By: Sarah Khan
Artificial Intelligence is now a necessary part of our way of searching, writing, and even thinking. However, as the applications of AI technologies become widespread in the media, educational institutions, and even private life of people, one of their flaws keeps popping up time after time ‘Hallucination’ which is the moment when artificial intelligence produces some false information. Of course, blaming the machine in such cases is quite easy but the hard question is whether people who develop and use these technologies are equally guilty. The above-mentioned issue was the core topic of the latest workshop held under the project titled “Digital Trust and Free Expression in the AI Era”, which is implemented through a collaboration of the U.S. Mission to Pakistan and the Pakistan-U.S. Alumni Network at Department of Media and Development Communication, University of the Punjab, Lahore.
The workshop was aimed at preparing future media workers to fact-check information produced by AI technologies before it is published. One of the workshop trainers managed to put a whole new light on the problem hallucination is not a defect which appears outside the algorithm, but in many cases, it is a built-in mechanism of AI models. Big Language Models are designed to choose the statistically most probable word, not to check the truthfulness of the information provided. As a result, when they run out of data, they do not stop working and fill in the blanks with plausible information. This is the fault of the machine. But that is not the whole picture. While it may be true that hallucinations can be baked right into the model’s mechanics, the reaction to them is largely dependent on the person behind the screen. The AI being used to draft an assignment and the one being used to grade the same assignment is fundamentally the same technology, and yet it behaves in a way that fits the prompt it received, as well as the human behind it. In a sense, therefore, AI does not break down by itself it is a result of the prompt and the intent and the expectation of the human being behind it. If it is humans who train it and control it and decide what to ask of it, then at least part of the responsibility lies with us and not with the system itself.
However, this is far from being the end of the story. As mentioned during the workshop, Dr. Ayesha Ashfaq, the project lead, has shared a perfect example of that very paradox. Having been asked by an AI-powered tool to describe a scene from a film, she got back the most vivid description of a scene which never happened in the film to begin with. The tool was not fibbing to her in a human way; it did exactly what it was meant to do produced a response which sounds confident and coherent regardless of the factuality of its content. The troubling aspect here was not that of AI making mistakes. It was that of a person taking this confident answer for granted.
Indeed, as has been established, the danger of AI hallucinations is only as significant as the willingness of people to accept it and promote it. There is no better illustration to this point than politics. A democratic system becomes weaker when competing political actors intentionally disseminate synthetic misinformation as if it was a fact and try to do it with the help of the new technology faster than ever. In its research on Pakistan’s 2024 general election, the Digital Rights Foundation (DRF) found a drastic rise in disinformation, with AI-generated content playing a significant role in the process and becoming a widespread phenomenon on all major social media platforms. This is not a problem with the machine. This is a deliberate choice of a person to use a flawed technology for his or her political purposes.
This is exactly why timing of such workshops is so crucial. In the absence of appropriate control measures, if AI is allowed to operate exclusively for the sake of gaining likes and clicks or is intentionally misused in the spreading of false information, the consequences of it can become rather dire precisely at the time when journalism is fighting hard for restoring its credibility among the audience. And the objective of this work in this field, as explained by Asad Baig, is to make sure AI can serve as a tool which will allow journalism to flourish in Pakistan in the upcoming years, if it will be used ethically.
Indeed, the issue of balancing between ethical and proper use of AI is the key here. Putting all blame for AI hallucinations on the machine is too convenient; it allows people to avoid their responsibility for it. It ignores the responsibility of the editor who publishes such content without verifying it, of the political operative who spreads false information, as well as of the regular user who shares an answer he or she has received from an AI-powered tool without any critical analysis of it. On the other hand, ignoring the role of technology itself in the process ignores its design and functioning. Finding the balance between those two extremes being mindful of our responsibility as humans while at the same time being aware of AI’s responsibility is crucial. But this workshop series, which seeks to teach the next generation of journalists to fact-check AI-generated content, shows the way in which this balance can be achieved.
The views, thoughts, and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the U.S. Mission to Pakistan or USEFP/PUAN.
The writer is a student of Media and Development Communication at University of the Punjab, Lahore.