Once again, I've noticed that the people behind the challenge have turned the whole thing into a game and not a viable idea that can be brought to the mass market. Its frustrating that you have a prototype that actually works (only 1 collision was reported during all the tests) and it isn't yet being incorporated into real cars.
Best place to recruit volunteers for computer science project
- Thread starter sunama
- Start date
Once again, I've noticed that the people behind the challenge have turned the whole thing into a game and not a viable idea that can be brought to the mass market. Its frustrating that you have a prototype that actually works (only 1 collision was reported during all the tests) and it isn't yet being incorporated into real cars.
Typically those challenges are more to uncover the human talent rather than the specific systems they make for the competition.
I recently took part in the MoDs version of that, the Grand Challenge (google it if you want) as part of team TumbleWeed.
While we didnt win, there have certainly been lots of opportunities that were borne from the competition, be it through inter-team relations or further work with the MoD.
While technically, the winners were the ones to get contracts and further funding etc, im not sure anyone has seen anything other than benefit from the experience.
What you see in the Darpa will eventually filter down into the market, either directly or by the people involved becoming involved in the market and bringing their ideas to the table. Either way, things take many many years!
I recently took part in the MoDs version of that, the Grand Challenge (google it if you want) as part of team TumbleWeed.
While we didnt win, there have certainly been lots of opportunities that were borne from the competition, be it through inter-team relations or further work with the MoD.
While technically, the winners were the ones to get contracts and further funding etc, im not sure anyone has seen anything other than benefit from the experience.
What you see in the Darpa will eventually filter down into the market, either directly or by the people involved becoming involved in the market and bringing their ideas to the table. Either way, things take many many years!
Soldato
- Joined
- 27 Mar 2004
- Posts
- 14,081
- Location
- Between Realities
It's never going to happen on 1 PC.Yep.
Normally, people have a sit at a computer terminal, typing out key words into a search engine to learn about a subject. I would like my program to be able to take verbal commands and then output the results using sound (aurally).
Imagine waking up in the morning and while brushing your teeth, you ask your computer what the weather is going to be like today. I would want the AI to be able to find this out and tell you before you come out of the bathroom.
Certainly.
This is where I think a lot of time will be spent. To give key information to the user without them having to babysit the AI, during the search process.
DEFINITELY.
Imagine buying the program and the from the moment you install it, it is "born". From here on in, it will live on your PC (or similar) and will account for every minute that it is on (ie. not in sleep mode, when the computer is off).
- From its birth, it keeps a record of EVERY conversation it has with you.
- It will learn to recognise your voice (voice recognition), just as you would expect a human to do.
- If the AI is hooked into the house and has access to telephones, in the event of a break in, it will be able to contact the authorities (and its master) of unrecognisable voices/movements in the house.
- It will continually learn the user's slang/language.
- It will keep a record of the user's likes and dislikes, what their hobbies are, who lives with them, what the user's name is, as well any pets that might live in the house.
- It will know everything about the user (that the user wants it to know).
- It will store all this information on hard disks (including back up hard disks).
- It will download information from the internet, on subject matter that the user finds interesting, so that when the user wishes, it can have an intelligent conversation, about the subject matter.
I have loads and loads of other ideas of what I would want this program to be able to do.
More pertinent to this thread, I will definitely need to hook up with people with diverse "computer skills".
Soldato
- Joined
- 7 Feb 2004
- Posts
- 9,511
Sounds like you have been watching too many Sci Fi programs.
What is your back ground with this? Are you educated in this field or are you some 14 year old with an idea?
No offence but some of the top minds in the world are probably working on this and although it is possible it is highly unlikely that you are going to crack it over night.
Rather than waste peoples time, could you tell us about yourself a little, education, do you have any other projects that have worked. Is this your first? Is it related to a degree of sorts that you are learning?
What is your back ground with this? Are you educated in this field or are you some 14 year old with an idea?
No offence but some of the top minds in the world are probably working on this and although it is possible it is highly unlikely that you are going to crack it over night.
Rather than waste peoples time, could you tell us about yourself a little, education, do you have any other projects that have worked. Is this your first? Is it related to a degree of sorts that you are learning?
One very interesting area I think you will find challenging is that of catagorisation (hence I said to really look into current ontological techniques).
How, for example, does the AI go away and download data from the internet and then verify that it is correct? Data classification is a huge challenge in itself (just look at the data mining community). Then, once it has formed this classified database of information, how does it re-organise it into knowledge as we know it. Certainly there are rules that we as humans use in order to catagorise data into what we call knowledge. Typically we use our penchant for cognitive speech to get across our point, sometimes using graphical or statistical representations of said data in order to backup our case.
What you are suggesting is not impossible, very little is impossible, but it is a very challenging idea that will bring together many different areas of CS and indeed other academic areas.
I can imagine, for example, chatting briefly with the AI about the possibilities of aliens one day, only for it to then spend a few hours collecting every single crackpots ramblings on the web about aliens, how would it catagorise what is factual and what isnt, how would it present that data. We know that the majority of alien "sightings" seen on the web are manufactured to decieve, as a result they are presented as fact on the web.
If the AI then spouted tonnes of garbage about aliens it would quickly get annoying. One way around that sort of thing would be to catagorise based on current verified scientific discovery, so if it hasnt appeared in a journal etc, then it can only be considered as rumour by the AI, but then there are lots of erroneous / hack scientific papers out there that get published with no knowledge of their peer reviewed ratings etc, so how do we distinguish which is reputable?
The list goes on, its a self replicating problem almost, designing an all encompassing Ontological descriptor for all and any data out there on the web would be a feat so impressive I reckon it would be worth a Noble prize.
Seriousely, though, as negative as I sound, I applaud anyone who even attempts to start anything of this magnatude and really would like to be kept updated, as i'm sure many others would.
How, for example, does the AI go away and download data from the internet and then verify that it is correct? Data classification is a huge challenge in itself (just look at the data mining community). Then, once it has formed this classified database of information, how does it re-organise it into knowledge as we know it. Certainly there are rules that we as humans use in order to catagorise data into what we call knowledge. Typically we use our penchant for cognitive speech to get across our point, sometimes using graphical or statistical representations of said data in order to backup our case.
What you are suggesting is not impossible, very little is impossible, but it is a very challenging idea that will bring together many different areas of CS and indeed other academic areas.
I can imagine, for example, chatting briefly with the AI about the possibilities of aliens one day, only for it to then spend a few hours collecting every single crackpots ramblings on the web about aliens, how would it catagorise what is factual and what isnt, how would it present that data. We know that the majority of alien "sightings" seen on the web are manufactured to decieve, as a result they are presented as fact on the web.
If the AI then spouted tonnes of garbage about aliens it would quickly get annoying. One way around that sort of thing would be to catagorise based on current verified scientific discovery, so if it hasnt appeared in a journal etc, then it can only be considered as rumour by the AI, but then there are lots of erroneous / hack scientific papers out there that get published with no knowledge of their peer reviewed ratings etc, so how do we distinguish which is reputable?
The list goes on, its a self replicating problem almost, designing an all encompassing Ontological descriptor for all and any data out there on the web would be a feat so impressive I reckon it would be worth a Noble prize.
Seriousely, though, as negative as I sound, I applaud anyone who even attempts to start anything of this magnatude and really would like to be kept updated, as i'm sure many others would.
Sure thing:
A Levels in Biology, Chemistry and Maths.
BSc in Mathematics and Statistics
MSc in [Object Oriented] Programming (SQL, Java, C++, project management)
This would be my first big project, apart from those that I've done for my Uni degrees and a project that I worked on turning my digital TV card into a PVR (before PVR software was available) with fully automated recording using DigiGuide.
Programming is a hobby of mine. My languages of choice are Java and Visual Basic. I can do C++, but I'm not a big fan of it (too easy to write bad code).
LOL it seems that I'm writing a personal statement here.
With regards to SciFi programs - a lot of ideas start from that and with time, they evolve into something else that eventually may not even resemble the original idea. Some SciFi ideas do actually make it into real life, but it requires vision and an open mind.
And this isn't about wasting peoples' time. All major ideas have to start somewhere. However, I realise that if I want to make this happen, I will have to listen to 1000s of people telling me, "Its just not possible and that I shouldnt proceed with this."
Ideally, the people who I would want to work with are those who are adventurous and are willing to throw out all conventional rules of what can or can't be done. A can do attitude is very important.
Oh and I'm not 14.
A Levels in Biology, Chemistry and Maths.
BSc in Mathematics and Statistics
MSc in [Object Oriented] Programming (SQL, Java, C++, project management)
This would be my first big project, apart from those that I've done for my Uni degrees and a project that I worked on turning my digital TV card into a PVR (before PVR software was available) with fully automated recording using DigiGuide.
Programming is a hobby of mine. My languages of choice are Java and Visual Basic. I can do C++, but I'm not a big fan of it (too easy to write bad code).
LOL it seems that I'm writing a personal statement here.
With regards to SciFi programs - a lot of ideas start from that and with time, they evolve into something else that eventually may not even resemble the original idea. Some SciFi ideas do actually make it into real life, but it requires vision and an open mind.
And this isn't about wasting peoples' time. All major ideas have to start somewhere. However, I realise that if I want to make this happen, I will have to listen to 1000s of people telling me, "Its just not possible and that I shouldnt proceed with this."
Ideally, the people who I would want to work with are those who are adventurous and are willing to throw out all conventional rules of what can or can't be done. A can do attitude is very important.
Oh and I'm not 14.
It certainly wont be completed overnight. In fact, as I stated in a previous post, the development of the AI is likely to take decades, however, working prototypes of a lower generation AI could be up and running in a much shorter time period, depending on how many people work on it.
There are indeed some very intelligent people working on this. But as seems to be the case with most AI projects, they come with a lot of theories of the direction that AI should take. These "minds" never seem to come out with a working prototype with a working AI for the masses and this is what I want to do. I dont want to theorise or debate, I just want to produce a software that joe public can use and develop it as quickly as possible.
Well without wishing to sound too crass, theorising and debate is the only way things get done properly.
You try diving in headfirst into producing Skynet and come back in a years time and let us know if those millions of scientific papers written by people over the past 50 years should have simply been considered during the development of one piece of software.
This is how science works, many people work on many different aspects, each building on each others work. Eventually the problems are distilled to the point where somebody thinks "hmm that could be useful actually!" and goes away and makes something simple. Then somebody else improves on it, ad infinitum.
Should Newton have discovered mavity as a result of him thinking one day "hmm i'd like to float", or is it better that he considered its existence first, resulting eventually in people thinking "hmm it may be possible to float, I wonder how we overcome this mavity stuff".
You are also assuming that there already arn't millions of people around the world working on fully fledged AI projects. Just because you dont have access to them directly, doesnt mean they arnt out there, running on some server in a University or company, being tested daily.
I still applaud your can do attititude, but I also stress that perhaps you are being too gung-ho in dismissing the fact that many are already doing what you suggest and are finding it very hard to get right. By all means, get stuck in, but personally I dont see the reason to re-do work (as you will inevitably do) others have already perfected, simply because you dont want to really read around the area and think a bit first.
You try diving in headfirst into producing Skynet and come back in a years time and let us know if those millions of scientific papers written by people over the past 50 years should have simply been considered during the development of one piece of software.
This is how science works, many people work on many different aspects, each building on each others work. Eventually the problems are distilled to the point where somebody thinks "hmm that could be useful actually!" and goes away and makes something simple. Then somebody else improves on it, ad infinitum.
Should Newton have discovered mavity as a result of him thinking one day "hmm i'd like to float", or is it better that he considered its existence first, resulting eventually in people thinking "hmm it may be possible to float, I wonder how we overcome this mavity stuff".
You are also assuming that there already arn't millions of people around the world working on fully fledged AI projects. Just because you dont have access to them directly, doesnt mean they arnt out there, running on some server in a University or company, being tested daily.
I still applaud your can do attititude, but I also stress that perhaps you are being too gung-ho in dismissing the fact that many are already doing what you suggest and are finding it very hard to get right. By all means, get stuck in, but personally I dont see the reason to re-do work (as you will inevitably do) others have already perfected, simply because you dont want to really read around the area and think a bit first.
One method might be to verfiy the data, from varying sources. EG. Michael Jackson's death. A VERY popular news story and notice that I've chosen a subject that is interesting to the masses, as the AI will be used by the masses and not necessarily by geniuses and professors.
The AI will have built into it, certain web sites/sources that it will visit in the first instance. For example, the BBC website. Once it identifies that the top news story is that MJ is dead, it will realise that this is a story that might interest the user. AI will then verify this news story with another popular news site and then another, collating information, before informing the user that, "Michael Jackson has passed away, would you like to hear more?"
There will be other information, for example, that might not be so popular and may be more difficult to verify. Lets say, "Whats the population of London?" AI will search for this information, come up with numbers and perhaps give a trimmed mean of numbers it finds. The result, "The population is approximately X, based on a trimmed mean, from multiple sources".
Obviously, we would need to make the output more user friendly and close to the way a human would speak, but you get the idea.
I want to re-iterate that AI must be attractive to the masses and not geeks/geniuses. Remember, this software will eventually have to presented/marketed to people who will know very little about computers. They'll need to see a working demo/prototype that works at a basic level.
Oh, I totally agree. In the first instance, it will be important to develop the AI to a point where it can have a basic conversation with the user. A later step will be to hook it upto the internet and increase its knowledge base, then to verify and organise this knowledge - but bear in mind that this would probably later down the line.
One method might be to ask the user for extra input. Consider the following conversation:
User: AI, I wish to discuss Aliens.
AI: I don't know much about Aliens, would you like me to look it up?
User: Go for it. How much time do you need?
AI: 3 minutes, max.
User: <waits>
AI: I have factual information and anecdotal/unverified opinions...which would you like to discuss.
User: Both
AI: OK you go first.
User: I believe aliens exist. There are far too many planets out there for Earth to be the only one with life.
AI: According to xxx source, the chances of life are ... There are also opinions from people that they have seen aliens at ...
AI could punctuate information with 'factual' or 'opinion'. It will never tell the user that what it is saying is fact (when it might not be).
Also, the user will have the option to ask AI to delete any information that he chooses. So, if the user feels that AI has got itself in a muddle, he could ask for certain sources to be deleted, keeping only the sources the user wishes to be kept in AI's memory banks (for later discussion).
Similarly, if the user feels that what AI has discussed is 100% correct, then the user will inform AI of this. AI could perhaps then categorise that bit of information under "factual". Alternatively, AI might have a database which stores information with certain level of perceived accuracy, rated anywhere between 0 - 100%.
Remember, AI will have a learning mechanism, which means that with time, it gets wiser, storing away factual information and discarding information that it or the user feels is incorrect.
Once searched and stored, AI will always have access to the information it previously gathered and verfied, negating the need to carry out the same searches. There is nothing to prevent it from "updating" its memory banks, though, perhaps while the user is away/asleep.
I do realise that its difficult and in our lifetimes wont be nailed 100%, if ever. However, there is no reason why as humans we can't attempt to make something that can replicate the human thought process and in time develop it so that we can get closer and closer to that 100% figure, where an AI can have a rock solid conversation about anything and everything, without getting itself in a muddle or requiring intervention from the user.
Actually, you are not negative. You are promoting good discussion and pointing out the problems that anyone developing an AI like I've outlined, will have to deal with. I have no problem with constructive criticism and enjoy finding solutions to problems.
I have no doubt that AI is being developed by many commercial and non-commercial organisations. But what I feel is that the AI systems others are developing are more specialised in a particular field. EG. Car makers are developing AI which can drive a car by itself, or the millitary are developing AI that can fly aircraft by itself, etc.
What I'm proposing is an AI system that can be used at home, by joe public. As I stated earlier, there is always room for products that fall into the same category...if this wasnt the case then there would only ever have been one brand of car, one brand of television, one brand of gym equipment, etc.
At some point, somebody has to stick their neck out and make a start on something that may or may not "make it". They then have to develop the product and then bring it to market. This is a long journey with many pitfalls. I totally understand this.
I will certainly read around the area. That goes without saying. At the stage that I'm at right now, I want to bounce ideas of like-minded individuals, who want to contribute to something that might change the way many people live their lives.
Man of Honour
- Joined
- 17 Nov 2003
- Posts
- 36,749
- Location
- Southampton, UK
Just doing that is a huge task. The semantic web is trying to help do this, but it's a good few years away from being realised.
Yep and that is why what I've written is something that a later version of AI will behave like. An earlier version will be a lot more primitive.