When artificial intelligence is used to support or make important decisions in areas such as healthcare and public administration, it becomes crucial to understand how these systems arrive at their conclusions. A new doctoral thesis from the University of Gothenburg presents a method for designing AI systems that can explain the evidence underlying their conclusions.
Today's AI assistants are based on large language models that generate responses from statistical patterns in text. However, these systems can also "hallucinate" and produce information that appears credible but is in fact incorrect. One high-profile example occurred recently when Sweden's Deputy Prime Minister Ebba Busch shared a quotation that an AI had falsely claimed to have found in two separate sources, when in reality the quotation had been fabricated.
In his doctoral thesis, Alexander Berman demonstrates how AI systems can be designed to explain, in dialogue with the user, the facts, rules or observations that underpin their conclusions - much as people do when they justify their reasoning.
- A key difference compared with language models is that the explanations genuinely reflect how the system arrived at its conclusions, rather than simply sounding convincing, says Alexander Berman, a doctoral student in computational linguistics at the University of Gothenburg.
Transparency is crucial
Instead of relying on a language model alone to generate both answers and explanations, the proposed method allows the system to access information about the evidence underlying its conclusions and use that information when explaining its reasoning to the user.
Language models can still be used as components within the system, for example to interpret the user's questions or adapt responses to the specific context, without compromising the system's reliability.
- Reliability and transparency are essential when AI systems are used to make important decisions. People must be given the opportunity to assess whether a decision is well founded, says Alexander Berman.
He hopes that the research will contribute to the development of AI systems that people can genuinely trust, and that the thesis will inspire further research in the field.
The thesis What do you base that conclusion on? Grounding explainable AI in human dialogue strategies will be defended at a public defence on 4 June at 10.15 a.m. in Jubileumssalen, Humanisten, Renströmsgatan 6, Gothenburg.
The defence can also be followed online via Zoom:
https://gu-se.zoom.us/j/67994493431
The thesis is available online:
https://hdl.handle.net/2077/91236
