AI-Generated Filters in Communication: Opportunities and Challenges

Exploring the Implications of AI in Enhancing Communication and Ensuring Security Artificial intelligence (AI) is rapidly becoming an integral part of our daily communication.

As I was replying to a few emails this morning, I realized I was using the Microsoft Copilot within Outlook and Google Premium AI embedded in Gmail more. I write a few lines for the reply and the "help me write" features enhance the quality and efficiency of communication. I do remove redundant sentences so it is not 100% AI generated. Many times the AI response makes me think more about my response. While Gemini in Gmail does not have this but Microsoft CopIlot has a "coaching" option that can look at your email and give you recommendations for tone, clarity and reader sentiment. I often use the coaching as a second step to a Copilot modified email and I get excellent results.

Prospects for AI-Enhanced Communication

The use of AI to improve communication by using the capabilities within the tool we use will grow and become second nature to all of us . AI can help draft emails, generate reports, and create personalized messages, thereby saving time and reducing human error. There is no inherent harm in using AI for these purposes, and it is not necessary to label every AI-generated email. However, as with any technology, there are potential downsides that need to be addressed.

Potential Risks and Security Concerns

The primary concerns with AI-generated emails is the risk of spoofing and identity concealment. Malicious actors could use AI to generate emails that appear to be from trusted sources, but contain dangerous links or fraudulent information. This raises significant security challenges that need to be addressed to protect users from such threats. Can the existing Cloud Email Security tools handle this?

At the Info-Tech Research Group we have rankings for Cloud Email Security Providers https://www.softwarereviews.com/awards/data-quadrant-awards-2024-cloud-email-security. We have had briefings with the providers. We have seen some of them have AI-Powered detection. We have not specifically asked about detection of AI powered communication and that can be a question for the next briefing. Please reach out if you have AI-detection technology for us to keep your company in mind when talking to our 40,000 members who are tech buyers.

Several initiatives are underway to address these concerns:

• Content identification tools: Google's SynthID, for example, embeds imperceptible watermarks in AI-generated content for identification.
• Platform policies: Facebook requires users to label AI-generated posts.

Key questions to consider:

• Do existing security tools adequately address AI-related threats, or is a new security category needed?
• Should a universal standard be adopted for watermarking AI-generated content?

There is some beginning to this with the The Coalition for Content Provenance and Authenticity (C2PA) is a project of the Joint Development Foundation, a Washington-based 501c6 non-profit, that brings together the efforts of the Content Authenticity Initiative (CAI) and Project Origin.

• What are the potential downsides of such a standard?

My perspective:

While existing security measures can likely mitigate some AI-related threats, continuous adaptation and innovation are crucial. A standardized approach to watermarking AI-generated content could enhance security, but potential drawbacks, such as limitations on creativity and potential for circumvention, need careful consideration.

The C2PA standard, supported by companies like OpenAI, BBC, Google, Adobe, Microsoft—and SCENES (a member since April 2024)—provides a solution.

Google's SynthID

A noteworthy tool in this domain is Google's SynthID, which identifies and watermarks AI-generated content across various media types, including images, audio, text, and video. SynthID embeds digital watermarks into AI-generated content that are imperceptible to humans but detectable for identification. The watermarking process is repeated throughout the generated text, creating a pattern of scores based on the model's word choices and adjusted probability scores. This pattern constitutes the watermark.

Availability and Application of SynthID

SynthID is open-sourced through the Google Responsible Generative AI Toolkit, making it accessible for broader use. If a standard is established for all AI providers to watermark their content, it would enable threat detection providers to effectively identify AI-generated material and address potential security threats.

The Downsides of AI Watermarking

While watermarking AI-generated content offers several benefits, it also has potential downsides. For instance, it may create a false sense of security, as not all malicious actors will adhere to watermarking standards. Additionally, it could lead to privacy concerns if the watermarks are misused for tracking and surveillance purposes.

It could be added incorrectly to non-Ai content. Discussions like thse are already happening on places like Reddit. "LinkedIn is still tagging any graphic I try to post with the CR watermark" "if the metadata of your images doesn’t include any AI-related tags or information that might trigger the watermark" will be issues to overcome

As the use of AI-generated filters becomes more common, it is essential to weigh the benefits against the potential risks. Engaging in a comprehensive discussion about the integrity and security of AI-enhanced communication is crucial to navigating the challenges and opportunities presented by this technology.

Conclusion

AI-enhanced communication offers numerous benefits, but it also introduces new security challenges. By implementing watermarking standards, enhancing existing security tools, and fostering collaboration between stakeholders, we can navigate this evolving landscape and ensure the integrity and security of our communication.