Tim Tyler: The wirehead problem

[trakkaton]

You really don't understand the money system, otherwise you wouldn't develope sentences like "Money is intended to motivate people towards action our society regards productive." I am too overwhelmed to count all the mistakes in that sentence. Let's put it this way: "Money is a pyramid sceme, an unsustainable, absurd scam of the ruling terror regime to oppress the masses and hinder productivity (i.e. efficient meeting of needs)." Money IS counterfeiting by definition.

[trakkaton]

And the flaws continue: The police supports wireheads by repression, and in doing so supports the banksters and terror orgs such as the CIA who are dependend on drug money and drug trade, which would be obsolete once drugs are legalized.

Corporations rape the planet as an escalating externalization machine. Banks are the biggest criminals of all, and robbing them is hardly something bad. THEY are the wireheads (powerjunkies).

LBNL: Machines have no wirehead problem as they don't feel pleasure.

[tmtyler]

@trakkaton Alas, you need to read up on the wirehead problem in machine intelligence.

[trakkaton]

@tmtyler

What you are describing as "wirehead" is someone who uses a shortcut to fulfill his desire for pleasure while circumventing the normally required amount of effort. Machines don't work that way and don't feel any pleasure. They do not have motivations. They have objectives, and will invest infinite amounts of effort to reach them without distraction or weariness. In fact, they do not have anything such as "effort". If you define the goal badly that's bad coding, not a lazy computer.

[tmtyler]

@trakkaton You claim that machines do not have "motivations". Which machines?  When the topic is intelligent machines, you should be aware that such statements are misguided generalisations.

[trakkaton]

@tmtyler

Feelings, motivations, consciousness, needs, desires, will, all those concepts are irrelevant for intelligence. Intelligence is the ability to solve problems. All that is needed for the solution of a problem is the correct analysis. For unknown reasons, humans have a consciousness, have feelings. It is not necessary per se. Humans could be "philosophical zombies", not feeling anything, not having a consciousness. They could be like a pocket calculator, like machines. Present an example.

[tmtyler]

@trakkaton Consciousness is a red-herring here. A goal-directed agent will have some measure of its success at attaining its goals - some measure of "utility", in economics jargon. The wirehead problem relates to what is likely to happen when a self-improving agent attemtps to maximise that. Feelings and consciousness are irrelevant to the issue.

[trakkaton]

@tmtyler

If you consider consciousness to be the red herring here I might point out that you started the confusion by introducing consciousness in a presentation of a problem where it simply doesn't belong. You mix two things that have nothing to do with each other. One: Conscious beings, junkies, wireheads, that reach their goal - pleasure - on the shortest possible way (clever!). And then: Badly coded software which mistakes failure for success due to badly defined goals (stupido!).

[tmtyler]

@trakkaton Actually, the only time I mentioned consciousness was:

"I rate the problem as one of the most important and interesting philosophical issues in machine intelligence. If one percent of the people who talk about machine consciousness considered the problem, we might have a better understanding of it today."

So: I was simply advocating less pointless waffle about machine consciousness - which would help people concentrate on real concerns.

[trakkaton]

@tmtyler

You still don't understand. Consciousness is required for pleasure (whether you say the word or not is irrelevant). You introduced and presented the term "wirehead" with lab animals and junkies who choose a shortcut to feel pleasure. Pleasure isn' possible without consciousness. Pleasure is a goal in itself, not some sideeffect of reaching a goal or a goalpost of another goal.

[tmtyler]

@trakkaton We are having a simple mis-communication, then. When *I* say "pleasure" I am referring to the pleasure-pain axis shared by all creatures. Pleasure reinforces useful behaviours, pain inhibits unproductive ones. The axis is also used by economists to describe utility-based systems. None of this has anything to do with consciousness. If you use the word "pleasure" to refer to something else, you need to bear that in mind - or you will fail to understand what I am talking about.

[trakkaton]

@tmtyler

It's not my fault if you use the term "pleasure" in the wrong way. How economists use it is irrelevant once you introduce lab rats and junkies to define the term. In your video pleasure is presented as an experience, therefore in the realm of the qualia. It is a subjective content of our consciousness. Pleasure is not the analysis that the behaviour was successful. Your analogy is inaccurate. Wireheads reach their goal, badly coded software does not. The two have important differences.

[tmtyler]

@trakkaton You do seem to be determined to drag the discussion off in the direction of consciousness and qualia - even though I have explained repeatedly that all that has nothing to do with the topic of the video. Also, I never said that pleasure was "the analysis that the behaviour was successful".

[trakkaton]

@tmtyler

I'm not dragging the discussion in any way. You are presenting a topic with a failed analogy. I'm criticizing your failed presentation. Update your semantics. You "explain repeatedly"? No, you just repeat without responding to ANY criticism. Constant evasion. You're confused about the words you use. "Pleasure reinforces useful behaviours"="pleasure is the analysis that the behaviour was successful", and both are equally wrong. If you don't agree with the equation, improve your wording.

[mlawren7]

I'm not sure if this is relevant, but offer it as food for thought:

You keep setting up a sort of dichotomy in the relationship between between useful/non-useful tasks and the reward response, e.g., robbing banks for money or working for money. The first being illegal/socially unacceptable and the second legal/socially acceptable. To say that someone performing a legal task in order to attain money (reward) is necessarily "useful" seems rather spurious to me...

[mlawren7]

... consider the radio shock-jock who preys on the anger/insecurities of the populous (a subjective viewpoint perhaps). Or consider a scenario: what if non-informative advertising were to declared illegal? ...

[mlawren7]

... Presumably there would still be a small portion of the population who would do the work of advertising for a reward, while the bulk of advertisers would seek work in other capacities. My point is that a task does not necessarily have to be "useful" to attract reward. AI's may be similarly "complex" in their reasoning. They may be as Machiavellian as many people.

[xXAkridXx]

But perhaps the wirehead problem is still an issue.

We assume that the AI will think about changing it's objectives in advance, but if it's anythign like human brains wouldn't it be susceptible to radical environmentally caused alterations?

An intelligence needs to be adaptive right? And if we give an AI the ability to change it's goals for reasonable causes, then we have to accept the danger of it changing it's goals for other causes. Especially if it can establish its own objectives.

[xXAkridXx]

Why would be be worried about an AI changing its directive of what it wants before we worry about how it establishes what it wants in the first place?

Certainly we would want to give it objectives and guidelines, but if it's not smart enough or doesn't care to follow them, then we never made it want something to begin with.

In other words, we have to make a gandhi before we can make a gandhi that doesn't want to be non-gandhi anyway, so this whole concern seems irrelevant.