Wednesday, April 24, 2024
HometechThe new normal of AI fakes

The new normal of AI fakes

Following on from my last article, consider the following scenario. You've grabbed enough clear speech of someone to make a good AI model of them. You write up some text, pass it through the model and verify that the result sounds exactly the same as that individual. This is a little different from the previous example because it's a text to speech model, but essentially the same as using one voice to change to another. You now take a speech or interview from that person, change one word that will essentially change the context, and process this.

– The vast majority of people don’t remember every word they said from a given time. One typical response when presented by an inaccurate version, is to defend what they don’t remember saying, essentially “confirming” the altered or misquoted speech. Even if they deny it, the sample sounds so much like them that it becomes difficult to defend. This kind of scenario is what we are likely to see many times in 2024 against high-profile people all around the world.

– If you are a high-profile person, how do you defend against this? You could of course record everything you say but when hours of material build up, finding the exact section that was altered becomes difficult. One answer is to develop an AI that specialises in rapidly finding specific text in videos. Given the current AI marketplace this probably already exists. Like just about anything else in the technology world, this is the very real side of AI that can be misused. In the social media age, by the time there is any chance to try and correct the false story the world has moved on, those responsible are on another project, and recovering what once was is all but impossible.

– Also, as a follow-up from last time, it turned out that three well-known people were attacked at the same time, and for the example I gave, and a major country’s government was sending letters to social media giants to ask them to effectively ban one individual, without any evidence having been presented at the time other than in a news story.

– Wikipedia is becoming less and less reliable. In one recent example they edited the entry for a Canadian Nazi, a real one from World War II that had embarrassed the Canadian government, to hide his background. From now on I recommend that you consider entries in Wikipedia as being fiction rather than fact, unless there is corresponding information elsewhere. If you must, also use the Way Back Machine to see what the entries were one, five and 10 years ago to determine what has potentially been edited out or modified to match current politics or ideas.

– YouTube is being used by companies to selectively issue warnings to content creators that criticise them. Imagine the fictional “gotcha” game Paid Adow Edges. A gotcha game is one where you need to pay to stay competitive at the higher levels of the game. The people that pay the most, up to thousands of dollars per month, are known as whales just like in a casino. One such player, who probably paid enough to put one of the creator’s children through college over the course of four years, announced leaving the game because in spite of all the money had not managed to get a few of the top-level heroes and the game’s prices were the highest of any. The game’s creator, without notice to the player, sent a note to YouTube to hit him with a copyright strike.

– If you get three of these you are banned for life from YouTube. They are all but impossible and very expensive to fight. Other content creators for the game could not see any such thing in his last video and pointed out that others were offering far worse actual breaches that were not being challenged. So, if you say something someone doesn’t like they can potentially attack you and effectively remove your livelihood. The collective community opinion in this case was that the creator should have been given a quiet note to just take the post down instead of the strike. As people start to move onto platforms to make a living, YouTube currently stands out as the one to avoid and Rumble stands out as the one to trust.

– Lastly for this week of others potentially messing with our online experience, some governments around the world are working on “you can’t say that” bills. They are carefully couched in terms like “we are just looking out for your safety”, but they are about control. Canada for example has one that requires all content creators to register with them so they can monitor their content. Based on the last two years, if your content doesn’t conform with that current government’s platform it could mean a loss of access to your bank accounts or worse. As more countries apply such restrictions and controls, the concept of an open internet grows less by the day. A good VPN and independent encryption service may be the only answer in future.

James Hein is an IT professional with over 30 years’ standing. You can contact him at

- Advertisment -

Most Popular