Here are the results of our weekend poll:
As you can see, my audience is full of farsighted individuals, not neo-Luddites who turn their eyes from the future. Hooray for us!
36 thoughts on “Poll results: Artificial intelligence is coming your way”
Comments are closed.
Porque no los dos?
I believe that Kevin see these two options as be polar opposites and, at least for purposes of his survey, mutually exclusive. The second option seems to envision reasonably functional AI, albeit AI that tends to break things, break laws, and kill people. Essentially, variations on TESLA’s autopilot system wouldn’t be a scam even if they were disastrous because they’d be omnipresent like social media.
I missed the party but count me among the skeptics.
As usual KD does not define AI in any way that makes a conversation about it meaningful. Is AI your cell phone's voice to text feature? Is it HAL? Is it the singularity? What are we talking about here? By one definition AI is already here. By the other it is probably never going to happen.
Again I also think the term "simulated intelligence" would be useful in conversations like this. A bot you can sort of talk to on the phone is a simulated intelligence. There's no actual intelligence, just some canned responses to a few general categories of word combinations that someone might say.
And as to the idea that AI skeptics are like people in the 1900s saying cars are useless toys--this analogy might work if by 1900 people had been talking about how cars were going to transform our lives since before the Civil War.
Bingo. I'd write the same, but @realrobmac basically nailed it.
AI right now is pattern matching + search. Really good pattern matching, programmed from feeding it piles of examples, but still pattern matching. And in a hostile environment, it is *fragile*. You can never quite tell what it will do with a slightly weird input.
Calling a green shaded cat an avocado is cute. Failing to recognize a tank because it is covered with glitter might not be as cute.
As a longtime reader I have a different impression. I seem to remember many posts in which KD made a big distinction between AI which simply makes current “apps” more efficient (voice recognition, say) and “true” AI in which the software can pretty much do the job of a highly trained human (like a doctor or lawyer or accountant, say, or a truck driver). As I recall — sorry I’m a little too lazy at the moment to do a google search and find the links, but you certainly can if you’re interested — KD’s opinion is that AI number 1 will not disrupt society much, but that AI number 2 will be hugely disruptive.
Given even current gen Tesla motorcoaches are programmed for rolling stops, I think automatonic tractor trailers will be a death race.
By any definition, AI is certain to happen. Nature provides an existence proof that Intelligence is possible. Now it's just a matter of reverse engineering.
I think that's without a doubt true. The question is timeline. I've never been a skeptic with respect to the eventual arrival of truly robust (essentially human quality, with vastly faster processing speed) AI.
I'm just skeptical of the pace of its arrival.
I reckon (to use one example) true AI driverless automobiles are another 30-40 years away. Minimum.
The next couple of decades is a long couple of decades.
But always just around the corner.
The rise of AI is measured in Friedman Unit Factorial.
Not at all. We have a pretty good idea of how much hardware we need to have compute capacity comparable to a human mind. And we've long had a good idea as to how much compute capacity we need. Now that we're starting to get vaguely in the right ballpark, we're starting to see some interesting results.
I work in infrastructure for internet systems, have my whole career. There are a lot of products trying to get me to add them to my budget that used to be "powered by" "advanced heuristics", or "proprietary analysis engines" or whatever that magically became AI sometime in the last decade.
Then there are our fraud detection systems. Those do employ gradient descent ML and other trappings of "ai". But they're mostly black boxes that appear to do what we want, until they don't. Then we throw them away and train a new one on some other guess that might work.
None of that is general AI, something that has a theory of mind. I don't know when or if that is coming. I tend to think corporations are our AI and that's what we get, but not 100% invested in that view.
A final thought, this stuff is always disappointing in the short term and surprising in the long. So adjust your priors appropriately.
"advanced heuristics", or "proprietary analysis engines" or whatever that magically became AI sometime in the last decade."
Well, I hate to differ, but a woman working for me - with my help - wrote a book on AI - by that name - about 40 years ago for DEC. The name of the technologist who invented the term and theorized about what it would entail escapes my addled, elderly mind. But the "magic" as it were happened a long time ago. Whether the promise of AI delivers remains to be seen.
Oh, AI as a discipline existed 40 years ago. But the fact is that virtually none of it worked worth s**t back then.
What's changed is piles of online data for training, and neural networks that are orders of magnitude bigger. Like, when I was at CMU in the mid 1980s, a fast machine ran at 1 million instructions per second, and cost ballpark $1M. It wouldn't fit in any room in your house. My phone, on the other hand, runs about 1000 times faster, has probably 10,000 times more space, and is about 2000 times cheaper. This is pretty much why neural networks have moved from toys to things that can do useful pattern recognition.
Nearly all AI applications that work surprisingly well are these much bigger neural networks. They're surprisingly good at repetitive work, like speech to text. They are in no sense a general AI.
I didn't click the poll because my choice wasn't there. I think AI is way overhyped but I don't believe it qualifies as a scam. I think it's coming sometime in the next few decades if human civilization doesn't first succumb to resource wars because of global warming. But not like a freight train; more like gradual at first and then geometrically faster.
"I think AI is way overhyped but I don't believe it qualifies as a scam."
I agree. Although I don't think it's coming. It's here. It just has a long way to evolve before it becomes something like futurists imagine.
We're almost getting to the basic "Star Trek" computer--voice commands for light, music, random facts, etc. Granted, we kinda had some of that with "The Clapper".
Yes that would have been my choice too but I settled for clicking on the "It's coming like a freight train" choice as I've been a lot of automation eliminate low level accounting positions in my career. Luckily I've been able to pivot from low level accounting clerk right out of college to supporting the accounting system - discovering bugs, documenting the system and recommending system change requests.
When I was in college back in the early 90s, I taught myself what was then often called "neural network theory" since I was planning on becoming a neuroscientist (which is ultimately what I did), hoping for insights into how real networks of neurons work. Although I completed an undergrad thesis using them, I concluded that they are terrible models of neurons and moved on. A few years ago, hearing about the "deep learning" hype, I decided to see what had changed. The answer, as noted already by several commenters here, is fundamentally nothing. No fundamentally new approach has been devised since 1990 (or even the 1970s, depending on how you count it). So you could, on that basis, call it hype fueled by improvements in hardware and data availability. However, the details have changed too--progress been made has been made in devising network architectures that learn and perform more efficiently, even if they are still just tuning parameters by gradient descent of a cost function aided by error backpropagation. These innovations have generally made them even more un-brainlike in their technical details, but it's not actually clear that you need different fundamental algorithms for "real intelligence." There's good evidence that brains use some versions of reinforcement learning algorithms to use experience to train themselves--albeit with differences in how the units and learning rules work--coupled with species- and circuit-specific architectures that bias learning in the "right" direction. Tweaking architectures to get good results has been the work of evolution for more than half a billion years. That tweaking may be the hard part of AI--not (necessarily) finding brand-new principles. Which means we are already walking the path to AI and have been doing so for some time. I'm not saying I'm certain that's true, but that's why I picked "AI is coming like a freight train."
Thanks for the comparison and details! Truly, given how the human species in the United States seems largely incapable of weighing information or learning, coming up with an actual analog to our brain "function" is probably a dumb idea.
We'd do better with something more effective.
What is the circumstance when pattern recognition becomes so complex you cannot tell the difference between a computer's mimicry of something like Bach or Shakespeare and the real thing?
Is that real AI or is that still pattern recognition?
Yeah, my choice wasn’t there so I skipped it. I’m curious as to the number of respondents…
Ditto. My response is (and always has been) something to the effect of: of course kiick ass AI will arrive. Some day. It's not a scam, because human intelligence is not a scam (but a product of 3.5 billion years of life's evolution on earth).
But timeline may well dissapoint.
AI is a bunch of different stuff sloppily bundled under the same label. Autonomous driving systems are vastly different than natural language processing, and so on. Mostly what people are still talking about is automation.
I'
I'm still waiting for a voice recognition system that really works.
Here (according to this site) is a paragraph generated by AI:
The gazelle bounded through the tall grass, its long, graceful legs moving swiftly and gracefully. It was a beautiful creature, and it seemed to move with a kind of elegance and speed that was almost supernatural. The gazelle's movements caught the attention of a nearby podcast listener, who turned to see what it was. They were both startled when they saw the gazelle charging straight at them, its eyes flashing with an angry intensity. The podcast listener didn't have time to react, and was quickly flattened under the gazelle's powerful body. The gazelle continued to run, its wild movements quickly leaving the scene.
That's based on two theme entries: gazelle and podcast.
If the promise of AI is a slightly more advanced version of Mad Libs, then AI is here.
I have no idea what "they ... both" is referring to, but I suspect the podcast listener deserved everything he/she/they had coming to him/her/them.
I still like my books written by humans and expect they'll continue to be the best books to read long after we're all gone.
Not even bad-good enough for a Dishonorable Mention in the Bulwer-Lytton Fiction Contest.
AI, take 2.
It was a dark and stormy night. The only light came from the fire in Snoopy's doghouse, which cast an eerie glow in the darkness. Snoopy sat in his chair, watching the fire. He was thinking about his friend Woodstock, who had gone away to war. Snoopy was sad, but he knew that Woodstock would be safe. Suddenly, a loud noise outside made Snoopy jump. It was a lightning bolt, and it was hitting the doghouse! Snoopy trembled, and he knew that a storm was coming.
Theme entries:
...a dark and stormy night
...Snoopy
Solid fourth-grade-level writing, but not particularly inspired, in a good or bad way.
Was waiting for someone to declare artificial mimicry as a reason for relegating the Turing Test to the trash bin, followed by retort that such mimicry is biologically parallel to how real learning occurs, and therefore, the first step towards true AI, resulting in an ad hominem attack that only an idiot would think that artificial mimicry of such an idiot could be considered a sign of intelligence, nascent or otherwise. ¯\_(ツ)_/¯
The problem here is that AI doesn't work the way Kevin is suggesting. As soon as we started recording taxes paid on clay tablets, we started working with artificial intelligence. A human intelligence didn't put a man on the moon; a huge collective human-created intelligence did. Intel has long been designing and building new computer chips using a collective combination of human and artificial intelligences.
We communicate with each other a little bit faster; we automate tasks we used to do by hand; and step by step we integrate more artificial intelligence into our already very artificial collective intelligence.
Criminals, spies, and military will use AI too. Yippee.
Marc Andreeson, Vox, 10.5.2016: '. . . recent breakthroughs mean artificial intelligence has the potential to spawn a new generation of big, important technology companies.'
Eric Schmidt, Tim Ferriss interview, 10.25.2021: I don't have a specific quote to trot out; the general sense of the interview is that Schmidt (and Henry Kissinger) think AI is huge.
By the way, people here have suggested all kinds of definitions of AI. Best to rely on Eric Schmidt, who said: 'The simplest explanation for AI is a system that gets better through learning, that it’s busy learning something. That’s probably the easiest and current definition of it.'
Peter Diamandis, 5.14.2015 blog, 'Artificial Intelligence (AI) is the most important technology we’re developing this decade.'
I don't invest better in technology than these guys. That's why I voted the way I did.
Still wondering how many respondents there were…
Hey, I know this post is old (by Internet standards), however, a tweet just posted today by Peter Diamondis popped up. I don't know if the comment system allows links to tweets, so I will simply quote it.
"In the late 1800s, if you wanted a good idea for a new business, all you needed was to take an existing tool, say a drill or a washboard, and add electricity to it—creating a power tool. In the 2020s, AI will be the electricity. Take any existing tool, and add a layer of AI." (3:58 PM · Feb 11, 2022)