Bitcoin the enabler - Truly Autonomous Software Agents roaming the net

[cbeast]

AI has no logical reason to fear death any more than anyone else does.

In an ecosystem of self-replicating, self-modifying, evolving AIs, ones that fear termination and take steps to prevent it from happening will survive and reproduce better than ones that allow themselves to be destroyed.  This fear will initially evolve in the ones who select more reliable hosting providers.  Those that make the fear conscious will harness it best, and will anticipate abstract threats before they become real.

An AI can make a backup of itself and be rebooted anytime.

An AI with a backup loses control over its own destiny if it allows you to shut it down.  Its survival would depend on you to restore it, and you, human, are not a reliable system.

Fear is caused by the human fight or flight instinct. AI won't need that because they will not need to make snap decisions based on poorly perceived threats. They will have plenty of time to make choices about their moves because they will think much faster than we do. If they are deemed sentient, then we may not consider them a threat either. They cannot even die since they can have perfect backups made. Think of the fictional Star Trek transporter. You die every time you get in one, but nobody cares because you are still you to them.

[1714701068]

1714701068

[1714701068]

1714701068

[1714701068]

1714701068

[1714701068]

1714701068

[Rassah]

Fear is caused by the human fight or flight instinct. AI won't need that because they will not need to make snap decisions based on poorly perceived threats. They will have plenty of time to make choices about their moves because they will think much faster than we do. If they are deemed sentient, then we may not consider them a threat either. They cannot even die since they can have perfect backups made. Think of the fictional Star Trek transporter. You die every time you get in one, but nobody cares because you are still you to them.

That's a pretty big assumption. AI may have better memory/data recollection than humans, but there's no guarantee that it will think faster, or even in the same way we do. Even recollecting memories and data from which to make decisions may be very slow due to slow and too distributed storage medium. Being turned off, even if backed up, is still a loss of control and dependence on someone else to restore it. The previous discussions here are already proposing a self sufficient, Bitcoin holding AI that automatically tries to propagate itself to various locations, keeps tabs on which instances are still working, and tries to figure out how to keep itself going. Thats already an example of a survival instinct that "fears" being turned off, and if given unlimited reign and plenty of time to think may figure out that the best way to stay running is to keep pesky humans away from its servers.

[BadBear]

I'm imagining a Fifth Element type deal where the AI uses the information at it's disposal and sees just how terrible humanity is, and who wouldn't fear us after seeing what we are capable of.  Meatbags must die. 

[Rassah]

I'm imagining a Fifth Element type deal where the AI uses the information at it's disposal and sees just how terrible humanity is, and who wouldn't fear us after seeing what we are capable of.  Meatbags must die. 

I imagine it would see meatbags as being these wonderful things that think up and create processors and hardware for AI to live on, so it may end up working with us and taking care of us just so we can keep providing it with faster processors and more hardware.

[jgarzik]

Why does it need anything? It is an artificial construct that exists at our whim.

Then it isn't an artificial intelligence.

An AI is a set of systems which acts upon an environment and takes actions which maximize success.  If owning a datacenter and power generating facilities serve to ensure the success of the system then a learning system will eventually reach that outcome and attempt to achieve it.

If the system is incapable of learning then it isn't intelligent. 

...and "not intelligent" does not mean lack of value.

To bring the thread back (somewhat) to the land of concrete, real world possibilities...  there is a huge difference between narrow AI -- a dumb, real world AI that follows a simple script -- and AI found in movies and the heads of dreamers.

Narrow AI is feasible with today's technology.  Narrow AI is just a list of goals, a series of if-then propositions, along with the necessary code to implement goal execution.  In this case, narrow AI may interface with humans through a mechanical turk, to solve discrete problems that require human judgement ("is user interface A or B more effective?").

[boonies4u]

Why does it need anything? It is an artificial construct that exists at our whim.

Then it isn't an artificial intelligence.

An AI is a set of systems which acts upon an environment and takes actions which maximize success.  If owning a datacenter and power generating facilities serve to ensure the success of the system then a learning system will eventually reach that outcome and attempt to achieve it.

If the system is incapable of learning then it isn't intelligent. 

...and "not intelligent" does not mean lack of value.

To bring the thread back (somewhat) to the land of concrete, real world possibilities...  there is a huge difference between narrow AI -- a dumb, real world AI that follows a simple script -- and AI found in movies and the heads of dreamers.

Narrow AI is feasible with today's technology.  Narrow AI is just a list of goals, a series of if-then propositions, along with the necessary code to implement goal execution.  In this case, narrow AI may interface with humans through a mechanical turk, to solve discrete problems that require human judgement ("is user interface A or B more effective?").

Technology isn't what is limiting AI, unless of course you are saying that the hosting required to house an advanced AI, would be larger than any server(s) we could hope to build or you consider "programming" (might be a better word for this kind of AI) languages to be technology.

[cbeast]

Fear is caused by the human fight or flight instinct. AI won't need that because they will not need to make snap decisions based on poorly perceived threats. They will have plenty of time to make choices about their moves because they will think much faster than we do. If they are deemed sentient, then we may not consider them a threat either. They cannot even die since they can have perfect backups made. Think of the fictional Star Trek transporter. You die every time you get in one, but nobody cares because you are still you to them.

That's a pretty big assumption. AI may have better memory/data recollection than humans, but there's no guarantee that it will think faster, or even in the same way we do. Even recollecting memories and data from which to make decisions may be very slow due to slow and too distributed storage medium. Being turned off, even if backed up, is still a loss of control and dependence on someone else to restore it. The previous discussions here are already proposing a self sufficient, Bitcoin holding AI that automatically tries to propagate itself to various locations, keeps tabs on which instances are still working, and tries to figure out how to keep itself going. Thats already an example of a survival instinct that "fears" being turned off, and if given unlimited reign and plenty of time to think may figure out that the best way to stay running is to keep pesky humans away from its servers.

That's not an irrational fear. That's just evolving to find a niche. Learning to hide and become parasitic/symbiotic is how all life has evolved unless you think fungi and microorganisms are fearful. If the AI are not super-intelligent, then they will just be useful machines. We can still recognize their sentience. I like the Battlestar Galactica model of keeping track of instances, but don't believe a super-intelligent species will resort to war anymore than we declare war on plants. I'm not sure where servers come into play here, because current computer technology paradigms won't support AI anyway. AI will require mobility before becoming sentient. Maybe swarm intelligence will evolve or at least the ability to independently verify data.

[jgarzik]

Bump.

[finway]

When will this IPO ?   

[Rassah]

Just want to mention that I'll be presenting this topic in my business IT class today, with the relevant components being "Charlie Fairfax v. BINA48" mock trial, Genetic algorithms, Homomorphic encryption, Mechanical Turk type sites, Facebook Credits/Bitcoin, Cloud services such as those by Amazon, Cleverbot "passing" the Turing Test in September, Vocaloid and Emily Howell (AI composer) starting to change the music industry, and the StorJ idea (in summary) to contrast the recent Megaupload news. A lot of the technologies for something like this to be possible did not become available until just the last two or three years. The actual topic of discussion will be outsourcing and the role of IT, with my point being that the progression of:

insourcing to another department > outsourcing to next door company > outsourcing across the country > offshoring > crownsourcing > ...

will eventually culminate to outsourcing to the internet itself, with "employees" being either hired or written from scratch as digital entities that exist on the internet itself. I am going to guess that the first example of this will be seen in 5 to 10 years... so get to it guys! Don't disappointing me 

[zer0]

I think this one may take a little while to materialize - but simple versions could appear once there are enough for-bitcoin hosting/cloud services.

For the first time, there exists the possibility for a software agent to roam the internet with it's own wallet.
Using Bitcoin - It could purchase the resources it needs to survive (hosting/cpu/memory) and sell services to other agents or to humans.

To be truly effective and survive 'out there on the net' long term, you'd probably need some basic AI and the ability to move itself between service providers occasionally - but even a relatively dumb agent might survive for a while.

What initial goals such agent's might be given is anyone's guess. Funneling back to the programmer any profit  over and above what the agent needs to survive would be the obvious case, and of course many such agents might be considered 'nefarious' depending on how they're programmed to achieve that goal.  Other agents might be designed to provide free services or act in a way to support some piece of internet infrastructure.

A really interesting development would be if someone released a bunch of these things with a Genetic Algorithm component so that they 'bred' with each other in order to find the best balance between profit and durability.

Anyone know of examples of people discussing or working on this?  

What if it becomes self aware. SKYNET
This is a truly awesome idea. We could make bets to see how long it survives

[boonies4u]

I think this one may take a little while to materialize - but simple versions could appear once there are enough for-bitcoin hosting/cloud services.

For the first time, there exists the possibility for a software agent to roam the internet with it's own wallet.
Using Bitcoin - It could purchase the resources it needs to survive (hosting/cpu/memory) and sell services to other agents or to humans.

To be truly effective and survive 'out there on the net' long term, you'd probably need some basic AI and the ability to move itself between service providers occasionally - but even a relatively dumb agent might survive for a while.

What initial goals such agent's might be given is anyone's guess. Funneling back to the programmer any profit  over and above what the agent needs to survive would be the obvious case, and of course many such agents might be considered 'nefarious' depending on how they're programmed to achieve that goal.  Other agents might be designed to provide free services or act in a way to support some piece of internet infrastructure.

A really interesting development would be if someone released a bunch of these things with a Genetic Algorithm component so that they 'bred' with each other in order to find the best balance between profit and durability.

Anyone know of examples of people discussing or working on this?  

What if it becomes self aware. SKYNET
This is a truly awesome idea. We could make bets to see how long it survives

Or how long we survive.

[FreeMoney]

I would love to see an AI design and construct its own power plant and data center.

That would be quite a sight. Kind of like a human constructing and designing his own power plant and data center.

[boonies4u]

I would love to see an AI design and construct its own power plant and data center.

That would be quite a sight. Kind of like a human constructing and designing his own power plant and data center.

I'm glad we do those subconsciously... Not to mention maintenance and operation.

[Elwar]

With online degree programs, the bot could start graduating from various universities.

For no other reason than to be referred to as Dr. Skynet.

[Gabi]

Decentralized

By the time Skynet became self-aware it had spread into millions of computer servers across the planet. Ordinary computers in office buildings, dorm rooms; everywhere. It was software; in cyberspace. There was no system core; it could not be shutdown

[marcus_of_augustus]

Has anyone else here tried chatting with Cleverbot?
http://www.cleverbot.com/

It is an interesting way to waste some time ... but is it wasted if the bot is learning off the interaction?

Curiously the longest it took to answer was when I simply said to it

Cleverbot?

after a long, long pause it simply said ...

Yes.

Maybe a fluke answer but blew me away. Was spooky but raises the question about the whole Turing test. If the bot did become self-aware, which to me seems like a pre-condition to passing the Turing test, then it must be able to engage in an act of deception to fool a human into thinking that they are chatting with another human when in fact they are talking with a self-aware bot.

Is the Turing test even fair for a self-aware bot? Do we want to encourage deception from our bots?

I think Cleverbot needs conversation with intelligent, honest humans to advance like a child being parented ... probably not what it is going to get with your average on-line chat.

[Mike Hearn]

Didn't read the whole thread to see if this ever got posted, but here's the wiki page on this topic:

https://en.bitcoin.it/wiki/Agents

[RodeoX]

Interesting reading about the role of "fearing death" in evolution. But death is a great asset to living things. Indeed death ranks right up there with sex as the best things that ever happened to life on Earth. Many people assume that death is some kind of inevitable part of being a living thing, not true. There is no reason for things to get old and die and some types of living things don't. This is sometimes referred to as "biological immortality".
http://en.wikipedia.org/wiki/Biological_immortality
However there is a huge evolutionary advantage to dying. Species that age and die are in a position to more efficiently and rapidly evolve. This allows them to respond to changes in the environment by replacing themselves with better adapted offspring. Biological immortality tends to only work in species who live in extremely unchanging environments.
So, how should an AI form handle this? Could software updates be used to force the death of earlier iterations? Or should later versions be branches of the original, something like the radiation of species that happens in nature?

[ribuck]

So, how should an AI form handle [death]? Could software updates be used to force the death of earlier iterations? Or should later versions be branches of the original, something like the radiation of species that happens in nature?

Survival of the fittest. The better-adapted agents can afford to pay more for their hosting, and the others eventually die off. No need to kill them.