TikTok’s AI message blurs the line between creativity and security

A recent post from TikTok declares its commitment to AI transparency, and introduces new efforts to that effect. The social media company says it will expand AIU literacy efforts with “new educational resources and continued investment in trusted expert partnerships.”

“We know from experts that education is critical to giving people control over their experiences with AI,” says the blog. TikTok’s new guide will aim to help promote the responsible use of AI tools, and a new in-app hub will teach “practical skills for spotting AI-generated content when they search for AI-related terms.”

The company also promises to test improved detection systems targeting so-called AI spam or slop accounts posting on “topics that could pose a risk to public trust or well-being.” In this, it moves toward likeness detection protections others have put in place to defend against deepfakes.

Finally, it is joining the C2PA Steering Committee to double down on its commitment to labelling and content credentials.

Please specify: AI-A or AI-B?

Generally, the post takes care to make sure readers know TikTok is committed to AI. Deep learning and generative AI, it says, “can transform how people share their creativity, discover new passions, and stay safe on our platform when used transparently and responsibly.”

It says the updates are part of “wider work to safeguard and empower positive AI experiences on TikTok.”

The statement exposes an emerging necessity for differentiating types of AI. The blanket term, shorthand for artificial intelligence, has rarely been useful beyond sci-fi and marketing: it is inaccurate, vague, not at all descriptive and burdened with a semantic history. As the technology it wraps is applied more broadly, there will need to be lines drawn between machine learning systems deployed for security on platforms, and those that “transform how people share their creativity.”

The latter generally refers to generative models and AI chatbots that users prompt to generate media content based on what they have ingested as training data. It is most often sensory in nature – i.e., audiovisual. These have a poor track record to date.

Security AI, and LLMs used as coding tools – or fraud engines – are its less overtly visible cousins, operating at system levels. Their purposes are fundamentally different: these prompt creativity of a sort that TikTok does not speak of in its talk of empowerment and “new passions.”

Raising the question: to what kind of AI is TikTok committed? And who will be tasked with sorting so-called good AI from so-called bad AI?

Tick-tock, TikTok: genAI deluge could swamp platforms

TikTok is presently under investigation by UK regulator Ofcom into whether or not its machine learning-based system for estimating a user’s age qualifies as highly effective age assurance. Like many big tech firms, it is selling AI as the future, and does not wish to be regulated out of its pitch, or the many use cases it has spun off for its models. Which means that, much like Meta, it is navigating a script of doublespeak that mixes security and privacy commitments with evangelizing and evasion.

But there is a flicker of concern in TikTok’s commitment to AI transparency, notably in the promised action on AI spam accounts. “Every day we see the amazing ways AI can unlock new storytelling opportunities for our community,” it says. “However, as this technology evolves, it can also be misused to mass-produce spam that crowds out authentic creators.”

The stated purpose is to weed out accounts peddling faked opinions on politics and current events, financial advice, and medical content. But it contains the kernel of a realization: the rule of scale suggests that a platform like TikTok is likely, over time, to be swamped by fake, haphazardly made AI content that nobody wants to watch. The AI boom says to embrace the change – but TikTok may not like what its platform becomes on the other side.

As it stands, it is the last of the Big Social apps to have found traction at scale. It launched a decade ago. The glyphs are on the screen for social media: ultimately, all traditional social platforms may be destined to follow the path of Facebook, toward the swamp of AI posts about vegetables and on-sale cookware that only old people look at.

Article Topics

AI fraud  |  deepfake detection  |  deepfakes  |  generative AI  |  likeness detection  |  social media  |  TikTok