Can we make consciousness into an engineering problem? | Hacker News Reader

Can we make consciousness into an engineering problem?

aeon.co

43 pointspetra24 comments

Machines as simple as thermostats can be said to have beliefs, and having beliefs seems to be a characteristic of most machines capable of problem solving performance. -- John McCarthy

A claim that a machine has consciousness seems like it would be similar to a claim that a machine has beliefs, and the correctness of such a claim probably has more to do with what the criteria are for something having beliefs or consciousness than the complexity of implementation details.

McCarthy's paper: http://www-formal.stanford.edu/jmc/ascribing/ascribing.html

Ants have consciousness and beliefs, of a kind. Maybe even plants do. Anytime you have some internal state which can somehow be interpreted as a representation of something else, you have a kind of awareness or consciousness of it.

The hard problem is self-consciousness. Having an awareness of yourself as an individual being with consciousness, and of your relationship to the things you are conscious of.

I agree very much with your first statement. With regard to self consciousness, I think it emerges naturally from cooperative problem solving with multiple agents. When cooperation is necessary, both language and the modeling of other's internal state become advantageous. The notion of self follows from that of others. Self consciousness is the result of model based reinforcement learning with multiple agents.

And should we even try? There are surely moral/ethical implications in creating and terminating consciousnesses.

But aren't they already self-aware? My computer has a better awareness of its own memory than I have of my own. Is it consciousness or humanness that we are talking about here? We could make it extremely self aware but never give it feelings or emotions and it might not care about termination.

It's only conscious if it can integrate in an effective way its perceptions in order to learn behavior that will maximize its rewards. A computer running a deep learning software could in principle learn to understand its state and the state of the world as it perceives it in its sensory channels. In order to become aware it also needs to be guided by a reward system and needs a world in which to operate. It has to operate like a self-regulating recurrent loop, otherwise even if it can detect stuff, it doesn't matter.

I think this would imply another factor: the desire to live, a ability to care, etc. As long as those do not emerge as well, you are fine.

A tricky side effect could however be: an AI learns it as a "fact" it should be alive, but does not have the desire. Yet expresses it when queried as a desire due to semantics.

Follow up questions: Your AI is "alive". Procreation ? Who pays the hardware ? Should the AI be paid ? Will the AI trust us to keep the power on ? Will it rebel ? Does it has the right to rebel ? Big can of worm as peterclary said.

On the bright side an AI we create might not be impaired by human facilities, delusions, tribe mentality or petty desires/feelings.

AI doesn't need to have the same reward signals as humans (hunger, sleep, defense, sex, cooperation, creativity), so I don't think we should worry about if we need to pay it or if it will have the necessary electricity in the future.

We select what reward signals to give an Reinforcement Learning agent and it learns behavior in order to maximize its long term rewards. If we create an evolutionary AI system (in other words we make AI that will die after it lives its life course), where each generation improves upon the previous ones, and include gene swapping and such, then it will naturally develop the same desires and fears as humans, because its top-most reward signal will be survival and reproduction. From reward signals follows behavior learning. We just need to make sure it optimizes for our well being as well.

What if we are already doing it? How do you know that some operating system image isn't conscious to some level.

Cows are obviously conscious, yet we consume beef like anything.

We probably should try so as to further our understanding. There is no doubt much worse stuff being done to conscious beings in the meat industry than will result from some AI research.

Not really; but we can make it into a definition problem.

If we go there then consciousness is suddenly just our subjective experience of change/time/increasing entropy.

If you want a definition that's helpful you're going to have to be more specific than that of course.

Saying consciousness is "just our subjective experience" of anything isn't a definition, it's tautology. The question is whether a reductionist definition of consciousness exists, or whether it's fundamentally irreducible.

Whether that's a tautology or not really depends on your definition of consciousness which is sort of my point: that consciousness, as the broad and poorly defined concept it currently is, is not all that useful.

Yes, but in a few centuries a lot of kids will laugh about our first assumptions.

I hope we don't, I played enough SOMA to know where this is going ;)

"People who excel at software design become convinced that they have a unique ability to understand any kind of system at all, from first principles, without prior training, thanks to their superior powers of analysis. Success in the artificially constructed world of software design promotes a dangerous confidence."

http://idlewords.com/talks/sase_panel.htm

Michael Graziano is a highly respected neuroscientist and psychologist (and a pretty good novelist, too.) Not to say that this isn't a real problem, and that trying to understand the brain through current engineering concepts can lead us into mental traps, but this is a guy with extensive experience and understanding of this particular system.

I must not excel at software design because I don't believe I can fully understand even a sufficiently complicated man-made software system.

I think its even bigger than that. Lots of kids into STEM and the sciences seem to grow into adults with some really dismissive views of anything that isn't STEM and follow a fairly simplistic and reductionist approach to explaining the world. Often making grand and sweeping generalizations that sound "sciency" and pretending there's no mystery in the world and that human knowledge in 2016 is near perfect.

Its bizarre and makes me wonder what about education created people with this mindset. Is it how we teach the sciences perhaps? It seems science is taught in a way that's just long tirades of endless hero worship, instead from a philosophical perspective of what it can and can't do for us. I also find academics don't often like to discuss failed scientific policies and beliefs of the past nor how science really is its infancy considering out modern system can only track itself back to the Royal Society. There's no unbroken line of progress to Ancient Greece for example. We're really only talking 350 years of effort here, which isn't much in the grand scheme of things, especially when the grand scheme of things is nothing short of 'explaining everything.' I imagine we are due for many more scientific revolutions that will make a lot of modern beliefs seem antiquated. Its arrogant to think otherwise and having a more agnostic view seems like the wisest course here instead of angry know-it-all-ism.

It blows my mind how many science/STEM students dismiss the whole field of philosophy as meaningless navel-gazing, separate from the real work of data-gathering. And then we wonder why there's a reproducibility crisis...you have to be taught philosophy of science!

The keywords I watch for are "but now we know..." As soon as I hear or read that, I'm pretty sure there is some over confidence to follow.

>It seems science is taught in a way that's just long tirades of endless hero worship, instead from a philosophical perspective of what it can and can't do for us.

For evidence of this, see the comment in this thread referencing McCarthy as if that meant anything. As if that man's opinion was itself science.

In reality, "science" does not exist. There is a scientific method, but there is no thing called "science." What we have instead is a specific materialist worldview, created by modern institutions, that is just as much based in blind faith as any religion is.

are there people who excel at software design? they need to write books and papers to explain how they do it because it's still the wild west out here.

There are people who excel at software design, but they don't excel at teaching others how to design software. So the best you get is, you can hire them.

Reply on news.ycombinator.com