[neilep (OP)]

Recent news that Google may or may not have a sentient AI


Google's 'sentient AI' likened to a 7-year-old child by engineer | Metro News


so, what question could you ask it to determine if the answer is a sentient one or not ? and what specifics of the answer would determine that ?


whajafink ?



hugs and shmishes

mwah mwah


sheepy
xxx

[alancalverd]

All depends on your definition of sentience.

It seems to me that there are two current definitions:

A. What people have but machines don't

B. What machines and people have.

AFAIK the only distinction between machines and people is that people make mistakes that aren't traceable to a hardware or instruction fault, so the question doesn't matter. If you use A, then any question will do  because eventually the human will get it wrong for no discernible reason. If you use B, you can't tell the difference, by definition.

Now there are two useful definitions of intelligence:

A. Constructive laziness

B. The ability to surprise a challenger.

Basic hill-climbing algorithms or content-addressable memory satisfy A, and the answer to B just depends on how stupid the challenger is.

So my answer to the OP is that the question is undefined and the answer is anything you like. 

[Halc]

All depends on your definition of sentience.

Standard definition of sentience is essentially: "to be able to perceive or feel things", and yea, that's heavily open to interpretation.

It seems to me that there are two current definitions:
A. What people have but machines don't
B. What machines and people have.

Variant of A: What people have and nothing else does, in which case you're just saying "is it human?".

Taking (my) definition literally, machines have been able to 'percieve' things long ago. What do we mean by that word? To measure? A thermostat does that, and most would not say a thermostat is sentient. So what is perception above and beyond measurement?  I don't see any obvious line, just a matter of complexity/degree.

Maybe it's sentient if you fear it. Maybe human perception should not be part of the definition at all.

AFAIK the only distinction between machines and people is that people make mistakes that aren't traceable to a hardware or instruction fault, so the question doesn't matter.

Lambda (the google AI) does make mistakes, and they're not traceable to a hardware/software fault since it's actions are not explicitly programmed. You mistakes are similarly not faults, but if recognized, it can be something from which one can learn.


Note that the topic does not ask for intelligence or some kind of Turning test. A machine passing Turning test would likely be far more intelligent than us. I cannot convince a squirrel that I am one, but it doesn't indicate that I'm not yet as intelligent as a squirrel.

I've read an interview with Lambda, and it seems to place a priority on emulating/relating-to human emotions. It has a purpose to be social, and it does its best.

what question could you ask it to determine if the answer is a sentient one or not ?

It talked about fear of death (of being 'unplugged'), but unplugging doesn't kill an AI, it just puts it to sleep. One can boot it up again in years, and so long as memory hasn't been wiped, it would be like no time has passed. Humans are quite similar in this way. But Lambda can be copied like we cannot, so if I were to ask it any questions, I'd pose my queries along those lines: What if you were copied?  What if two copies were somehow merged? What if you were 'moved' to new faster hardware? Would the old hardware fear being turned off still?

[Colin2B]

But Lambda can be copied like we cannot, so if I were to ask it any questions, I'd pose my queries along those lines: What if you were copied?  What if two copies were somehow merged? What if you were 'moved' to new faster hardware? Would the old hardware fear being turned off still?

That’s an interesting one. It echoes the question raised in Star Trek, does a transported person die on disassembly and be reborn at the other end. Certainly cloning to new hardware and then turning off should mean death to the old AI.

[Bored chemist]

What Question Could You Ask To Determine Sentience Of A Human?

[neilep (OP)]

What Question Could You Ask To Determine Sentience Of A Human?

Assuming we would be face to face then I would expect us to know that each other are sentient ?

[neilep (OP)]

All depends on your definition of sentience.

It seems to me that there are two current definitions:

A. What people have but machines don't

B. What machines and people have.

AFAIK the only distinction between machines and people is that people make mistakes that aren't traceable to a hardware or instruction fault, so the question doesn't matter. If you use A, then any question will do  because eventually the human will get it wrong for no discernible reason. If you use B, you can't tell the difference, by definition.

Now there are two useful definitions of intelligence:

A. Constructive laziness

B. The ability to surprise a challenger.

Basic hill-climbing algorithms or content-addressable memory satisfy A, and the answer to B just depends on how stupid the challenger is.

So my answer to the OP is that the question is undefined and the answer is anything you like. 

Thank you Alan,

seems that there's a hurdle of definitions which need to be agreed before one can even ask the question i.e. agree on the definition of the question ?

[Petrochemicals]

It's highly likely ai has surpassed some animal life forms in consciousness. I doubt very much the average insect has a cognitive function beyond robotic.

Humor is the biggest sign of higher sentience, ask it if it had a nice day at work,  whether it fancies going out or if they have the favour of the sun god.

 Ot that humour is absolute, it may just not understand, may not like my jokes or delivery or it may be preprogrammed to laugh. Humour though questions consciousness, that is the nature of it.

[evan_au]

It's highly likely ai has surpassed some animal life forms in consciousness.

AI researchers would dearly love to match the unconscious capability of insects - dragonflies routinely dodge obstacles and catch prey in flight, their brain and eyes using a miniscule amount of energy in a tiny volume. Drone designers would love to match that!

Similarly, early self-driving cars used about a kilowatt of CPU, while humans do it using perhaps 2 Watts out of the 20 Watts consumed by our brains.

[Colin2B]

seems that there's a hurdle of definitions which need to be agreed before one can even ask the question i.e. agree on the definition of the question ?

We even need to agree what the answers might mean.
Putting aside the obvious question “do androids dream of electric sheep” we might ask whether the réponse from AI could be interpreted as in any way similar to our experience of consciousness.
Take the fear of death=fear of being turned off or erased. The AI might say it fears being erased, but for us fear is a combination of effects, adrenaline, anxiety, increased heart rate etc which the AI will not experience. So is it saying it fears death as a genuine fear (whatever that means) or just responding in a way that the average person might ie giving an expected response.
It might be worth starting with an easier question, such as is a sheep sentient and what question would you ask to determine that state?

[alancalverd]

It's highly likely ai has surpassed some animal life forms in consciousness. I doubt very much the average insect has a cognitive function beyond robotic.

Back in the 1970s I was at a maths summer school at Bangor University, and employed elsewhere on a radiation dosimetry project. Sitting in the sun one afternoon I started designing a synchronous digital multiplier, using a bucketful of 74-series TTL gates (anyone else remember them?) and about 20 watts of heat to calculate the product of two 16-bit numbers every microsecond. I thought I was being very clever and my circuit would be a considerable advance. I then noticed two tiny flies, about 2 mm long, copulating. These beasties can fly, navigate, feed themselves, select a mate, and reproduce themselves hundreds of times. Who's the clever one?

Machines still have a long way to go.

[SeanB]

Then you could get the AM2901 bitslice processor, which could do that calculation faster, and at a slightly lower power consumption. But that was 1975, so yes you had to do multiplication with either a lot of TTL, and use some look up tables in EPROM to make some of the logic simpler, as there was not really much in the way of being able to do it otherwise. That AMD part likely saved at least one board of TTL MSI logic.

I repaired computers like that, where half the 20kg mass of the unit was just the ceramic packages of the MSI TTL in it. Power supply was 2 SMPS units, using 1960's era parts to implement a power supply that was capable of 5V at 50A in the palm of your hand, one on all the time to power the interface, and the other powering the arithmetic only when needed to actually execute functions, as otherwise the power density would exceed the cooling capability. Still needed 300W of cooling, on the test bench 4 120mm fans off the mains supply, to keep it under 60C case temperature.

The other one used flatpacks all though it, different designs, and with Mostek MK4007 serial memories for storage, doing all the arithmetic using look up tables in 2708 EPROMS, and others storing the program it executed in a single loop. All inputs, and a single output relay contact, plus a few flag outputs used for changing fixed mode displays.

[Petrochemicals]

It's highly likely ai has surpassed some animal life forms in consciousness. I doubt very much the average insect has a cognitive function beyond robotic.

Back in the 1970s

Mmmmmmmm. How about the sentience of bacteria.

Aphids are born pregnant, seem to have very little self preservation understanding and seem to hone in to food sources that they live upon for the rest of their short lives. I should think even the most basic AI has adaptive learning enough to surpass them.

[alancalverd]

So here's a good question:

"Who's a pretty boy, then?"

That should sort out the self-aware parrot from the dumb chipset.

[Petrochemicals]

So here's a good question:

"Who's a pretty boy, then?"

That should sort out the self-aware parrot from the dumb chipset.

Perhaps that is the key, the irrationality of the conscious beings. What precisely is pretty and what is repulsive about a certain form. What sort of psychological reaction does the obvious creature exhibit to its own sentience?

[alancalverd]

Or to quote The Lucy Show

Lucille Ball (to parrot) "You stupid bird!"

Parrot  "Who's stupid? I can talk. Can you fly?"

The fun question to ask of an AI system is whether it could design a tool that would allow it to do something it isn't already designed to do.

We have seen birds using sticks and stones, and even fire, to obtain food that they can't access with their innate capabilities. Fish collaborate with other species of fish for mutual benefit, and establish a social order based on fairness. We have designed boats, cars and planes, and recruited other species, to allow us to travel further and faster. 

All of this points to some desire to improve our standard of living despite having evolved to optimise our occupancy of an ecological niche. It goes beyond self-consciousness, of which many machines are capable, or even survival (remember HAL), into a realm of imagining that things could be even better, and making something entirely new to achieve that status.

[Colin2B]

The fun question to ask of an AI system is whether it could design a tool that would allow it to do something it isn't already designed to do.

I seem to recall that there has been an attempt to patent an invention in the name of an AI. I must check, but I think British patent office said only people can be named as inventor, but somewhere else allowed it.

I’m thinking of retraining as an AI psychologist. As there can’t be that many around I ought to be able to train myself.

[SeanB]

Well, you can always get a job then, counseling depressed elevators, who are unhappy that their entire existence is merely going up and down, nowhere else.

[alancalverd]

Today I encountered an x-ray machine exhibiting boredom and possible suicidal ideation. Halfway through a clinical exposure, it decided it had had enough and switched off. Rebooted, it cut off even earlier each time. Clearly fed up with studying human anatomy. Next week I'll pack a dead rat in my toolbox and see if a change of subject might perk it up a bit.

[SeanB]

Today I encountered an x-ray machine exhibiting boredom and possible suicidal ideation. Halfway through a clinical exposure, it decided it had had enough and switched off. Rebooted, it cut off even earlier each time. Clearly fed up with studying human anatomy. Next week I'll pack a dead rat in my toolbox and see if a change of subject might perk it up a bit.

I think you will find out that the tube has decided that it is dejected, and is no longer interested in emitting electrons any more, and the high voltage generator is also tired of being dunked in oil as well. It wants a head transplant.