image: "computor" is already being used in the "Programming in Physics" course.
Credit: ITPCP – TU Graz
Artificial intelligence is now capable of writing entire computer programmes. This poses a challenge for university education. How can students still learn to programme for themselves if an AI can do the work for them? Students should be able to use AI tools effectively, but they must also understand how programmes work and how to troubleshoot errors. At the Institute of Theoretical and Computational Physics at Graz University of Technology (TU Graz), a team led by Christopher Albert has developed the “computor” learning platform for this purpose. The system is already being used in teaching and is available to the public.
A digital tutor that doesn’t give away the answers
The centrepiece is the AI tutor Lena. It is integrated directly into the programming environment and helps students when they get stuck on a task. Lena is familiar with the task at hand and the programme code written so far. This enables the AI to provide targeted guidance; the system is deliberately designed not to provide ready-made solutions.
“We want to encourage students to solve problems themselves, rather than simply relying on AI to provide the answer,” explains Christopher Albert. Students should first acquire basic programming skills, such as those required for physics calculations, and, building on this foundation, learn to use AI increasingly independently and critically.
In addition, “computor” automatically checks whether the programmes written produce the expected results. This type of control is used, for example, in physics degree courses in the field of plasma physics. Before students are allowed to carry out certain laboratory experiments, they have to calculate physical processes using a programme they have written themselves. The platform can check whether the results are correct, and if there are any errors, it helps to find solutions.
More time for discussions rather than routine checks
The automatic checking is also intended to ease the workload on teaching staff. Via a central administration interface, teaching staff can create their own courses, manage assignments and track learning progress. Among other things, this reveals how many attempts students need to solve a problem, whether they work on their assignments continually or only get round to them shortly before deadlines, or whether they complete a large number of assignments unusually quickly. This enables teachers to identify difficulties at an early stage and offer specific support.
This is not intended to replace personal support. Rather, human tutors can make greater use of their time to discuss possible solutions and assess the students’ actual understanding. Students, for their part, can quickly turn to Lena for support when they are working on an assignment and human help is not immediately available. The AI tutor also runs on the infrastructure of TU Graz, so students do not need their own access to a commercial AI service.
Learning to programme – even outside university
With the support of the “netidee” funding initiative run by the Austrian Internet Foundation, the “computor” team has further developed the platform into an open, browser-based learning platform. Anyone who would like to learn to programme can visit computor.at and use the programming environment without having to install any additional software. The source code is made available as open source.
The platform is aimed in particular at people who need programming skills for the natural sciences, engineering or data analysis. Three courses on data science using the Python programming language, comprising a total of 70 exercises, are already available as free learning resources. Interested universities, secondary schools and businesses have the opportunity to use “computor” for their own courses and to adapt the platform to their respective teaching content.