Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

MIRI's Approach – Machine Intelligence Research Institute

intelligence.org

Post reply on HN
Print view View original

51–60 of 78 posts

Re: MIRI's Approach – Machine Intelligence Research Institute

#51

Post by rasz_pl »

reasonattlm wrote: post #10

MIRI is one of the success stories to emerge from the transhumanist community of the 80s to 00s. Others include the Methuselah Foundation, SENS Research Foundation, Future of Humanity Institute, and so on. When it comes to prognosticating on the future of strong AI MIRI falls on the other side of the futurist community from where I stand. I see the future as being one in which strong AI emerges from whole brain emula…

>whole brain emulation

we dont need whole thing

http://rifters.com/real/articles/Oliveira-et-al-2012-Revisit...

Re: MIRI's Approach – Machine Intelligence Research Institute

#52

Post by chriswarbo »

sampo wrote: post #46
sanxiyn wrote: post #38

Earlier quoted context omitted.

You don't consider AIXI part of the actual AI research community? Also, AIXI is non-computable, but AIXItl is.

> You don't consider AIXI part of the actual AI research community? I just briefly tried to read up on AIXI. The concept is ...not without merit, but I think it totally sweeps the central problem (We have no idea how to program an AI) under the rug in two ways: (1) The problem of programming ...programming anything at all, is swept under the rug by assuming a formalism in which every possible algorithm/program can be…

You're right, of course; AIXI doesn't attempt to tackle the practical problem of building an AI. However, it does give us a concrete, non-hand-wavey algorithm which can reasonably be considered "AI, if we ignore resource constraints".

Consider that it took ~50 years to go from the inception of AI to a formal model like AIXI; or ~40 years from the definition of NP-completeness (ie. the realisation that scalability and resource usage are the real challenges for AI).

In that sense, AIXI provides two things: a hypothetical "gold standard" for AI builders to compare their research to; and a formal model which can be studied right now by those who aren't directly building AIs (like MIRI).

Consider an analogy to space flight. The engineering contains all kinds of resource constraints (eg. launch mass, strength-to-weight ratios, etc.), but it's still useful to ignore them temporarily and ask: what if we had as much fuel as we could ever want? What if we could keep the crew frozen during the journey? and so on. In other words, if we manage to overcome our current difficulties, what could we actually do with this tech?

Re: MIRI's Approach – Machine Intelligence Research Institute

#53

Post by JulianMorrison »

idlewords wrote: post #15

Are there any examples of substantive AI work to come out of MIRI? And have they succeeded at all at engaging the actual AI research community? The last time I looked at them, they were consumed with grandiose philosophical projects like "axiomatize ethics" and provably non-computable approaches like AIXI, not to mention the Harry Potter fanfic. But I'm asking this question in good faith - have things changed at MIRI…

Given that what they are trying to do is, pull things out of the domain of philosophy and into the domain of comp sci, then it's quite likely that they are not trying to do what you'd consider substantive AI.

Re: MIRI's Approach – Machine Intelligence Research Institute

#54

Post by TeMPOraL »

idlewords wrote: post #15

Are there any examples of substantive AI work to come out of MIRI? And have they succeeded at all at engaging the actual AI research community? The last time I looked at them, they were consumed with grandiose philosophical projects like "axiomatize ethics" and provably non-computable approaches like AIXI, not to mention the Harry Potter fanfic. But I'm asking this question in good faith - have things changed at MIRI…

I don't have enough knowledge to evaluate MIRI's productivity, but I've noticed an interesting thing in this subthread. On the one hand, it is widely understood (especially here on HN) that the "publish or perish" culture of modern academia is a source of lots of bad science and pointless work, and yet here we are, using number of papers as a metric for productivity. So which way is it?

Re: MIRI's Approach – Machine Intelligence Research Institute

#55

Post by TeMPOraL »

JoshTriplett wrote: post #31
reasonattlm wrote: post #10

MIRI is one of the success stories to emerge from the transhumanist community of the 80s to 00s. Others include the Methuselah Foundation, SENS Research Foundation, Future of Humanity Institute, and so on. When it comes to prognosticating on the future of strong AI MIRI falls on the other side of the futurist community from where I stand. I see the future as being one in which strong AI emerges from whole brain emula…

Note that MIRI's current position no longer suggests the development of actual artificial consciousness , just the development of human-equivalent optimization processes. In other words, they argue that you can develop a process capable of solving human-level and harder problems without giving it self-awareness. And that seems like a feature: if you avoid building self-aware machine intelligences, you don't have to w…

Keep in mind that this does not sidestep the biggest practical concern with AIs, namely misalignment of values. You don't need a self-aware, conscious being to have a system with wants and values. In context of AIs, it's good to understand intelligence (including that of ourselves) as a very strong, multi-domain optimization process.

Re: MIRI's Approach – Machine Intelligence Research Institute

#56

Post by TeMPOraL »

dmfdmf wrote: post #24

I agree with their financial approach which is funded through voluntary donations, so more power to them. But then I think NASA should be funded voluntarily and would probably have a larger budget if they did. I would donate to NASA if they repudiated government money. I reject the "how many peer-reviewed, university-associated journal articles have they published" as any measure of success. The university system is…

Chinese Room is not an old, still unanswered philosophical problem. The answer is obvious - it's not the person that understand Chinese, but the system of person + room, with the room setup probably doing most of the work.

Re: MIRI's Approach – Machine Intelligence Research Institute

#57

Post by eli_gottlieb »

dmfdmf wrote: post #24

I agree with their financial approach which is funded through voluntary donations, so more power to them. But then I think NASA should be funded voluntarily and would probably have a larger budget if they did. I would donate to NASA if they repudiated government money. I reject the "how many peer-reviewed, university-associated journal articles have they published" as any measure of success. The university system is…

Machines without "subjective experience" or "semantic content" can still fire guns.

Re: MIRI's Approach – Machine Intelligence Research Institute

#58

Post by eli_gottlieb »

sampo wrote: post #32
pjscott wrote: post #17

Earlier quoted context omitted.

If you'd like to give them another look, here's an up-to-date list of their publications: https://intelligence.org/all-publications/

Counting only articles and conference papers that look like they are in at least somewhat established journals or conferences, I quickly count: 2015: 4 2014: 6 2013: 1 2012: 5 2011: 1 2010: 4 That would be about the level of 1 or 2 very mediocre early career scientists. And very little for MIRI's 8 people staff and 13 Research Associates. MIRI publishes a lot more, but they mostly operate outside of traditional scien…

How many of their staff hold PhDs, particularly in cognitive science, foundations of math, or AI? It would seem like a useful thing for them to get.

Re: MIRI's Approach – Machine Intelligence Research Institute

#59

Post by eli_gottlieb »

shoo wrote: post #50
anonmeow wrote: post #48

Earlier quoted context omitted.

AIXI was neither invented by MIRI nor have they proved any nontrivial theorems about it. AIXI is Marcus Hutter's work: http://www.hutter1.net/ai/aixigentle.htm

Indeed! From a superficial inspection, Hutter's back catalogue features 89 * @inproceedings, 3 * @book, 12 * @techreport, 51 * @article, and 1 * @compression prize Over an elapsed period of '87 (from before he had completed his masters degree) to the present, on a published_artefacts/year basis, Hutter is more productive than MIRI.

Of course, how many of those publications are Yet Another AIXI Paper?

Re: MIRI's Approach – Machine Intelligence Research Institute

#60

Post by eli_gottlieb »

chriswarbo wrote: post #52
sampo wrote: post #46

Earlier quoted context omitted.

> You don't consider AIXI part of the actual AI research community? I just briefly tried to read up on AIXI. The concept is ...not without merit, but I think it totally sweeps the central problem (We have no idea how to program an AI) under the rug in two ways: (1) The problem of programming ...programming anything at all, is swept under the rug by assuming a formalism in which every possible algorithm/program can be…

You're right, of course; AIXI doesn't attempt to tackle the practical problem of building an AI. However, it does give us a concrete, non-hand-wavey algorithm which can reasonably be considered "AI, if we ignore resource constraints". Consider that it took ~50 years to go from the inception of AI to a formal model like AIXI; or ~40 years from the definition of NP-completeness (ie. the realisation that scalability and…

>However, it does give us a concrete, non-hand-wavey algorithm which can reasonably be considered "AI, if we ignore resource constraints".

No, not really. "If we ignore resource constraints" is ignoring most of the problem. Using Kolmogorov complexity in the Solomonoff Measure also constitutes ignoring the problem of generalization by assuming an optimal compressor into existence, which again is an issue of the cognitive resources of training data and processing power. Bayesian updating means it will achieve optimal expected reward, but also that AIXI can be "fooled" by the hierarchical nature of real environments' variance[1].

And the whole thing pays no attention to knowledge representation whatsoever.

It's basically a grand victory for the fields of AI and Machine Learning that still tells us basically nothing about how an actually existing, embodied mind has to function, except that statistical learning is most likely the core mechanism in some fashion (after all, neural networks show that the real thing isn't even necessarily Bayesian in any sense).

[1] Benjamin B. Machta, Ricky Chachra, Mark K. Transtrum, and James P. Sethna. Parameter space compression underlies emergent theories and predictive models. Science, 342(6158):604–607, 2013.

Post reply on HN