…and the quiet solution emerging to save AI from itself.
There’s been a lot of talk about training sets for AI, especially the LLMs that form the foundation of GenAI’s “intelligence,” but most of the discussion is around compensation for copyrighted information used in training, or the impact of the output itself on modern workforces. However, for all the very real importance of these two issues, the biggest question facing AI is not around monetization, copyright compensation, content creation, or even human impact: it’s the danger lurking at the heart of the LLMs themselves. It’s the distortion of information…and the straightforward solution to this enormous problem.
You see, every LLM is based on sets of information. In the past, those sets have been closed: controlled environments of trusted inputs. As Meta and Google move forward, however, they are rolling towards dynamic training sets, much as Google did with search. Just as every user query and click-through informed every search that followed, in GenAI, every input for a new cat picture or a short explainer video will train the LLM further. And as Meta and Google go, so will OpenAI and its competitors, with the entire planet moving to a “dynamic,” or open, training model that responds in real time.
Yet no one is talking about the dangerous elephant in the room. In search, private corporations invented scheme after scheme to distort search until Google just did away with relevance in the name of advertising dollars. Social media was worse: what little controls existed in search engines, social platforms like Facebook, Instagram, and TikTok abandoned entirely, happy to show you more of whatever the algorithm thought you wanted, even inadvertently spreading child pornography, as the The Wall Street Journal so recently reported. You would think, given the dangers we’ve already experienced, our giants of technology would be ready for the very same threat to attack AI…except they’re not. And the looming threat to this new technology is not only very real, it’s so much bigger than anything we’ve faced before.
The Chinese and Russian governments have long ago decided that they need not wage direct war against the United States; instead, they simply focused on getting us to fight each other through disinformation spread through fake social accounts, rampant fraudulent news sites, and bots programmed to instigate conflict. They weren’t wrong. So what happens when they unleash the full force of their technological wherewithal on the dynamic training sets at the heart of Western society’s entire AI infrastructure?
It won’t just be bad cat pictures and parodies of our politicians. Instead, the tools that Americans have come to trust will rip at the very fabric of the truth itself. AI can be distorted to tell people any falsehood is real. That vaccines will kill you; your neighbors are planning a race war; election day has been changed. Bad actors won’t just lead you to the wrong location for a restaurant; they can poison the entire well of information that exists. We’ve already demonstrated the power of misinformation to harm this country. What happens when misinformation is all that’s left? Unprotected AI could allow bad actors to destroy whatever shred of integrity remains in any information at all.
So what’s the answer? We can’t cancel the internet and resort to printed encyclopedias published by the government with a stamp of approval. Instead, we must, for the first time, move faster than our bad actors. Thankfully, there’s one technology ready to serve as the Iron Dome protecting AI: spatial behavioral context.
For years, a few select spatial behavioral technologies have been training to predict your next move or desire based on physical behavior, such as learning what restaurants you like to visit and suggesting new ones, or being able to show you customized destinations on a map suited to what you want to do or where you want to go. But the technology under these engines can go further. Using the intelligence gathered from trillions of human movements, spatial AI can now understand who is real and who is not…and with that knowledge comes the ability to flag and eliminate fraudulent bad actors in real time. To know what is human and what is not.
This is transformational, and it’s not just the next evolution of AI, it’s probably worth twice as much as the current AI market. Think about the economic implications of near-perfect identity verification. Financial fraud cut in half, or more. Secure transactions. Increased protection from cyberattacks. Perhaps even a restoration of trust in other humans by verifying they are real humans, every time. Is anything more powerful than that?
For the first time, we can know who is real and what is real on the internet. And that change is groundbreaking.
The speed at which technology moves means only technology can save itself…by moving faster and smarter. Spatial behavior AI can and will ensure GenAI’s dynamic training sets respond only to real, human inquiries and movements, and as a result, deliver information you can trust. In fact, without Spatial AI’s ability to verify real human activity, GenAI and potential for trillions of dollars in economic value will disappear into a cloud of cyberterrorism and unrelenting bot attacks. The compelling, incredible future that Dario Amodei, Sam Altman and Sundar Pichai are selling is only possible when their non-human systems rely, ironically, on humans. In the process, we may be able to even turn back the recent tide of misinformation everywhere, restoring peoples’ trust in information, and maybe even each other.