In theory, could an AI become sentient, as in learning and becoming self-aware, all from its source code?
In theory, if one could build a computing device that matched or exceeded the cognitive capabilities of a sentient being, it should be possible.
(Singlarity adherents believe we will one day be able to transfer the human mind into an artificial computing platform, and it logically follows that one could "hack" such a mind, or build from the ground up, to create a truly Artificial Intelligence.)
But this may be like fusion power, where the old adage is that it is "always 20 years away."
Sentience is another AI term without a standard definition. It can mean that the system has sensory input, that it has emotion, or that it forms a subjective opinion on what it senses. None of these require any type of non-causal or metaphysical elements, so the limits imagined for AI by those who propose that AI will never have free will or a soul don't really apply.
Whether AI has sensory input is a function of the total system design. AI already has sensory input in several products working in their field today.
Whether AI will have emotion is likely, since emotion is a more general state that affects many neurons and has some momentum (since it is based on chemical secretion, agonism, antagonism, uptake, and metabolism) and such design features will probably enter into the AI tool box.
Whether AI will form subjective opinions will depend on how deeply cognition will be realized in computers. Some believe that by adding depth to artificial networks and different interconnecting designs of them, cognition will be realized in those networks. Others (including some members here) are proponents of fuzzy logic as the best match for cognition.
Self awareness isn't the same thing as sentient although related. If computer vision is pointed at the computer and recognizes itself, then it is self-aware at a sensory level. If a computer likes itself or becomes sad because it doesn't have a robot body so it can go shopping, then that's self-awareness at an emotional and cognitive level.
Although there is no proof that these higher level things can be accomplished, there is no theoretical reason yet discovered to doubt their potential development.
Yes, an AI program can become sentient. Ray Kurzweil while giving a lecture at Singularity University on The Accelerating Future stated that human body is basically composed of approximately 23,000 little software programs called GENES. If you think about it, they are actually programs, composed of sequences of data. They are not written in C++ or Java, instead they use 3-D Protein Interaction. They evolve with time and their evolution is the reason that species are able to survive even when their surroundings experience tragic changes.
We are on the edge of a breakthrough where software will be able to do the same (evolving by themselves) efficiently. Today this is done one a basic level. Artificial Neural Network is a good example.
It is predicted that we will be able to reverse engineer human brain by 2029. Prior to this we will be able to write codes that can stimulate human brain.
AI programs can be categorized into three:
- Artificial Narrow Intelligence (ANI): This is a basic AI program that is good at good one thing. These programs are prominent nowadays. AI programs playing board games (like Chess, Reversi etc.) are example of these. They are good in only one thing.
- Artificial General Intelligence (AGI): This is level 2 AI. This will be having a IQ level equivalent of humans. It will be able to do multiple tasks efficiently just like humans. This is where a program can have understanding of it's environment just like humans. Perception, rational behavior and others will be part of this program.
- Artificial Super Intelligence (ASI): This is basically the ultimate level of AI. Average predicted date for a successful ASI is between 2045-2080. Ability of this program will be way more than that of combined intelligence of all humans on the planet. Things this program can do and think, will be beyond any (or all) human(s) to understand or comprehend.