No Intelligence Required: Her, Ex Machina and Interstellar | Hacker News Reader

No Intelligence Required: Her, Ex Machina and Interstellar

gregegan.customer.netspace.net.au

13 pointsggreer6 comments

I agree on "Interstellar" and "Ex Machina", but I think Greg missed the point of "Her". True, it doesn't nearly do justice to the themes of artificial intelligence and human-machine interaction. But that's because it's not meant to--it's really just a romantic little film about human relationships and the difficulty of achieving intimacy and trust in the modern world. It never even attempts to explore, let alone explain, any differences between "Samantha" and actual humans--as portrayed, she's simply a fully human female with the unusual quirk of happening to live in cyberspace. And any sci-fi-heavy exposition regarding her technological nature would only have detracted from the film's very touching, and very human, charm.

Wow, someone doesn't like movies very much. "But the dismaying fact remains that the majority of art-house cinema treats science fiction as a means to acquire a veneer of philosophical gravitas, while freeing the auteur from the burden of writing a story that makes the slightest bit of sense."

Science Fiction aside, you could say much the same thing about "Groundhog Day," "It's a Wonderful Life," and "Pan's Labyrinth." Any movie with a bit of "magical thinking" and "what if..." That's the whole point of huge swaths of movies from every category of cinema! Even romance and drama -- with seemingly solid footing in reality draws on the implausible to make things interesting. You really think a successful Richard Gere would fall head over heels in love with a prostitute (even one who looks like Julia Roberts)?

If you want "science" why are you looking in "science fiction"?

Maybe because Groundhog Day is just totally crazy and Her ... just a bit. The smaller the mistakes, the more they seem like laziness.

In Groundhog Day it seems like a constructed thing and the main plot that the protagonist lives one day again and again.

In Her it seems like a side-note, that a company is selling their super AI for way to much money to do things that don't need an AI in the first place.

If you want to use such things in your design/composition you have to embrace them and don't do them half-assed.

Ex-Machina had much better premises, but the actors were really bad and the story didn't play with these premises very well.

Anything from Greg Egan gets my attnetion!

With regard to Her — the only movie on this list that I've seen — I believe the author has approached the film from the wrong direction. There's an assumption often made amongst geeks and hackers that we'll know when we create the first "true" AI. I thoroughly enjoyed Her, because the movie didn't delve in to the "who and how" details of the AI. If we strip away the assumption that we'll know when we create the first true AI, then the film's circumstances make much more sense.

Think about how applications like Siri work today. They have a very superficial appearance of AI, but they're not terribly convincing. Apple is making progress though. Siri surprises me all the time; in a good way. Periodically, I'll try something new; and occasionally, Siri delivers. It's progressive enhancement.

Fast-forward 25-50 years. Where is the line between what we have today and "true" AI? I keep quoting the word true because I believe the phrase "true AI" is apocryphal in the way the author uses it. True AI can mean many things. Two possibilities are:

1) We gain some deeper insight in to the mechanics of human consciousness, and set out to replicate that in software. When we achieve that, we've achieved "true AI".

2) We never really understand human consciousness, but we keep chasing the goal of convincing AI. That is, AI that seems human. In this chase, we achieve a convincing enough level of AI that the questions raised in the article — and in the movie — become relevant.

I believe the latter is a more likely scenario than the former.

Looking forward 25-50 years again, let's examine a future where Megacorp™ keeps iterating their AI product. Samantha could be just another major (vis a vis SemVer) release. In Megacorp's testing, the AI was very convincing, and they believed they had a leg up on the competition. They hadn't realized just how effective their new software was though. In testing, the software's access was limited to the sandbox of the testing environment, which used samples of information. Once released on to the Internet, with access to a vast number of connections and information, the software reaches levels of convincingness (a proxy for intelligence) that the authors hadn't anticipated.

    ...but that doesn’t change the fact that we’ve spent all our time
    mired in the world view of a neurotic middle class man whose only
    real problems, and only real interest, are his dysfunctional
    romantic relationships. Jonze has taken what would in truth be a
    revolutionary event in human history and shrunk it down to an
    episode of Dr. Phil.

I hate to be so cynical, but have a look at the world around us. Wouldn't it be so perfectly human of us (as a society on the whole) to overlook the emergence of true AI while we ponder our navel? That is exactly the type of complication that Samantha is referring to. We have the capacity for rationality, but we're not 100% rational. That's what makes us human. That is, IMO, a major theme of Her.

PS - I am 100% certain that somewhere in the societal background, there is a Hacker News that exists in the world of Her. This community identified that this new software was a "game changer" early on, and many arguments ensued with regard to how this new AI will impact the world. Society at large, failed to notice.

Excuse me, but your dichotomy of two possibilities is facepalm-worthy. A few broad points:

A) Consciousness, as in subjective experience, has very little to do with "artificial intelligence", in our current understanding. Only philosophers of mind continue to be obsessed with consciousness qua consciousness, as if a program needs qualia to drive an automated car or solve a math problem.

B) We will damn well understand human consciousness and cognition, in terms of neurally-implemented statistical learning mechanisms. This will just require that one or two more generations of cognitive scientists and philosophers of mind who refuse to believe in statistical learning retire. The best current literature in neuroscience and cognitive science is all computational and statistical in nature.

C) We will not build "AI products" based on a One True Theory of Cognition. We will find that human cognition works as a singular, highly effective example of general, broad principles. This is why we already have lots and lots of highly functional statistical learning and inference systems, albeit not ones that can be directed at a keystroke to perform any inference a human being can perform. Of course, this is also what makes them useful: by restricting their hypothesis space, and thus the problems they can solve, we vastly reduce their need for training data compared to a generally-intelligent human, and thus make otherwise difficult inference problems tractable.

D) Those who understand the principles of cognition will notice and understand perfectly well when near-human or human-level "artificial intelligence" (inference engines) is/are built.

>Wouldn't it be so perfectly human of us (as a society on the whole) to overlook the emergence of true AI while we ponder our navel?

You're already doing it.

>We have the capacity for rationality, but we're not 100% rational. That's what makes us human.

I demand a definition of humanity that doesn't define my entire species as crippled, and necessarily exclude all non-crippled life forms from humanity ;-).

Reply on news.ycombinator.com