Add thelocalreport.in As A Trusted Source
As almost all Australians say they have recently used artificial intelligence ,aye, toolKnowing when and how they are being used is becoming more important.
Consultancy firm Deloitte recently partially refunded money to the Australian government because a report they published contained AI-generated errors.
A lawyer also recently faced disciplinary action after false AI-generated quotes were found in a formal court document. and many more universities are concerned about how their Student Use AI.
Amid these examples, a series of “AI detection” tools have emerged to meet people’s need to identify accurate, trustworthy, and verified content.
But how do these devices actually work? And are they effective in detecting AI-generated content?
How do AI detectors work?
Several approaches exist, and their effectiveness may depend on what type of material is involved.
Detectors for text often try to predict AI involvement by looking for “signature” patterns in sentence structure, writing style, and predicting certain words or phrases being used. For example, the use of “delves” and “showcasing” has skyrocketed since AI writing tools became more available.
However the difference between AI and human patterns is decreasing. This means that signature-based tools can be highly unreliable.
Detectors for images sometimes work by analyzing the embedded metadata that some AI tools add to the image file.
For example, content credentials inspection tools allow people to see how a user has edited a piece of content, provided it was created and edited with compatible software. Like text, images can also be compared to verified datasets of AI-generated content (such as deepfakes).
Finally, some AI developers have started adding watermarks to the output of their AI systems. These are hidden patterns in any type of content that are invisible to humans but can be detected by an AI developer. However, no major developers have yet shared their detection tools with the public.
Each of these methods has its own shortcomings and limitations.
How effective are AI detectors?
The effectiveness of AI detectors can depend on several factors. These include what tools were used to create the content and whether the content was edited or modified after generation.
The tool’s training data can also affect the results.
For example, the major datasets used to detect AI-generated images do not contain enough full-body photographs of people or images of people from certain cultures. This means that successful detection is already limited in many ways.
Watermark-based detection may be good enough to detect content created by the same company’s AI tools. For example, if you use one of GoogleAI models like Imagen, Google’s SynthID watermark tool claim to be able to recognize the resulting output.
But SynthID is not publicly available yet. For example, if you create content using ChatGPT, which is not created by Google, it still does not work. Interoperability is a major issue among AI developers.
AI detectors can also be fooled if the output is edited. For example, if you use a voice cloning app and then add noise or reduce the quality (by making it smaller), this may mess up the voice AI detectors. The same is true with AI image detectors.
Clarification is another major issue. Many AI detectors will give the user a “confidence estimate” of how certain it is that something is AI-generated. But they don’t usually explain their reasoning or why they think something is AI-generated.
It is important to understand that it is still early days for detection AI, especially when it comes to automated identification.
A good example of this can be seen in recent efforts to detect deepfakes. The winner of Meta’s deepfake detection challenge identified four out of five deepfakes. However, the model was trained on the same data it was tested on – somewhat as if the answers were seen before taking the quiz.
When tested against new material, the model’s success rate dropped. It correctly identified only three out of five deepfakes in the new dataset.
This means that AI detectors can and do get things wrong. These can result in false positives (claiming it is AI-generated when it is not) and false negatives (claiming it is human-generated when it is not).
About the authors
TJ Thomson is Senior Lecturer in Visual Communication and Digital Media, James Meese is Associate Professor in the School of Media and Communication at RMIT University and Aaron J. Snowswell is Senior Research Fellow in AI Accountability at Queensland University of Technology. This article is republished from Conversation Under Creative Commons license. read the original article,
For the users involved, these mistakes can be devastating – such as a student whose essay is rejected as AI-generated when they wrote it themselves, or someone who mistakenly believes an AI-written email came from an actual human.
It’s an arms race as new technologies are being developed or refined, and detectors are struggling to keep up.
Where to go from here?
Relying on any one tool is problematic and risky. It is generally safer and better to use different methods to assess the authenticity of a content.
You can do this by cross-referencing sources and double-checking the facts in the written material. Or for visual material, you can compare suspicious images with other images taken during the same time or location. You can also ask for additional evidence or clarification if something seems suspicious or questionable.
But ultimately, trusting relationships with individuals and institutions will remain one of the most important factors when detection tools fall short or other options are not available.