If you combine enough open source AI libraries will it seem like a person?

Question

I listened to a talk the other day by someone who creates language robots by casting natural language to vectors, creating a robot that understands the difference between, for example, a vehicle, a colour and an abstraction like 'the' or 'it' (the example given in the talk). At another talk there was an AI that beat humans at Go. When asked if robots could be like humans the guy who did the Go-bot presentation said no, because they are focused on one task but humans are flexible. Which made me wonder...

Answer 1

Apart from assembling the modules and libraries, the problem lies within training the resulting AI model. For an AI to perform well as a "person", the current methodology may require it to be trained among humans, perhaps in a community. Calculating the fitness of the model would also require some thought, like how are we supposed to measure how "human" it is, for each iteration.

Answer 2

If "a general purpose to exist" doesn't appeal to you, then perhaps an alternate, more plain text expression would be, "What makes a person, a person"? In the entire history of human civilization, never have there been something built to be just what it is, without a clear motive / objective.

I would define humans to be the problem solvers, the programmers. The ones who identify problems and come up with algorithms to solve them. If that is the case, it comes down to the same problem:

Can a programmer, program and create a new programmer? Can one who solves problems, create a machine which solves all other problems?

Answer 3

@HatsyRei that's kind of what I was considering -- the AI would need some sort of meta-model for deciding between code libraries. I.e. does the machine like playing Go on a Sunday morning, but not a Friday night, as a simple example.

Answer 4

I think we are all aware of the technical limitations. Artificial intelligence is the study of creating machines which exhibit human intelligence. We know that the current state of AI exists for specific purposes, to solve specific problems. In order to come up with a "general AI person", humans will first need to identify their general purpose for existing, only after that can they ever hope to incorporate that general purpose into a machine / system to create "real AI". Should their be any bias or mistake in this process, the result can either be catastrophic, or simply another attempt falling short of imitating human behaviour.

Answer 5

@HatsyRei I disagree on the purpose for existing angle. Appears to me humans just are... Or alternatively, the different general purpose for existing is defined locally, or maybe globally applied to instance members...?

Answer 6

alternatively, we could just program a robot whose reason for being was to collect social media likes and that should be pretty much indistinguishable 😉

Answer 7

@MartinTaylor that answer doesn't address the details of the question. No-one is suggesting AI is 'simple'. The details of the meta-implemention and even getting the open source libraries to work with each other would no doubt present very significant challenges.

Answer 8

@HatsyRei individually the act of building appears to have a motive, but the meta-motives of those who pick building as a specific sub-motive *appear* largely to be on a per individual level and to include the irreducible sum of everything that person is, except that those which continually pick motives that cause them not to exist in a smiliar form tend not to exist in a similar form.