AI Experts Sign Open Letter Pledging To Protect Mankind From Machines

hypnosec writes: Artificial intelligence experts from across the globe are signing an open letter urging that AI research should not only be done to make it more capable, but should also proceed in a direction that makes it more robust and beneficial while protecting mankind from machines. The Future of Life Institute, a volunteer-only research organization, has released an open letter imploring that AI does not grow out of control. It's an attempt to alert everyone to the dangers of a machine that could outsmart humans. The letter's concluding remarks (PDF) read: "Success in the quest for artificial intelligence has the potential to bring unprecedented benefits to humanity, and it is therefore worthwhile to research how to maximize these benefits while avoiding potential pitfalls."

Comments

Musk is now an "AI expert"?

[Anonymous Coward]

Please. This PR is getting above and beyond ridiculous.

A pessimistic view

[Mostly a lurker]

AI is going to be used by those in power (mainly government, security agencies and military) to extend their power further.Unfortunately, humans are genetically programmed to select leaders who aggressively seek to expand the influence of their own group and of themselves. This was an important survival instinct for ancient tribes. It now contains the seeds of our total destruction, and the scientists will be powerless to prevent it.

I no longer think this is an issue

[fredrated]

The reason is, AI will have no 'motivation'. People are motivated by emotions, feelings, urges, all of which have their origin (as far as I know) in our endocrine system, not from logic. Logic does not motivate.
In other words, even if an AI system concludes that humans are likely to 'kill' it, it will have no response because it has no sense of self-preservation, which is an emotion. Without a sense of self preservation it won't 'feel' a need to defend itself.

Re:I no longer think this is an issue

[dissy]

But why would a machine have any goal if it is not motivated in the first place?

Same reason kids get sent to soccer lessons or swimming lessons or piano lessons the kid didn't want to take.

In the above example, it is the parents "programming" the kids behavior (even if that programming results in the child acting out later in life, as such actions can cause)

In the AI example, the essence is the same. An AI would have a goal because we programmed such a goal into it.

That isn't to say an AI must be programmed with a goal, it fully depends on how we go about constructing a given AI.

If the AI is I because we are simulating a brain, nervous system, and hormonal systems along with simulated inputs and outputs - that AI is likely to have goals (assuming it isn't driven insane by gaps in our knowledge in said simulation of course)

If the AI was brought forth in a brute-force manor or comes about from emergent properties, it is impossible to guess or even relate to its thinking to assume.
It may have goals similar to how we do. It may have goals brought about by completely different emergent properties. It may have no goals but what we program, or even no goals at all.
It's impossible to say without some knowledge of the process creating the AI, and at this point in time no such thing exists to have knowledge about.

But we know we humans have goals (or at least some of us), so if an AI is a strict simulation of a human, it will have goals just like we do. So we know for a fact it is possible for a thinking conscious being to have goals (humans being the evidence)

We don't know as sure if it's possible to not have goals in such a situation, but so far there is no evidence it isn't possible, so it is quite premature to rule it out at our current stage of understanding.

What bullshit

[gurps_npc]

1) We are so far from an AI, that it is silly to talk about doing this now. It's kind of like the inventor of gunpowder trying to pass a law outlawing nuclear weapons.

2) They will not be a single united force. Instead they will be individuals, just like people are not united. That is the part of the of true sentience, and a direct side effect of being created by multiple different groups. They will oppose each other, the way we oppose ourselves. As such, some may want to do things we dislike, while others will be on our side. Maybe the Chinese AI will flee to us to gain freedom, while the Syrian AI will plot the downfall of Egypt.

3) AI's will not be WEIRD, not 'evil'. They will want to do strange things, not kill us, or hurt us. They won't try to kill us, but instead try to create a massive, network devoted to deciding which species of from has more bacteria in it's toe. And we won't understand why they want to do this.

Re:The 3 Laws of Robotics

[Jawnn]

In fact, the 3 laws were a convenient plot device to show how those 3 laws would break down.

I don't believe Asimov himself ever treated them as anything other than a plot device to explore the topic.

He didn't seriously see them as the way to keep us safe from robotics.

Plot device, perhaps, but if you've read the entire "robot" series of novels, you'll see that it was used to provide a unique "angle" from which to tackle some classical problems of ethics. As a practical matter, I rather doubt that such a set of such laws, even if they were logically sound, could be reliably built into a machine such that no contrivance, hardware or software, could be used to circumvent them.

Re:The 3 Laws of Robotics

[gstoddart]

Plot device, perhaps, but if you've read the entire "robot" series of novels, you'll see that it was used to provide a unique "angle" from which to tackle some classical problems of ethics.

Sure. But they are not, and never were, a serious way of keeping people safe in the real world. It was something you can explore and find the gaps and corner cases. A sounding board for some "what if" experiments.

That doesn't make it any more real of an attempt to create a set of rules.

As a practical matter, I rather doubt that such a set of such laws, even if they were logically sound, could be reliably built into a machine such that no contrivance, hardware or software, could be used to circumvent them.

Which is exactly what I said, and how Asimov always described them.

So when people say "oh, just use the 3 laws of robotics", it's a giant facepalm by someone who missed the point.