# Tag Info

56

Yes, indeed, neural networks are very prone to catastrophic forgetting (or interference). Currently, this problem is often ignored because neural networks are mainly trained offline (sometimes called batch training), where this problem does not often arise, and not online or incrementally, which is fundamental to the development of artificial general ...

29

To answer this question, first we need to know why developing conscious AI is hard. The main reason is that there is no mathematically or otherwise rigorous definition of consciousness. Sure you have an idea of consciousness as you experience it and we can talk about philosophical zombies but it isn’t a tangible concept that can be broken down and worked ...

19

Yes, the problem of forgetting older training examples is a characteristic of Neural Networks. I wouldn't call it a "flaw" though because it helps them be more adaptive and allows for interesting applications such as transfer learning (if a network remembered old training too well, fine tuning it to new data would be meaningless). In practice what you want ...

10

tl;dr I always like to think of Neural Networks as a generalization of logistic regression. I too don't like that, traditionally, when introducing Neural Networks, books start with biological neurons and synapses, etc. I think its more beneficial to start from statistics and linear regression, then logistic regression and then neural networks. A ...

9

I can say that among AI researchers I interact with, it far more common to view it as wild speculation than as settled fact. This is borne out by surveys of AI researchers, with 80% thinking strong forms of AI will emerge in "more than 50 years" or "never", and just a few percent thinking that such forms of AI are "near". Software Developers are not the ...

7

What is consciousness? There are some real challenges in setting up consciousness as a goal, because we don't have that much scientific understanding yet of how the brain does it or what balance there needs to be between long-term memory, short-term memory, the implicit work of interpretation, the contrasting conscious modes of automatic processing and ...

7

Utility is a fundamental to Artificial Intelligence because it is the means by which we evaluate an agent's performance in relation to a problem. To distinguish between the concept of economic utility and utility-based computing functions, the term "performance measure" is utilized. The simplest way to distinguish between a goal-based agent and a utility-...

7

Here are my suggestions Her, the AI part (movie spoiler): Ex Machina the AI part (movie spoiler): Eagle Eye, the AI part (movie spoiler): Big Hero 6, the AI part (movie spoiler):

7

An important thing we're going to need is what is called the "Expected Grad-Log-Prob Lemma here" (proof included on that page), which says that (for any $t$): $$\mathbb{E}_{\tau \sim \pi_{\theta}(\tau)} \left[ \nabla_{\theta} \log \pi_{\theta}(a_t \mid s_t) \right] = 0.$$ Taking the analytical expression of the gradient (from, for example, slide 9) as a ...

6

We have not been able to create a truly intelligent AI yet, according to your definition. So we have no real life proof-of-concept that shows that it actually works. But based on the current research, there is no known property of the human brain that cannot be modeled in software/hardware. We do not understand the human brain enough yet - and most likely ...

6

Since a neural network does iteratively learn its own weights I assume you mean the structure of the neural network - the number of layers and nodes per layer. If what I said above was your question, then yes, it most definitely is being explored. Even when a neural network is allowed to learn its own structure it still needs to be suited to a specific ...

5

Algorithms can be racist, sexist, and otherwise bigoted. When we feed them data produced by systems that are biased against groups of people, the algorithm will learn to behave that way. We're used to garbage in garbage out, now we have to worry about racism in racism out. See: Facial Recognition Is Accurate, if You’re a White Guy Rise of the Racist Robots –...

5

There are a lot of approaches you could take for this. Creating a realistic artificial analog for fear as implemented biologically in animals might be possible, but there is quite a lot involved in a real animal's fear response that would not apply in simpler AI bots available now. For instance, an animal entering a state of fear will typically use hormones ...

5

2001 (1968) HAL 9000 is a great example of an artificial general intelligence that goes astray, where the humans don't understand the reasoning process as values dis-align. (This is a nod to Asimov in the sense of humans not understanding the implications of a logical framework. Marvin Minsky was an adviser on the film.) BladeRunner (1982) The critical ...

5

What you are describing sounds like it could be a deliberate case of fine-tuning. There is a fundamental assumption that makes minibatch gradient descent work for learning problems: It is assumed that any batch or temporal window of consecutive batches forms a decent approximation of the true global gradient of the error function with respect to any ...

5

Interpreted languages allow for a faster development cycle, as they don't require time for compilation, and fragments can often be run without having a complete program. They often also have fewer constraints for variable declaration or typing. That means they can be used to quickly scope out a problem and try different solutions. The drawback is the slower ...

4

We don't even know how to prove that humans think (or not) yet. Or maybe it would be better to say that we don't really know what thinking is. In either case, there doesn't seem to be much reason to think (heh) that a "machine thinking" needs to be the same as a "human thinking". So no, I don't think it's been formally proven that a machine cannot in ...

4

I'm going to refer you to one of my favorite AI philosophers, Phillip K. Dick, who thought deeply on this subject and wrote about in some detail in Do Androids Dream of Electric Sheep. Essentially, replicants (artificial humans) had a design flaw--they lacked empathy. This flaw was allowed to persist because it had a useful side-effect in that replicants ...

4

I believe you are referring to something that Arend Hintze wrote about in his article "Understanding the four types of AI, from reactive robots to self-aware beings". Here are the four types from his article: Type I AI: Reactive machines The most basic types of AI systems are purely reactive, and have the ability neither to form memories nor to use past ...

4

Algorithms can learn to lie: SEE: Robots Evolve to Deceive (MIT Tech Review, 2007) Robots 'Evolve' the Ability to Deceive (MIT Tech Review, 2009) Evolving Robots Learn To Lie To Each Other (Popsci, 2009) Deception as a strategy has been observed in animal populations: Do Animals "Lie"? Yes, Even to Their Own Kind, Biologist Says (Rochester University, ...

4

Nice Question! This is a perennial topic of discussion among AI researchers. The short answer is "we don't really know which topics are hard in general, but we do know which we haven't got good techniques for yet." Let's start by explaining why AI is not concerned with notions of computational complexity like NP-Completeness. AI researchers figured out in ...

4

It is a difficult question to answer, as — for a start — we still don't really know what 'intelligence' means. It's a bit like Supreme Court Justice Potter Stewart declining to define 'pornography', instead stating that [...]I know it when I see it. AGI will be the same. There is no single event (almost by definition), as that's not general. OK, ...

3

Human intelligence is very general / broad in its scope. This is self-evident, and whatever AI ends up to be, we'd like it to be a general problem solver as well (cf. Simon and Newell). Taking liberal interpretations of your question... Why AI in a computer? Computers, to the extent that we can frame problems in general as a solvable computational ...

3

If your "AI" doesn't have the ability to move and perform physical manipulations in the real world then there is no way it could do something like this.

3

Judea Pearl's 2018 comment on ACM.org, in his To Build Truly Intelligent Machines, Teach Them Cause and Effect is piercing truth. All the impressive achievements of deep learning amount to just curve fitting. It may be less sensational and more technically correct to state that it is not, "Just curve fitting," but rather, "sophisticated surface fitting." ...

3

In principle, yes, what you are proposing can be done. The exact details of how to do it are an open research question. The details would also depend on exactly what your goals for the system are. If you're just trying to build some domain specific system that learns a very specific kind of knowledge, then that's probably going to be easier than building ...

3

This question has been studied academically for decades, and is really an extension of the work on Philosophy of Mind that was done in the two or three centuries before that. A good resource is Mind Design II, though it's getting a little bit old now. The modern schools of thought are: Cognitivism. This is in decline, but was extremely popular in the 70's ...

3

There is at least one very important and serious AI scientist that apparently believes in the creation of true artificial general intelligence and possibly superintelligence: Jürgen Schmidhuber, who is the co-author of the LSTM, among many other important contributions. In fact, he recently founded NNAISENSE for this ultimate purpose, that is, to build a ...

3

I'd call any system that receives input to be "perceiving", and that can constitute a basic definition of "awareness" or even "consciousness", distinct from "self-awareness" or consciousness in the human sense, which we tend to think of in terms of sentience. What the Chinese Room seems to argue is that sentience is based on qualia, which may be difficult ...

3

No, Neural Networks do not have such a guarantee. In fact, I don't believe any kind of classifier in the entire field of Machine Learning has such a guarantee, though some may be slipping my mind... For an easy counterexample, consider what happens if you have two instances with precisely identical inputs, but different output labels. If your classifier is ...

Only top voted, non community-wiki answers of a minimum length are eligible