Alignment of Algorithms

AI algorithms

Glasses that tell a blind user what is in his/her path. An AI system that can indicate the authenticity of a work of art based on factors such as composition, paint, and method. Algorithms based on Artificial intelligence (AI) are being used more and more. Tilburg University is an expert in it.

AI algorithms interpret complicated, social issues

Shouldn't we be doing something with AI? It's a question that companies and government agencies ask themselves on a regular basis. The need is there, but more effort and work force is needed to answer the question. Pieter Spronck is Head of the Department of Cognitive Science and Artificial Intelligence (CSAI) and leader of the AI Algorithms branch within TAISIG. Spronck: "AI can interpret complex questions in the field of climate, the COVID pandemic, the energy transition, and developments in healthcare. An algorithm knows how to choose the right one from a huge range of possible solutions and conditions. You can no longer escape taking AI seriously and using it.”

Computing power and competition

"In recent years, the development of AI has been accelerating," observes Pieter Spronck. "This is due to the increase in computational power plus the rethinking of existing algorithms; there is more computational power and there are possibilities that make AI more useful. Globally and locally, it's playing an increasingly large role." That large role also brings tension, observes Marie Postma, she is founder of the AI degree program in Tilburg and, as an associate professor, involved in TAISIG. "Companies want to work with it, want to perform better through AI to be competitive. The tricky thing for them is that AI is developing at lightning speed. For example, in 2020 the world's largest language model, the GPT-3 model, was introduced by Open AI. GPT-3 is able to answer all kinds of questions meaningfully because it is trained on all texts of the Internet through underlying technology, i.e., transformers. Our Department has been applying this technique for several years but industry is still hardly doing so. An ideal scenario is therefore when a researcher is partly employed by a company and partly works for the university like the PhD candidates in the national Innovation Center for Artificial Intelligence (ICAI) labs. For example, our researchers are closely involved in KPN’s Responsible AI ICAI lab.”

Pieter Spronck

An algorithm knows how to choose the right one from a huge range of possible solutions and conditions

AI in development

What is AI anyway? "The domain has been around since the 1950s," answers Postma. "Basically, it is automated information processing, with the goal of building intelligent systems. In the current development, machine learning and deep learning are central." Spronck: "AI is becoming more and more creative. Ten years ago, for example, it was not possible to make a really good translation based on an algorithm: simple, imperfect translations were possible, but not correct ones. Now you throw in a text and it often comes out better than a human translation."

Humans and computer

"Humans are needed to assess the outcomes of AI," he continues. "A computer is far from being able to do everything. AI is used in relation to people, with the goal of helping people and society move forward. In Tilburg, we always pay attention to how something fits into society. We work on improving algorithms so that the interaction with human experts improves." "It is a misconception that AI researchers are only concerned with algorithms, that the rest of the context such as ethical or legal aspects escapes them," adds Postma. "Many AI researchers are working on developing algorithms that actually support experts. For example, when you make a decision based on human experience as well as AI, as in the medical field, you often achieve a better result. For example, our researchers are working with neuropsychologists and neurosurgeons at the Elisabeth-Twee Steden Hospital (ETZ) in Tilburg to improve prognoses for neurosurgical procedures. ”

Expertise more widely available

Tilburg University is home to some 65 AI researchers and offers various degree programs. Postma: "There is a broad need for more people with AI expertise. AI has the image that it is only suitable for the most technical students. However, more people are needed who understand how AI works, also from perspectives such as culture, economy, and society. As Tilburg University, we are also taking steps in that direction."

There is a broad need for more people with AI expertise

Marie Postma - Nilsenova

Collaborations

The university collaborates with a wide range of external parties in labs, consortia, and in smaller partnerships. A recent example is the Region Deal within MindLabs. In close cooperation with the regional business community, five research projects are carried out under the leadership of Professor Max Louwerse, which started in September 2020. These projects include the use of AI for decision support, logistics, and education. Another example is the collaboration between TU Eindhoven, JADS, and Tilburg University in the field of smart industry within the Certif-AI project. In this project, an AI toolkit is developed for industrial applications consisting of algorithms and methods.

Various themes

"The main topics we focus on within AI research in Tilburg are natural language processing, machine learning (including deep learning), agents and robotics, decision making, and computer vision," says Marie Postma. "The applications of these AI core areas are broad," states Spronck. Postma: "A great example is the combination of natural language processing and computer vision techniques to teach a system to learn a language like a child does, through auditory and visual input. An AI system would then generate and understand speech. That results in an application like glasses that translate visual information into speech for blind people."

Challenges

AI algorithms are based on data. Access to good quality data is therefore of great importance. Postma: "In addition, it is difficult to improve understanding of how some powerful AI techniques produce results, the so-called explainable AI. Here, technical researchers can work closely with domain experts in socially relevant fields such as economics, behavioral sciences, and law. Tilburg University has an excellent starting position for this."


Contact