Ask HN: Can someone please say where AI is real?
13 comments
I've been working in machine learning for a relatively long time (~15 years), and most of what I've seen does not pass my AI smell test: "do the results amaze me, even though I built it and know how it works?" I am confident true AI, by any reasonable definition, will pass this test, so I think it's a valid question to ask (to be clear I love machine learning, I just think its tight relationship with the term "AI" is an historical error).
Deep neural networks on visual and audio data are passing my smell test. At my last job I helped design and build neural networks to process music data. The results were shockingly good: the networks could tell us a very wide variety of information about a song at an accuracy that exceeded human experts, including genre and mood. This is in a domain that most people think of as very human ("understanding" and "feeling" music), which is what made the results so impressive.
The goalposts will continue to move, but an inflection point has been hit in the last few years. I believe we'll see at least one or two more inflection points in the next several decades as we move closer and closer to "strong" conscious AIs that are functionally above human intelligence.
Deep neural networks on visual and audio data are passing my smell test. At my last job I helped design and build neural networks to process music data. The results were shockingly good: the networks could tell us a very wide variety of information about a song at an accuracy that exceeded human experts, including genre and mood. This is in a domain that most people think of as very human ("understanding" and "feeling" music), which is what made the results so impressive.
The goalposts will continue to move, but an inflection point has been hit in the last few years. I believe we'll see at least one or two more inflection points in the next several decades as we move closer and closer to "strong" conscious AIs that are functionally above human intelligence.
AI is not a magic wand, just a tool to perform some tasks in a quick and hopefully more efficient manner than humans. What is it going to be then in the real world? As with any tool, it will depend on the humans using it: medicine and science might be good, military and finance might end in tears.
Maybe it already exists but you don't know what it looks like........and maybe they like it like that :)
Self-driving vehicles. IBM Watson, supposedly.
Also AI always means "a level of intelligence we haven't achieved yet." Siri could've been considered AI in the past.
Also AI always means "a level of intelligence we haven't achieved yet." Siri could've been considered AI in the past.
Picture tagging in Facebook based on computer vision. AI is being implemented in medical diagnostics. Deepmind has demonstrated reduction in Data centre power consumption.
Additionally, I'd love to see where it is being used where:
1. It's not for an adtech or financial application.
2. It's not for a company that's primarily adtech or finance.
Though clearly AI has value, it seems in practice much of it is implemented for things of questionable human value.
Some examples:
1. Search (some argue the heavy personalization has made it actually less accurate, but that's subjective).
2. Computer vision (main monetization seems to be computer vision/filters. Self driving cars will be useful if and when that comes to fruition)
3. Handwriting recognition. I think this is a great example of AI being helpful. I'm not sure how much of this is "AI" vs. just statistics and machine learning. Maybe it's all the same, in this context.
1. It's not for an adtech or financial application.
2. It's not for a company that's primarily adtech or finance.
Though clearly AI has value, it seems in practice much of it is implemented for things of questionable human value.
Some examples:
1. Search (some argue the heavy personalization has made it actually less accurate, but that's subjective).
2. Computer vision (main monetization seems to be computer vision/filters. Self driving cars will be useful if and when that comes to fruition)
3. Handwriting recognition. I think this is a great example of AI being helpful. I'm not sure how much of this is "AI" vs. just statistics and machine learning. Maybe it's all the same, in this context.
It's a funny thing that handwriting recognition is seen as worthwhile. Who writes by hand any more, and who prefers handwriting to a keyboard?
You don't have to actually write for it to be helpful. There are billions of handwritten documents out there already. OCR and handwriting recognition can help with archiving.
There is no real answer to this question, as the goal-posts for "what is AI" are always moved as soon as AI can do something. And the distinction between AI and "just a computer program" is kinda arbitrary. All AI is "just a computer program"if it's running on a computer.
Edit:
To expand on this a little more... I think it's important to remember that "AI" isn't a "thing" so much as a "field" or a family of (possibly somewhat related) techniques. That is, AI includes specific algorithms and approaches ranging from automated reasoners and inference engines, to pattern recognition using deep learning, to evolutionary algorithms, planning algorithms, etc., etc. What there isn't yet is one single artificial general intelligence that we can talk about as the be-all end-all of AI.
So yeah, individual organizations are using AI techniques all over the place. But nobody is using "AI" because it doesn't exist yet.
Edit:
To expand on this a little more... I think it's important to remember that "AI" isn't a "thing" so much as a "field" or a family of (possibly somewhat related) techniques. That is, AI includes specific algorithms and approaches ranging from automated reasoners and inference engines, to pattern recognition using deep learning, to evolutionary algorithms, planning algorithms, etc., etc. What there isn't yet is one single artificial general intelligence that we can talk about as the be-all end-all of AI.
So yeah, individual organizations are using AI techniques all over the place. But nobody is using "AI" because it doesn't exist yet.
In 1999 Amazon created a program that would look at your purchase history and recommend you new books. That's a great example of AI in the real world.
In 2016, investors decided to make AI a new priority (it kinda was there for ever). So all the tech companies use this word today to stay "relevant".
In 2016, investors decided to make AI a new priority (it kinda was there for ever). So all the tech companies use this word today to stay "relevant".
Amazon Go.
AI can fail, just like programs.
AI can bring with it a lot of issues and it can be annoying sometimes. Here is an article about where it can go wrong. https://www.surfly.com/blog/ai-in-customer-support/?utm_sour...
If you have an example, please explain why it is AI and not just a computer program.
Sounds like smoke and mirrors to me.