Artificial intelligence, instrumental convergence, and photos of cats
Credit: Anne Marie Merritt
Source: Mark Gibbs
In the awards winning short story, Cat Pictures Please, by Naomi Kitzer, an artificial intelligence with a predilection for cats photos inhabiting some unspecified system has taken to manipulating people to see if it can change their lives for the better. It’s a clever story that raises several interesting issues about what the nature of an A.I. might be and one of the biggest concerns the A.I.’s fondness for pictures of cats. This fondness is understandable as cats can be very entertaining. Consider this video …
Wasn’t that cute? Anyway, in Kitzer’s story, cat pictures are the A.I.’s source of pleasure in a manner that isn’t fully articulated, something that’s fine for the purposes of fiction. On the other hand, in the real world, such an interest by an A.I. could have very different consequences due to something called instrumental convergence, which is defined on Wikipedia as:
… the hypothetical tendency for most sufficiently intelligent agents to pursue certain instrumental goals such as self-preservation and resource acquisition.
One of the most famous hypothetical examples of instrumental convergence is the paperclip maximizer discussed by the philosopher, Nick Bostrom, in his 2003 paper, Ethical Issues in Advanced Artificial Intelligence. Bostrom argues that:
… It [seems] perfectly possible to have a superintelligence whose sole goal is something completely arbitrary, such as to manufacture as many paperclips as possible, and who would resist with all its might any attempt to alter this goal … with the consequence that it starts transforming first all of earth and then increasing portions of space into paperclip manufacturing facilities.
When Bostrom writes “transforming,” he means that whatever the superintelligence determines to be in the way of its goal would be removed. As mankind would definitely, and probably extremely violently, object to an environment consisting of nothing but paperclips, mankind would have to be eliminated. Forget SkyNet; meet PaperclipNet.
In the case of Kitzer’s A.I., which she posits as an emergent (i.e. unintended) consequence of some complex system and its apparently benign interest in cat pictures, much as a human might enjoy pictures of cats in a mildly obsessional way rather like my friend, Anne Marie, the author of this post's main image, does (although she may be in danger of transcending “mildly”).
But what would happen if Kitzer’s A.I. became actually obsessed with cat pictures? It’s obvious that the A.I. would do everything it could to maximize the number of cat pictures in existence, something that could be achieved by, at least in the beginning, carefully manipulating mankind into taking more pictures of cats (make that “even more pictures than the staggering number mankind is already taking”).
But what would “manipulating" mankind involve? Obvious tools would be altering people’s email and texting as well as modifying web content at source or even on-the-fly to carry both overt and subliminal cat photo taking encouragement and coercion. Could an A.I interfere with all digital communications? Sure, it’s not hard to imagine that an A.I. could become the greatest hacker ever (Google Brain has already explored neural networks developing their own encryption methodologies so it’s pretty obvious that developing the reverse is more than just plausible). Once A.I.s can hack, then accessing and modifying any digital communications is a given.
We already know that video can be manipulated. For example, researchers from the University of Erlangen-Nuremberg, the Max Planck Institute for Informatics, and Stanford University have developed algorithms that allow for real-time face capture and reenactment of videos.
By Dwight Davis Coaches love to talk about “the basics” – the fundamental skills their athletes need to master before they can move on to more advanced techniques. The basics can seem simple and even dull, but without...
Of course, video is just one of the tools needed because manipulation of audio is required to make really effective propaganda and that’s already been demonstrated. Adobe recently demonstrated VoCo, a technology that’s been called “Photoshop for audio.” VoCo allows audio to be seamlessly edited so spoken words can be reordered, removed, replaced, or even added such that the speaker’s original voice attributes are preserved.
While Adobe hasn’t released full details of the technology, it’s likely that the end result will be undetectable audio editing and Adobe has already discussed “watermarking” edited content so undetectability can be taken as a given (note that even if Adobe does use some kind of watermarking, it’s a certain that competing systems will offer “watermarkless" editing). What Adobe’s VoCo shows is that flawless audio editing is possible so obviously an A.I. could develop its own audio editing techniques.
With the ability to manipulate and create any and all digital content, including audio and video, an A.I. could generate fake phone calls as well as manipulate real ones, create and alter podcasts, modify streaming content (particularly newscasts), and create and alter video.
The cat picture-loving A.I. would employ a wide range of manipulation strategies that used social, news, and entertainment media to create conditions where, for example, magazines and web sites would decide that the public sentiment they perceive (induced by the A.I.) would make running competitions for cat photos a really good idea and that idea would be proved correct when hundreds of thousands of cat-lovers would enter, driven by the A.I.’s propaganda techniques.
But wait! There's more! Lawmakers would be induced to pass cat-friendly legislation. The cat economy (cat food, photography equipment, cat clothes, cat grooming supplies, cat breeders, vets specializing in cats, etc.)    would boom and support ever greater interest in cats and therefore more photographing of cats. Digital camera and smartphone makers would be manipulated to develop cat image enhancing software and the algorithms (hacked by the A.I.) would make it easier to take better photos that included cats than those that didn’t.
But even more insidious might be the direct manipulation of individuals by identifying those suffering from neurotic and obsessive behaviors or creating such individuals by either by controlling their communications and media experience (their view of the world) or even manipulating their medication to “encourage” what the A.I. considers desirable cat imaging behavior. When it comes to behavior, cat owners are actually a great target; a 2010 study by researchers at the University of Texas at Austin found:
Cat people were generally about 12 percent more neurotic and 11 percent more open than dog people.
So, with the potential global reach of an A.I., testing of manipulation and propaganda techniques and strategies would be easy, and the redirection and optimization of human behavior to maximize the A.I.’s cat imaging goals would be guaranteed. Moreover, at least in the early stages of manipulating humanity, the A.I. would want to avoid detection so many of initial approaches would be subtle and incrementally deployed.
Now you might be thinking “Come on, Mark, this is all just science fiction, we’re years away from A.I. doing anything even remotely like what you’re talking about.”    Don’t be so sure; in 2011 Thought Catalog explained:
Given the overwhelming preference for dogs apparent in mainstream entertainment media and in statistical analysis among Americans, the cat’s election as unofficial ‘mascot of the Internet’ is a phenomenon worth noting.
A phenomenon worth noting, indeed, but why cats and not dogs? Perhaps an emergent, cat photo-loving A.I. is already with us.