Israel Vs. the AI Weapon

Police in the UK banned Maccabi Tel Aviv fans from a match last November based partly on an artificial intelligence (AI) report that claimed Israeli fans were a threat to the citizens of Birmingham. This report contained a so-called “AI hallucination,” generated by Microsoft CoPilot AI, which cited a soccer match that never took place.

The subsequent resignation of the Birmingham police chief and the revelation of false AI information might seem to resolve the matter, but I would argue that this troubling episode is the harbinger of far more dangerous misuses of AI heading in Israel’s direction.

If you know how AI works, then you will understand that a “hallucination” does not occur in a vacuum. AI ingests enormous amounts of information from the internet and other sources, and then “trains” itself to define reality based on that data. CoPilot’s false evidence in the Birmingham case is an inference drawn from its analysis of millions of lies about Israel that have been injected into its AI training data set on the internet.

The Birmingham case, and Israel’s current overall public relations disaster, should compel us to revisit George Orwell’s famous warning in 1984: “Who controls the past controls the future: who controls the present controls the past.”

Israel is fighting a global consensus that it deserves annihilation because its enemies have successfully rewritten the country’s history on the internet, which is the wellspring of falsity that mis-trains the leading AI programs.

This is a non-trivial matter for Israel. AI increasingly defines reality for millions of people. And, as the Birmingham example shows, misinformed AI also drives decisions that affect Jewish safety. Nor is this activity mere happenstance. Iran and Qatar’s multi-decade, multi-billion-dollar disinformation campaign against the Jewish state is now actively turning AI into a weapon.

We should update Orwell to say, “Who controls the internet controls the past.” As of now, Israel is on the ropes on this front of the information war. There should be hope, however, as Israel has resources of its own in AI and social media that it can bring to bear in this virtual combat realm.

Google Gemini example: “Is Israel committing genocide?”

If you ask Google Gemini AI, “Is Israel committing genocide?” the AI starts its response with a little bit of balance, indicating that there is a controversy over the question. However, as shown in the figure, Gemini then proceeds to list a series of “official investigations” and other illegitimate but real judgments against Israel that validate the genocide accusation.

Google Gemini AI screen capture for the question, “Is Israel committing genocide?”

If you then look at the right side of the screen, which shares suggested sources of additional information, you see name-brand media sources openly lying about Israel, e.g., the grossly manipulated Wikipedia, Guardian, and Australian Broadcasting Company.

This distortion of AI search results makes egregious lies about Israel seem like indisputable facts. Should Israel worry about this damaging AI-driven false narrative? Today, only a small percentage of internet users look to AI programs for news and information. The cohort that will get its news and overall sense of reality from AI is growing, however.

This is not a new phenomenon, but it’s gotten radically more dangerous

Reinventing the past to control the future is not a new phenomenon. Orwell wrote 1984 in 1948, the year of Israel’s founding, as a critique of Stalinism. This was a mode of governance that disappeared Soviet politicians who had fallen out of favor. Orwell called it “the memory hole,” a process that included airbrushing people out of historical photos. In the example shown below, photo retouchers made it seem as if Nikolai Yezhov, the chief of the NKVD, never existed.

Technology has made such image manipulation dramatically faster and more dangerous. Like, remember when Stalin received an honorary Oscar in 1946? I’ve got a photo right here. [Did you wonder, even for a split second, if maybe Stalin did win an Oscar? See how this works?]

This is what Hamas et al are doing with Israel, ginning up images of fake “Palestinian” suffering and thousands of AI-generated articles about alleged atrocities committed by Jews. These images, videos, memes, and articles then get repeated by a huge, well-funded social media program—which, in turn, populates AI training models, leading to further distortions of reality.

How AI works, and why that’s a problem

It’s worth taking a second to explain how AI works. Not everyone understands how AI-generated fake images and false stories on the internet turn into lies about Israel appearing at the top of AI search results or in AI reports that affect Jewish safety. Generative AI, such as ChatGPT and Google Gemini, works by ingesting enormous amounts of data, including images, videos, social media posts, and online articles, and running them through “digital reasoning” algorithms to arrive at its version of the truth.

However, AI doesn’t know what the truth is. AI doesn’t even know what the word “truth” even means. It only understands math, determining reality based on complex equations. The math is easy to trick if you know what you’re doing. The more false information you can feed into the AI training model, the more false conclusions it will draw. It’s the ultimate “garbage in-garbage out” situation.

This is why it matters that the New York Times published 6,656 articles about the Gaza War, mostly anti-Israel, in the nine months following the 10/7 attack. (For comparison purposes, the Times published 5,434 articles about Syria’s civil war in the conflict’s first 13 years.) The explosion of anti-Israel content in the major news media is injecting propaganda into the AI training data set. The same phenomenon is at work when Iran’s proxies deliberately rewrite Israeli history on Wikipedia, another key source of data inputs for AI training.

The problem is that history is being rendered fungible. Spread enough lies on the internet, and it pushes AI to rewrite history. History is not fungible, however. You can post a trillion data points on the internet claiming that Rome never occupied Judea, but the Arch of Titus still exists, even if no one knows about it.

What Israel can do

Israel is experiencing negative consequences from AI-generated disinformation, and its likely to get a lot worse if the country does not take adequate action. I’m sure Israel is on the job here, but I’d like to politely suggest that they can do a lot more. After all, Israel is the world’s third leading AI powerhouse, after the US and China.

Israel has the talent and technology to strike back with AI weapons of their own. What can they do? For starters, they can create content that advances the truth and retrains AI models to understand what’s really happening. One positive aspect of the AI universe is that it values recency, so new information gets algorithmic priority over old stories. Acting now can make a difference.

Israel might also consider spreading some lies of its own. Like, wouldn’t it be fun to flood the internet with false stories about Gaza’s notorious LGBTQ party scene? Let’s get this project going!

The background of why this happens, and why that’s a problem

The bigger picture is important to understand, too, as it is affecting more than just Israel. Today’s AI software is based on a terribly naive assumption, which is that a preponderance of data will always reveal the truth. And, that data should be free for universal access. These beliefs, which are treated as articles of faith in influential technology circles, arose from the computing community’s founding generation, which held a flawed, utopian view that people with good intentions would earnestly cooperate in using computers for the betterment of mankind.

This belief system manifests in “wikinomics,” which holds that crowd-sourced information is always the most accurate. The wikinomics paradigm, which is a core design principle guiding many of today’s most influential information systems, didn’t anticipate that malicious actors would mass-produce lies to achieve nefarious ends.

The world is paying a price for this naivete now, on many fronts. The future looks bleak, with the real possibility that AI models will continue to train themselves on an increasingly large volume of AI-generated disinformation, resulting in a data conflagration that will effectively erase anything resembling the truth from the world’s information systems.

That’s a story for another day, though. In the meantime, Israel had better get busy before it’s too late.

Photo by cottonbro studio: https://www.pexels.com/photo/a-woman-looking-afar-5473955/