Does AI truly possess free will in its actions?

[rootone]

I don't think it can because whatever it wills is predicated by assumptions made by the person who coded it.

 

[jedishrfu]

This question can’t be answered with such a simple answer. AI smarts are not from its coding alone. The coding defines how the AI can learn from the data it processes. If the coding is flawed then of course the AI will not learn what it could learn.

Today’s AI systems utilize neural nets where data is provided to the inputs and an answer appears at the outputs. In effect, a matrix of data values is applied to an input vector of numbers to produce an output vector of the answer.

Where the coding gets interesting is how we determine the data values in the matrix through backpropagation to create the output vector we want from the input we give. This requires large amounts of data for training and for testing.

To see how this works, watch this video series:

 

[Filip Larsen]

How would you define free will so that it is also a measurable characteristic?

Is free will something that could be interpreted to be in the eye of the beholder, like beauty and intelligence, or is it something else? For instance, you could measure the level of intelligence of an AI via experiments using a Turing test, but can you do the same for free will?

Free will seems to relate a fair bit to predictability but also not quite. There are compelling experiments that indicate free will in humans perhaps is just an illusion and that our decision are predictable in the same way as weather and other chaotic systems are predictable for a short time. If that is true and human conscious thoughts in principle are deterministic can humans then still have free will? If yes, does that also mean the weather in some sense has free will?

 

[Borek]

I don't think it can because whatever it wills is predicated by assumptions made by the person who coded it.

At present the state of art is that we know how to design an AI for a specific task (like Alpha Go). We already don't know how it really works nor how it really learned to win, we just defined - in some general terms - input and output conditions. They proved to be good enough for the same (or at least very similar) AI to learn how to become chess master just by playing with itself. Next obvious stage is to generalize input and output conditions even further. My bet is that at some point it will shift the state of art to the point where the assumptions no longer matter.

 

[DennisN]

"Can an AI have free will?"

I don't think it can because whatever it wills is predicated by assumptions made by the person who coded it.

I have studied and also programmed AI, and I can not answer that question without getting philosophical. And I don't want to get philosophical, because that could terminate the thread . So I will just say one thing: "free will" is in itself a very tricky topic, with or without AI.

 

[Vanadium 50]

How do you know *I* have free will, and not some convincing illusion? Until that has a well-defined answer, making it more complicated by bringing computers into it is just a waste of time.

 

[jedishrfu]

Many Zen stories touch on that very fact. The master will do something spontaneous and odd, a student will respond creatively to it and the master will smile. A second student tries to outdo the first and get rebuked because he/she in essence copied and enhanced the first student's response.

Inventing is similar, sometimes an inventor comes out with a completely novel way to do something and at other times its an enhancement of a common practice used uniquely. The question is which shows free will more.