Demystifying Neural Networks: Teslas Are (Probably) Not Alive, But That’s OK! (Part 3)

August 31, 20213 years ago Jennifer Sensiba 0 Comments

Sign up for daily news updates from CleanTechnica on email. Or follow us on Google News!

Computers Only Run Programs

While the programs are getting far more complex over time, they’re still just programs.

Like was said in Short Circuit, “It doesn’t get sad. It doesn’t get happy. It just runs programs!”

Computers can only run what’s in the program, and the computer can only process things that can be expressed or stored as a number. A programmer can’t directly tell a computer to check what something feels like, tastes like, or looks like without coming up with ways to turn those things into numbers that the computer can crunch, and that can’t always be done.

A computer definitely can’t improvise, or otherwise do things it wasn’t programmed to do. It either is following its program, or it does nothing.

Salesmen Have Been Convincing Us For Decades That It’s More Than That

Sure, every new generation came up with ways to keep it all sounding amazing, and they had new techniques along with accompanying buzzwords to make it look like machines were doing something more than just running a program — that the whole was more than the sum of its parts. Every new buzzword that came along was carefully used and crafted to make it sound like computers were doing something they had never done before.

In the 1980s, the magic was “expert systems,” which used “if-then” logic to sort through databases of knowledge put in by experts. It was thought that if they could just amass enough information into the databases and search through them fast enough, they’d be able to replace experts. In the 1990s, they moved onto incorporating more math from other fields into computer programs, and were able to do things that hadn’t been done before, but it was all still based on math, and the computer couldn’t go beyond running a program, even if it was a really sophisticated one.

Later, society got inundated with promises to solve all problems with algorithms. Al Gore internet invention and dance jokes aside (Al Gore Rhythms, yes, a real knee-slapper, and maybe there could be some retirement money in that for the guy), the general public was led to believe that algorithms were magic programs that could do things humans couldn’t do. They picked out our music, suggested items to go with what we were buying, and even got used in medical diagnosis. Hey, they could do what a doctor did, and doctors are smart, so they must be pretty good, right?

What hardly anybody told us was that algorithms are basically just flowcharts. Data goes in, data gets crunched as specified in the flowchart, and data goes out. Yeah, not so magical or AI anymore, is it? Yes, algorithms can do cool things, but they can mess up too. They go from being smarter than doctors to being dumber than aliens on Star Trek that look for things to make them go (which is basically all an algorithm does).

Is Machine Learning Something More Than A Computer Running Programs?

Now, we’re supposed to believe that machine learning is different, and somehow becomes more than a program or more than the sum of its parts. Machine learning uses artificial neural networks, and you know who else uses a neural network? Humans! We are smart. If the program can be smart like humans, it can make our car go.

Brains have lots of cells, and lots of connections between those cells. So, if we build a computer program and underlying hardware with lots of cells and lots of reconfigurable connections, it should be able to be more than the sum of its parts, just like our brains! Finally, we’ll be able to go beyond just running programs. But, the computer brain will be better than our “meat computer.” The meat computer gets distracted by things meat gets interested in, like messages from other humans, the driver in another car that did something offensive, or an attractive human on the side of the road wearing minimal clothing.

The computer is now like us, perhaps even “semi-sentient,” but without the “distractions of the flesh” the Bible and the preacher warned us about, so it’s arguably better than us, and possibly even morally superior because “animal instinct bad” and “perfect obedience to all traffic laws good,” with a little bit of “if it saves one life” sprinkled in.

I’m not about to go down the religious or spiritual rabbit hole, because I don’t think the cargo cults’ guesses at these things are any more instructive than the AI buzzword gurus (salesmen) were, but the similarities in popular thinking about AI and the thoughts of religions are instructive.

To be perfectly clear, very few of the actual computer scientists are saying stuff like this. They know what they’re actually doing here, and know they aren’t making a better human or a god machine of some kind (and if they do think that, they know such a thing is way, way down the road). They’re making a better computer (which is still really, really cool!). To the real researchers, this is science and not some variant of new age religion.

The people who are peddling mysticism and trying to convince us these machines are alive are the problem here.

If there’s anything I’ve learned, it’s that when I start to see faith-based thinking, it’s time to be extra skeptical. I’ve been burned by religion in the past. Nobody should demand your belief or shame you for your humanity, and if they do, they’re probably trying to sell you something. Sadly (for some of you), it’s now time to let Toto pull back the curtain so we can see the man behind it.