must leave for the next generation. Overall, I’m optimistic. Greece has extraordinary potential and can achieve much more than it has so far.”