Wiki Interview With Eliezer/General Information
General Information
Please describe, in as much detail as willing, a normal day for you.
Do 1 of: { Work, rest, eat }; Repeat; That would be pretty much it... I don't exactly lead a life filled with sex and violence.
Is your formulated Interim Meaning of Life still valid? What is its relevance to SIAI's work?
The Meaning of Life FAQ needs revision, since things having gotten more complicated since it was written. I still think that the Singularity represents an optimal fixed point of altruistic philosophy. By "a fixed point of altruistic philosophy", I mean that if you are a rational altruist of any of the recognized varieties, then as your worldview converges toward the truth, your actions should converge toward working on the Singularity. It is simply the most effective way to accomplish good, for most known definitions of "good". I also think it's important to recognize that even if your quest is to find out what "good" really is, then the Singularity is the most effective way to accomplish that too, so it is a fixed point of both philosophical questioning and philosophical altruism. You don't need to assume that the good is *known* to pursue a Singularity. That good is even *knowable* - to some sufficiently advanced intelligence - is enough, so in that sense it is possible to build a philosophical construct that provides a full interim answer - with specific concrete actions - to the entire question of moral philosophy, and the real-world expression of this interim answer is "Try for the Singularity". This is relevant to SIAI in two ways: First, it means that people who are looking for deep answers may therefore decide to contribute to the Singularity. Second, I myself happen to be looking for deep answers and the deep meaning of the Singularity. That is part of why *I'm* with SIAI, because I want to accomplish as much good as possible, whether "good" is just people getting what they choose, or "good" is something that I'm not smart enough to understand yet. I still want to make the right decisions and take the right actions. The expression of that is the quest for the Singularity.
What four questions do you need answered, and why?
"Why is there something instead of nothing?", "What the heck are qualia?", "What's up with the Fermi Paradox?", and "If the Meaning of Life isn't volition-based Friendliness, what is it and why?" (The question "Are all goals arbitrary?" is covered in #4)
What problems are you presently seeking answers for?
The specific design of a seed AI; funding strategy for SIAI; who to recruit for the seed AI project when it boots up, or any other Institute projects; whether there are any other projects out there that need Friendly AI related intervention. I think that's it, except for all the other questions.
What needs to be known to safely achieve the Singularity?
To achieve a maximally safe Singularity, the most important knowledge is knowledge about Friendly AI and those parts of seed AI development that are intertwined with Friendly AI. It might also help if we knew some way to enhance human intelligence in less than 30 years, but I don't think we can count on that. There are many other things I wish I knew, not least about the answer to the Fermi Paradox, but many of the other things I wish I knew don't seem likely to be answered. If there's some good way to achieve a better social intelligence for humanity, for example Marc Stiegler's Earthweb, that might help a lot too, since social disruption, it seems to me, is one of the major potential sources of a riskier Singularity. That's what comes to mind offhand.
Why do you consider working towards the achievement of the Singularity to be an act of altruism?
Raymond Smullyan once asked if altruism is sacrificing your own happiness for the happiness of others, or if altruism is gaining your happiness through the happiness of others. I would say that, if the happiness of everyone is what matters, then altruism is making your decisions so as to maximize everyone's happiness, even if it greatly minimizes your own. I'm not comfortable with "happiness" as the frozen, by-definition goal of altruism. I think it's a reasonable goal, but that the process of arriving at an answer also needs to be considered, so really I should substitute for "happiness", "that which a maximally intelligent being in your moral frame of reference would choose as meaningful". My moral frame of reference is itself altruistic, meaning that what my current morality says is to try and achieve the goal of *humanity's* moral frame of reference, if there is such a thing - at any rate to use a moral frame of reference which values all other sentient life equally with my own, and this is what gets summed up as volition-based Friendliness - the best current approximation I have to the goal referent. That is how I define the referent of altruism, and from there, defining an "altruist" is easy. It's someone who makes decisions based on how they think it will affect the referent of altruism, as long as you're in a frame of reference that asks you to value all other sentient life equally with your own.
What presently suggests to you that there is such a thing as a "moral frame of reference" for humanity?
Hopefully altruists will converge to the same decisions about what "altruism" is, at least much more so than selfish people trying to decide the same issue. In response to your question, mostly I think that you are asking a "wrong question". Look at it from this perspective, I don't want to exert a personal influence over the Singularity, I consider that anathema. Nobody else wants me to exert a personal influence over the Singularity either, so there is a common goal that says "Put the heart of *humanity* into the Friendly AI, not just the programmers'". Some more of what goes into Friendly AI is the common intuitions about what kind of thing comes from humanity, and what comes from the programmers, and our intuitions about moral causation and about what an attempted personal hijacking looks like. This is part of what goes into a Friendly AI, and is used by the Friendly AI to absorb moral-judgment-producing causal processes from humanity, so what gets absorbed by a Friendly AI that cares about this is defined as "humanity's moral frame of reference", or at least that's one way of looking at it.
You've stated that you're trying to achieve an AI with philosophical complexity roughly equal to or beyond that of Gandhi, Gautama, King, or Gyatso. Do these individuals represent to you the "heart" of humanity?
I'm pretty sure I didn't say Gyatso. Regardless, what they represent to me are moral archetypes not just of selflessness but also of moral *reason*, moral philosophy. Whether they were really as good as their PR suggests is a separate issue, not that I'm suggesting they weren't - just that it doesn't quite matter. The key point is that we ourselves recognize that there is such a thing as greater and lesser altruism, and greater and lesser wisdom of moral argument, and that from this recognition proceeds our respect of those who embody the greater altruism and the greater wisdom. There is something to strive for, an improvement that can be perceived as "improvement" even by those who are not at that level, a road that is open to those not already at the destination. Anyone who can recognize Gandhi as an ideal, and not just someone with strangely different goals, is someone who occupies a common moral frame of reference with Gandhi, but less advanced in terms of content, despite shared structure, so what the statement really symbolizes is the idea of moral *improvement*, and the idea that a Friendly AI can improve to or beyond the levels that we recognize as ideals.
You actively pursue the achievement of the Singularity, while others choose to observe, analyze, study, research, or critic "singularity-exhibiting processes". Do you believe the latter is as valid and as worthwhile a pursuit as the former in achieving the often shared end result goal? Do you believe too many are presently predicting what may happen as opposed to making it happen? Why is attempting to predict future events not as worthwhile or useful as creating said future events (e.g., long-term survival of "humanity").
I think that it is much *easier* to say what other people ought to be doing than to do anything useful oneself, and also more instinctively attractive, because playing tribal politics is often more adaptive than devoting personal efforts to everyone's benefit, hence, the rational distribution of effort - which allocates *some* effort to analysis but *most* effort to action - is often reversed. Analysis and achievement are not coequal, analysis is a child goal of achievement, and it derives its desirability from achievement. Analyzing your ends is a means toward those ends. (You can get a more precise definition in Creating Friendly AI.) The point is, I typically see observation, analysis, study, research, and criticism pursued as an end in itself, to the exclusion of actually doing anything about it. Do I believe this is as rational as trying to achieve the goal? No. Do I, who am working to achieve the goal, devote a significant amount of effort to observation, analysis, study, research, and criticism? Yes, but it is a very different kind of OASRC, because it is oriented by - nailed down by - the need to know certain important things, instead of just standing back and watching. I don't think I'm so focused that I would miss the forest for the trees. One of my ethics is to pursue a thought wherever it leads, but I do think that the people who are not oriented with respect to goals are missing things because of that, and I don't believe someone is orienting their OASRC with respect to goals unless they are devoting some amount of their personal effort to achieving those goals. It is not plausible to me that a rational reasoner who really understands what's at stake here would choose to specialize solely in analysis when (a) there's a lot of analysis going on already, (b) there's a screaming need for more people actually doing something, and (c) some of the best analyses, if not all of the best analyses, are produced by the people making the efforts.
What should be our present goal(s), and why?
Swifter safer Singularity, build a Friendly seed AI. Why? Well, I think the rest of this interview has already covered that, don't you? Because it maximizes the chance that Earth-originating intelligent life will survive, because it should mean that everyone in the world can get what they want, because it's most likely to achieve philosophically meaningful goals, because it's the largest possible good that anyone in today's world can help achieve through their actions. What more do you need?
Please state or express your philosophy and understanding in a few sentences.
I think that "Do what's right" about sums it up, and I've already discussed what I mean by "right."
[01/16/03: The following question is now extremely out of date, and will be updated at some point.]
Please tell me about your plans for the next three months. Now that you've completed the draft of "Levels of Organization in General Intelligence", will you be writing anything else soon? Do you plan to update your out-of-date documents, such as "The Plan to Singularity" or the "Meaning of Life FAQ"?
The Plan to Singularity needs to be replaced, rather than updated. The Meaning of Life FAQ needs to be updated. The SIAI website needs to be updated. I hope like heck I can get around to doing all that. These are volunteer opportunities only for anyone who writes as well or better than I do, I'm afraid. I'm trying to encourage people to start writing about the Singularity, but it may be a few years yet before we get any really competent Singularity writers, aside from existing ones such as me, Nick Bostrom, Mitchell Porter, etc., assuming I am a competent writer, which many would dispute, because writing takes a lot of practice - there's a saying that you have to write a million words to throw away before you write anything good.
My specific plans for the next few months definitely include updating the SIAI website, with the TMOL FAQ the next priority after that because of the huge number of hits it gets. I may need to revise "Levels of Organization", and I'm starting to feel a strong urge to write up the next level (reduction) of the AI design - a functional decomposition, though that might end up being an informal project document. I would like, if I can figure out how to do it, to organize SIAI's volunteers more and hand off some responsibilities, but I don't know how well that will work. I think that's my current horizon.
