'Could a robot do my job?'

Rotation Advertisements

We hope you enjoy your visit to this forum.

If you are reading this then it means you are currently browsing the forum as a guest, we don’t limit any of the content posted from guests however if you join, you will have the ability to join the discussions! We are always happy to see new faces at this forum and we would like to hear your opinion, so why not register now? It doesn’t take long and you can get posting right away.

Click here to Register!

If you are having difficulties validating your account please email us at admin@dbzf.co.uk

If you're already a member please log in to your account:

Pages:
1
2

'Could a robot do my job?'
Tweet Topic Started: Sep 14 2015, 07:06 PM (1,416 Views)
+ Steve	Sep 15 2015, 01:09 AM Post #16
Greetings. I will be your waifu this season. Posts: 19,471 Group: Legend Member #5,710 Joined: February 7, 2011 Gender Male Country United Kingdom Favourite Anime Hunter x Hunter(2011) What Browser do you use? Mozilla Firefox Favourite Manga Berserk all day erry day	Common2 Sep 15 2015, 12:52 AM @Steve oh my god I'm not even going to start this again I don't mean robots taking over the world I mean robots being used instead of soldiers. It's basically inevitable at this point, it reduces the lives lost of the rich military units that could afford to use bots. Not quite robots yet but remote controlled machines are a definite. Why send a group of soldiers in to a building at night when you could send a bullet proof RC machine with IR camera's. Fairly obvious direction war will go in we already use drones.
	Definitely not a succubus, fear not


Tinny	Sep 15 2015, 01:16 AM Post #17
Posts: 6,060 Group: Super Moderator Member #12,355 Joined: May 24, 2015 Gender Not Specified Favourite Anime Silver Spoon, Fate/Zero, Legend of Galactic Heroes, Space Patrol Luluco, Psycho-pass, Miss Kobayashi What Browser do you use? Chrome, Edge Favourite Manga Wizard's Soul ~Holy War of Love~	I assume Steve is referring to us heading more in the direction of dreaddive warfare becoming bigger and bigger. Though I doubt the biological soldier will ever go away without AI that can think feel and learn.
	Above signature created by Graffiti


Common2	Sep 15 2015, 01:23 AM Post #18
Posts: 171 Group: Member Member #11,263 Joined: April 25, 2014 Gender Male Country United States	Kyo Sep 15 2015, 01:03 AM AI is also developed by us. I don't see your point. Also machine learning algorithms are a subset of AI. Your thread didn't have any response to what I posted. My point isn't that they're developed by us, my point is that ML algorithms are just a tool. They're not intelligent enough to "do research" without any guidance (in fact ML algorithms are themselves a major object of study in AI research). By "do research", I mean do things like design algorithms, or write research papers, or do a literature survey of current work in the field, or come up with proofs, or anything else that a professor or a graduate student would do, with the goal of advancing the field of research. Being a researcher takes lots of creativity. Natural language is a major roadblock (among many others things) towards a robot capable of actually replacing researchers.



Mihawk	Sep 15 2015, 01:50 AM Post #19
Posts: 4,804 Group: Prodigy Member #4,783 Joined: March 2, 2010 Gender Not Specified	Common2 Sep 15 2015, 01:23 AM Kyo Sep 15 2015, 01:03 AM AI is also developed by us. I don't see your point. Also machine learning algorithms are a subset of AI. Your thread didn't have any response to what I posted. My point isn't that they're developed by us, my point is that ML algorithms are just a tool. They're not intelligent enough to "do research" without any guidance (in fact ML algorithms are themselves a major object of study in AI research). By "do research", I mean do things like design algorithms, or write research papers, or do a literature survey of current work in the field, or come up with proofs, or anything else that a professor or a graduate student would do, with the goal of advancing the field of research. Being a researcher takes lots of creativity. Natural language is a major roadblock (among many others things) towards a robot capable of actually replacing researchers. ...And AI is just a tool, and ML is a subset of AI. You probably mean there isn't any creative AI yet which doesn't mean there isn't AI that is currently replacing researches. Researches are being replaced through active learnings algorithms. You're seeming to imply that machine learning isn't AI, which in fact it is. It might not be that advanced but it's still AI. And there are multiple common ways to develop algorithms through AI out there. This part of your post is correct "write research papers...or anything else that a professor or a graduate student would do" (I'd just change it to most things a professor graduate student would do"). Natural language processing is a common facet to AI now. Anyone that knows what they're doing in the field can train a program analyze sentiment, develop main ideas, and perform more powerful searches through chunks of texts that humans wouldn't be able to. A lot of Google's technology is based on these features. And yes Google's search is a ridiculously intelligent AI.



Common2	Sep 15 2015, 02:52 AM Post #20
Posts: 171 Group: Member Member #11,263 Joined: April 25, 2014 Gender Male Country United States	Kyo Sep 15 2015, 01:50 AM Common2 Sep 15 2015, 01:23 AM Kyo Sep 15 2015, 01:03 AM AI is also developed by us. I don't see your point. Also machine learning algorithms are a subset of AI. Your thread didn't have any response to what I posted. My point isn't that they're developed by us, my point is that ML algorithms are just a tool. They're not intelligent enough to "do research" without any guidance (in fact ML algorithms are themselves a major object of study in AI research). By "do research", I mean do things like design algorithms, or write research papers, or do a literature survey of current work in the field, or come up with proofs, or anything else that a professor or a graduate student would do, with the goal of advancing the field of research. Being a researcher takes lots of creativity. Natural language is a major roadblock (among many others things) towards a robot capable of actually replacing researchers. ...And AI is just a tool, and ML is a subset of AI. You probably mean there isn't any creative AI yet which doesn't mean there isn't AI that is currently replacing researches. Researches are being replaced through active learnings algorithms. You're seeming to imply that machine learning isn't AI, which in fact it is. It might not be that advanced but it's still AI. And there are multiple common ways to develop algorithms through AI out there. This part of your post is correct "write research papers...or anything else that a professor or a graduate student would do" (I'd just change it to most things a professor graduate student would do"). Natural language processing is a common facet to AI now. Anyone that knows what they're doing in the field can train a program analyze sentiment, develop main ideas, and perform more powerful searches through chunks of texts that humans wouldn't be able to. A lot of Google's technology is based on these features. And yes Google's search is a ridiculously intelligent AI. Dude, you don't have to tell me ML is a subset of AI, I know this already. I'm a PhD candidate in computer science who's taken AI and ML classes so there's no need to educate me. NLP has always been a major area of AI. I've sat through the lectures of a prominent NLP researcher here and have talked to one of his PhD students, and they both told me we're far from solving NLP. The best algorithms still have difficulty parsing and interpreting common news media like the New York Times or Twitter. There are so many exceptions in natural language, and language is so volatile, which is why it's hard to formulate one static model that captures entirely any natural language. And NLP is one very necessary aspect of a hypothetical automated researcher. I am aware that sentiment analysis is a thing, but beyond "positive" and "negative" emotion I'm not sure how far we've advanced there; regardless, sentiment analysis isn't very relevant to doing research. Developing main ideas can, to some degree of success, be automated. I've heard of programs that automatically summarize text for you, though again I'm not sure how successful they are, and they're marginally important to doing research anyways. What makes research research is the process of coming up with new ideas or hypotheses or algorithms or proofs. As a researcher, these things are hard and are not amenable to naive algorithmic processes. For example, s*ty papers often times get published. Earlier this year I was working on completely rewriting and formalizing a paper written back in the 90s, and I find it unbelievable that any kind of algorithm could come up with anything close to what I wrote. The paper was filled with missing definitions, massive jumps in reasoning, false claims, and ambiguous writing. At the very least, to do my kind of work, one would need NLP and an automated theorem prover. Automated theorem proving is a very specific area of AI that has major problems of its own, though would be necessary to do automated research in most areas of science and engineering. The hard problems automated research are countless, so I'm not sure why you're so eager to claim that "researcher" will be one of the first jobs to do, when jobs like accounting still exist. If you're aware of any published, significant work on algorithms that do actual research, I would be extremely surprised but happy to hear about it, but as far as I'm concerned, the closest thing we have to automated research is SCIgen. Google's algorithms are great but they're not new and not any kind of AI breakthrough. Their latest advances come from breakthroughs in computer architecture, not AI. Neural networks have been around for decades. And based on conversations with my friend in ML/robotics, and from my understanding of deep learning and neural networks, we don't actually know what the f* we're doing. We're literally messing with the parameters of our neural networks, hoping to come up with smart algorithms. That's what Google's "dreaming" s*** is about.



SpeedoTrunks	Sep 15 2015, 07:03 AM Post #21
Posts: 908 Group: Member Member #9,238 Joined: March 22, 2013 Gender Not Specified	I work in IT support, so I presume one day they'll just build machines that work properly lol. Then I'll be out of work hah.



peep	Sep 15 2015, 09:17 AM Post #22
Posts: 1,104 Group: Member Member #9,326 Joined: April 7, 2013 Favourite Anime Cowboy Bebop, Rurouni Kenshin What Browser do you use? Chrome	SSj4 Gotenks Sep 14 2015, 11:41 PM I could see my job (accountant) being replaced by robots in the near future. Well, they could take over the bookkeeping part, but some of the main tasks of an accountant are to analyze data and make decisions regarding finances, and from what I've come across, AI still has a long way to go before being able to do that. Stuff like advances in tax software might put tax preparation firms out of business, but that's just one facet of accounting.
	thx accelerator for the sig it is my first sig ever and i love it and i love u


Notaka	Sep 15 2015, 10:35 AM Post #23
Always Wright Posts: 4,070 Group: Member Member #10,814 Joined: January 31, 2014 Country Algeria What Browser do you use? Waterfox Favourite Manga Dragon Ball, JoJo no kimyou na bouken, One piece, currently reading Shaman King	Well who wants to start a robot building business ? I sure do if they manage to replace.
	*"My name is Kakyoin Noriaki!"* Thanks to Graffiti for the Sig!


* Yu Narukami	Sep 15 2015, 10:39 AM Post #24
Izanagi! Posts: 12,330 Group: Retired Staff Member #6,446 Joined: July 3, 2011 Gender Not Specified Country None	With most jobs, you'd have to develop the AI to the point when it can look at things from different points of view. As in, it could see the option that would result in the most profit for the company, but it would realise the larger implications of that option and the impact that it could have on the economy, employees, the company's reputation, etc. I have no idea how far AI has come and whether this is feasible right now or not. Edited by Yu Narukami, Sep 15 2015, 10:40 AM.



+ Steve	Sep 15 2015, 01:22 PM Post #25
Greetings. I will be your waifu this season. Posts: 19,471 Group: Legend Member #5,710 Joined: February 7, 2011 Gender Male Country United Kingdom Favourite Anime Hunter x Hunter(2011) What Browser do you use? Mozilla Firefox Favourite Manga Berserk all day erry day	Considering how much we've learned and are still learning about our brains I wouldn't be surprised if AI took a sudden leap if we designed it around how we know brains work, we're further along with brains than AI so what better way to do it than to combine both sciences.
	Definitely not a succubus, fear not


Mihawk	Sep 15 2015, 03:25 PM Post #26
Posts: 4,804 Group: Prodigy Member #4,783 Joined: March 2, 2010 Gender Not Specified	Common2 Sep 15 2015, 02:52 AM Kyo Sep 15 2015, 01:50 AM Common2 Sep 15 2015, 01:23 AM Kyo Sep 15 2015, 01:03 AM AI is also developed by us. I don't see your point. Also machine learning algorithms are a subset of AI. Your thread didn't have any response to what I posted. My point isn't that they're developed by us, my point is that ML algorithms are just a tool. They're not intelligent enough to "do research" without any guidance (in fact ML algorithms are themselves a major object of study in AI research). By "do research", I mean do things like design algorithms, or write research papers, or do a literature survey of current work in the field, or come up with proofs, or anything else that a professor or a graduate student would do, with the goal of advancing the field of research. Being a researcher takes lots of creativity. Natural language is a major roadblock (among many others things) towards a robot capable of actually replacing researchers. ...And AI is just a tool, and ML is a subset of AI. You probably mean there isn't any creative AI yet which doesn't mean there isn't AI that is currently replacing researches. Researches are being replaced through active learnings algorithms. You're seeming to imply that machine learning isn't AI, which in fact it is. It might not be that advanced but it's still AI. And there are multiple common ways to develop algorithms through AI out there. This part of your post is correct "write research papers...or anything else that a professor or a graduate student would do" (I'd just change it to most things a professor graduate student would do"). Natural language processing is a common facet to AI now. Anyone that knows what they're doing in the field can train a program analyze sentiment, develop main ideas, and perform more powerful searches through chunks of texts that humans wouldn't be able to. A lot of Google's technology is based on these features. And yes Google's search is a ridiculously intelligent AI. Dude, you don't have to tell me ML is a subset of AI, I know this already. I'm a PhD candidate in computer science who's taken AI and ML classes so there's no need to educate me. NLP has always been a major area of AI. I've sat through the lectures of a prominent NLP researcher here and have talked to one of his PhD students, and they both told me we're far from solving NLP. The best algorithms still have difficulty parsing and interpreting common news media like the New York Times or Twitter. There are so many exceptions in natural language, and language is so volatile, which is why it's hard to formulate one static model that captures entirely any natural language. And NLP is one very necessary aspect of a hypothetical automated researcher. I am aware that sentiment analysis is a thing, but beyond "positive" and "negative" emotion I'm not sure how far we've advanced there; regardless, sentiment analysis isn't very relevant to doing research. Developing main ideas can, to some degree of success, be automated. I've heard of programs that automatically summarize text for you, though again I'm not sure how successful they are, and they're marginally important to doing research anyways. What makes research research is the process of coming up with new ideas or hypotheses or algorithms or proofs. As a researcher, these things are hard and are not amenable to naive algorithmic processes. For example, s*ty papers often times get published. Earlier this year I was working on completely rewriting and formalizing a paper written back in the 90s, and I find it unbelievable that any kind of algorithm could come up with anything close to what I wrote. The paper was filled with missing definitions, massive jumps in reasoning, false claims, and ambiguous writing. At the very least, to do my kind of work, one would need NLP and an automated theorem prover. Automated theorem proving is a very specific area of AI that has major problems of its own, though would be necessary to do automated research in most areas of science and engineering. The hard problems automated research are countless, so I'm not sure why you're so eager to claim that "researcher" will be one of the first jobs to do, when jobs like accounting still exist. If you're aware of any published, significant work on algorithms that do actual research, I would be extremely surprised but happy to hear about it, but as far as I'm concerned, the closest thing we have to automated research is SCIgen. Google's algorithms are great but they're not new and not any kind of AI breakthrough. Their latest advances come from breakthroughs in computer architecture, not AI. Neural networks have been around for decades. And based on conversations with my friend in ML/robotics, and from my understanding of deep learning and neural networks, we don't actually know what the f* we're doing. We're literally messing with the parameters of our neural networks, hoping to come up with smart algorithms. That's what Google's "dreaming" s* is about. Well your post seemed to imply machine learning isn't real AI so I thought I'd point it out. Well you don't seem to know a lot of machine learning and AI, just the abstract ideas and concepts. I've actually worked in their application with large companies (can't its name because of things like NDAs) and their AI software. Quote: NLP has always been a major area of AI. I've sat through the lectures of a prominent NLP researcher here and have talked to one of his PhD students, and they both told me we're far from solving NLP. The best algorithms still have difficulty parsing and interpreting common news media like the New York Times or Twitter. There are so many exceptions in natural language, and language is so volatile, which is why it's hard to formulate one static model that captures entirely any natural language. And NLP is one very necessary aspect of a hypothetical automated researcher. The key to using NLP across a machine is training it across a dataset you're using. If you don't train it it's more or less useless. One of the main keys of machine learning is error reduction, and if you aren't starting their you're just plugging and playing libraries. Quote: I am aware that sentiment analysis is a thing, but beyond "positive" and "negative" emotion I'm not sure how far we've advanced there; regardless, sentiment analysis isn't very relevant to doing research. Developing main ideas can, to some degree of success, be automated. I've heard of programs that automatically summarize text for you, though again I'm not sure how successful they are, and they're marginally important to doing research anyways. What makes research research is the process of coming up with new ideas or hypotheses or algorithms or proofs. As a researcher, these things are hard and are not amenable to naive algorithmic processes. Sentiment analysis, among other NLP tools are useful in carrying conversations when trained against a dataset. With the right datafeeding it should be possible to write papers based on research data in specified fields. Quote: For example, sty papers often times get published. Earlier this year I was working on completely rewriting and formalizing a paper written back in the 90s, and I find it unbelievable that any kind of algorithm could come up with anything close to what I wrote. The paper was filled with missing definitions, massive jumps in reasoning, false claims, and ambiguous writing. At the very least, to do my kind of work, one would need NLP and an automated theorem prover. Automated theorem proving is a very specific area of AI that has major problems of its own, though would be necessary to do automated research in most areas of science and engineering. True, I agree AI at this point would be very hard pressed at writing well written papers. Quote: utomated theorem proving is a very specific area of AI that has major problems of its own, though would be necessary to do automated research in most areas of science and engineering. The hard problems automated research are countless, so I'm not sure why you're so eager to claim that "researcher" will be one of the first jobs to do, when jobs like accounting still exist. Well now I know you meant a field researcher rather than say, a market researcher. You were referring to a "researcher" you were familiar with, when I was simply referring to the general term of employable researcher. For example law companies have people that do only research, and they can easily be replaced through AI technology which searches far better than them. Quote: Google's algorithms are great but they're not new and not any kind of AI breakthrough. Their latest advances come from breakthroughs in computer architecture, not AI. Neural networks have been around for decades. And based on conversations with my friend in ML/robotics, and from my understanding of deep learning and neural networks, we don't actually know what the f we're doing. We're literally messing with the parameters of our neural networks, hoping to come up with smart algorithms. That's what Google's "dreaming" s* is about. Google's search has been one of the least replicable form's of AI today. We know exactly what we're doing with neural networks, lol. I've used and made them and they're more of a simple concept than people realize (at least in abstraction)



+ Green	Sep 15 2015, 06:41 PM Post #27
Flashy Thing! Posts: 5,436 Group: Legend Member #1,181 Joined: February 25, 2006 Gender Male Country United States What Browser do you use? Fire fox Favourite Manga Gantz	All my work is in social service and emergency services. I don't foresee robots taking over any time soon or ever.
	Don't you ever worry that you're on the Devil's side without even knowing it? - DCI John Luther Black power ranger; I do not belong... B.o.B - Mr. Mister


Common2	Sep 16 2015, 02:41 AM Post #28
Posts: 171 Group: Member Member #11,263 Joined: April 25, 2014 Gender Male Country United States	@kyo sorry will reply tomorrow too busy being a grad student



Common2	Sep 19 2015, 06:35 AM Post #29
Posts: 171 Group: Member Member #11,263 Joined: April 25, 2014 Gender Male Country United States	Quote: Well your post seemed to imply machine learning isn't real AI so I thought I'd point it out. Well you don't seem to know a lot of machine learning and AI, just the abstract ideas and concepts. I've actually worked in their application with large companies (can't its name because of things like NDAs) and their AI software. I have no idea where you go that idea but for the record I know machine learning is a subset of artificial intelligence. I don't know where you got the idea that I don't know much about machine learning and artificial intelligence. Quote: The key to using NLP across a machine is training it across a dataset you're using. If you don't train it it's more or less useless. One of the main keys of machine learning is error reduction, and if you aren't starting their you're just plugging and playing libraries. Obviously you need to train the model for it to realize any kind of success. How successful of general NLP algorithms do we have though? These things are usually trained for specific domains, and even then, they not all that great. NLP algorithms have to have a very high success rate to be useful in practice, especially for something as important as writing academic papers. Quote: Well now I know you meant a field researcher rather than say, a market researcher. You were referring to a "researcher" you were familiar with, when I was simply referring to the general term of employable researcher. For example law companies have people that do only research, and they can easily be replaced through AI technology which searches far better than them. I was referring to academia, or creators of knowledge like professors and grad students. I just didn't want to have to type "professors and grad students" each time and I thought the word "researcher" would've been clear enough given the context. I do not know what research at a law company is or how easy it is to automate that. I am referring specifically to people who publish original, scholarly work, and I think it's ridiculous to say that their jobs will be replaced by machines any time in the near future or at all. Quote: Google's search has been one of the least replicable form's of AI today. We know exactly what we're doing with neural networks, lol. I've used and made them and they're more of a simple concept than people realize (at least in abstraction) The single biggest reason why no one can replicate Google's algorithm is because other institutions simply do not have the physical resources that Google has available. Peter Norvig himself told us this last year. One of the AI/robotics professors here admitted that it is becoming more and more difficult for universities to compete with industry in the field of artificial intelligence, for this very reason. Companies have both the interest and the resources to conduct research in these areas and are pulling ahead of academia. This kind of research is slowly becoming obsolete as industry takes over the field. Other than fiddling with the parameters of neural networks, we really do not know at a fundamental level what we are doing. We have basic principles like "more layers will make for more accurate ANNs" but that's about it. We use things like genetic algorithms to optimize ANN parameters, another class of algorithms which we hardly know anything about. In my opinion, the "simplicity" of neural networks lies in the statement of the Universal Approximation Theorem, which as you may know, states that any continuous function on a compact subset of Euclidean space can be approximated to an arbitrary precision by a neural network of perceptrons with one hidden layer. This theorem adequately justifies the use of neural networks in the first place, but the approximation guarantee of the theorem is very weak. It says nothing about how to compute parameters that define a neural network with a certain precision. So we have to just try parameters and see what works best, and we have some heuristics to guide us in parameter selection, but they're not very good. In this sense we don't know what we're doing. This kind of problem more broadly affects AI algorithms. For example, value iteration is guaranteed to converge, eventually, but we hardly know anything about rates of convergence. It is my impression that there's lots of room for theory to be developed in these fields, and until then, we're stuck with dumb old heuristics. Edited by Common2, Sep 19 2015, 06:35 AM.



1 user reading this topic (1 Guest and 0 Anonymous)


« Previous Topic · Deep Discussion · Next Topic »

Pages:
1
2