What is the Uncanny Valley of AIs and what will happen when we cross it

“Are we there, are we there, are we there?” but also “how much longer, how much longer, how much longer?”: when we talk about Uncanny Valley and artificial intelligence, we are all a bit like children in a car on a road trip. We want to arrive, we don’t know how much is left until the arrival and nobody tells us.

The Uncanny Valley Theory was described in 1970 by the Japanese Masahiro Mori, engineer and robotics scholar, but it is becoming very topical again now, over fifty years later, when we are in fact about to cross it, this metaphorical valley. Or maybe even we’ve already gone through it.

What is the Uncanny Valley

The Uncanny Valley concept it has nothing to do with the famous Turing Test (or with the Voight-Kampff, for those who have seen Blade Runner) but to really understand it, it is important to understand what the English word uncanny means: in Italian it is often translated as incredible (a known case is that of Uncanny X-Men comic, which we became The Incredibles X-Men), but it is an inaccurate translation. More properly, uncanny means puzzling, destabilizing, inconceivable, beyond human understanding and therefore incomprehensible. And for this also scary.

A widespread and acceptable translation it is perturbing: the Uncanny Valley is the Uncanny Valley and it is that area of ​​contact between us and robots that are increasingly similar to us. Mori’s hypothesis is that as these machines become more human, in appearance and in interaction capabilities, our emotional reaction becomes more empathic and positive until it turns into repulsion. It happens (the theory is) because robots are so perfect that they look human but are not yet fully human and therefore there is some detail, in their way of acting, speaking, moving their eyes (for example) that generates us some nuisance: it is like talking to a person who doesn’t behave the way a person would. This is baffling to us, and this is the Uncanny Valley. We’re almost there, but not quite there yet.

Again according to Mori, the next phase is that of overcoming the Uncanny Valley: the appearance and capabilities of robots continue to improve, they become less and less distinguishable from those of a human, the imperfections disappear and our emotional reaction becomes positive again. It’s like talking to a person who actually behaves as a person would behave. And so we are comfortable again.

Examples of Uncanny Valley in pop culture

An easy way to understand what are we talking about when we talk about the Uncanny Valley (and how it affects us) is to watch some recent film made with extensive use of computer graphics and special effects.

The most common example is that of Final Fantasy: The Spirits Withinwhich was talked about a lot in 2001: according to many newspapers of the time, viewers would have felt negative sensations in front of the faces of the protagonists, almost real but not quite real, and many film critics spoke openly of “faces that look totally fake” in their desire to seem real and of “coldness in their looks and mechanical movements”.

There are other examples: movies Polar Express, The legend of Beowulf, A Christmas Carol And Milo on Mars, all produced by Robert Zemeckis between 2004 and 2011, have been described as gruesome, full of zombies, scary, creepy and repulsive. Always for characters who were perfect enough to look human but not good enough to actually be. Again: in 2020, the release of the film Sonic came postponed for 3 months, paradoxically to make the protagonist’s appearance less humanafter the negative reactions of viewers to the first trailers, and last year also the character of She-Hulkin the Disney series of the same name, has received a lot of criticism for being almost but not quite there.

youtube: Final Fantasy: The Spirits Within trailer

Space

So NASA and SpaceX could save Hubble. And open up to the on-orbit maintenance business

by Matteo Marino

09 January 2023

Pass the Uncanny Valley (and get to the Strong AI)

Something that is already completely there, and maybe even beyond, is ChatGPT: second Professor Hiroshi Ishiguro, whom we were able to meet during his visit to the IIT in Genoa, the popular chatbot “is perhaps already beyond the Uncanny Valley, because it gives convincing answers and is able to hold a conversation that resembles the human one and faithfully reproduces it”. She looks like one of us and she does without scaring usto simplify.

But it depends on the context: during the tests we conducted on Italian Tech at the beginning of last December right on ChatGPT, some responses in which we were invited to prepare ourselves “for the impact of AI on the labor market”, with the potential loss of a lot of employment, had given us some shivers. In the same way, some concern is aroused by the experiments of researchers at Lancaster University and the University of California, who demonstrated that by now people would believe deepfake faces more real than real ones. Faces non-existent and created by an AI, but so well that they seem real and not disturbing.

In short, to answer the initial questions of imaginary children, we are actually almost there and indeed we are so close that in some cases it seems to have really already arrived. Perhaps it is also necessary to really conclude the journey that the so-called Strong AI be developed (things?), that is, a true intelligence that knows how to reason and solve problems independently. Paraphrasing Ishiguro, we need machines to be like us not only on the outside but also on the inside.

@capoema