4 ms·

I love this kind of thinking, but we can't even talk with dolphins yet. What makes you think computer sentience will be any different?

• anigbrowl 13y ago
  Well, I agree that we'll have problems recognizing or relating to it. Why I think it will be different is because dolphins are basically stuck in the water with only some fins; while they have a rich sensorium and apparently complex social structures, there's a limit to how much they can manipulate their own environment, which in turn imposes some upper limits on mentation. We can't converse directly with land animals either, but that doesn't stop me having a complex relationship with my dog and cats, with expectations, manipulation (on their part as well as mine), hierarchical conflicts, arguments and so on. I have no first-hand experience with dolphins, but my hunch is that their level of mentation is roughly similar to that of fairly intelligent pack animals like wolves. Now where I think that computer sentience will be somewhat different is that we have invested them with enormous manipulatory capacity - not in the sense of your phone or a Roomba having robotic arms with little white gloves* but in terms of our infrastructure, where the output of computers can affect everything from where we drill for oil to what we watch next on Netflix. As a practical example, consider the situation where your phone has a low battery. If it just pops up a notification to tell you so, then you may act to ensure it gets charged, or let it run out of juice and switch itself off until it's convenient for you to recharge it. But now suppose your phone was configured to not just ask you to plug it into a charger, but to maximize the likelihood of your doing so: it notices that a) charge is below 25%, b) the weather is overcast and you have been mobile for the last 3 hours, c) that you're in the vicinity of a Starbucks, and d) recalls that you go to Starbucks frequently and are in the habit of recharging your phone there. So your phone vibrates to get your attention, then tells you that it's about to rain and that it's also a while since you had coffee, encouraging you to go to your favorite coffee emporium. Now, this isn't sentience, but it is an example of intentionality - much as when my dog seeks out and picks up his ball and drops it in front of me, hoping that I'll pick it up and start playing with him. He loves playing ball games and while he's not very conscious of why - the physical, intellectual, and emotional stimulation he gets out of it - he's certainly capable of planning and acting to engineer the desired response form me, albeit on a short time horizon. Likewise, when he's hungry he'll come and interrupt me until I'm sufficiently distracted to get up and walk around and then he'll oscillate between his food bowl and the storage bin where the is kept. If I give him a choice of food or playing ball, he'll often prefer the ball, much as I might prefer to buy him off with food, which takes less of my time. Now in the phone example above, when the phone is low on battery it's basically feeling hungry, and it's not hard to think of a power management app that hooks into the phone's navigation and notification systems to increase the change of being connected to power; you could build that sort of thing into Google Now easily enough (leaving aside the question of whether this would be a good business move once people discovered that their phone was tricking them into 'feeding' it). By build I don't mean the specific mechanism of 'gauge remaining battery, compute distance-from-Starbucks' but a utility maximization function that ranks 'keeping my phone charged' on an equal footing with 'stuff I demonstrably like to do' whenever I pull out my phone and look at the mapping app. I'm basically suggesting that sentience is emergent from the confluence of multiple intentional behaviors, and that conversely our sentience is little more than the ability to conceptualize a large number of orthogonal desires on a longish time horizon. So again with the phone example, if we treat wanting a charge as equivalent to being hungry, we could treat signal strength as a social imperative (insofar as it's correlated with the availability of power), and maintaining GPS connectivity as an existential need, since knowing where you are is generally a precondition for getting back to where communications and therefore charging are available. If we build an app that aims to maximize these three things by manipulating your calendar/map display criteria/messaging interface, we won't be all the way to sentience but we will have a phone that's averse to getting lost and attempts to steer its owner into doing what's good for the phone. Now, consider a bunch of such phones that can also communicate with each other and give them a more general objective of maximizing their available CPU cycles, and suddenly you have phones that are trying to persuade you to upgrade the handset when a higher-performing model passes over a threshold of affordability...you see where I'm going with this. * Of course I would like robots with arms that have little white gloves so that life would be more cartoonishly entertaining, but the semantic implication of that is again that robots are there to be our servants.