29

Here's a snippet from an article by Gary Marcus In particular, they showed that standard deep learning nets often fall apart when confronted with common stimuli rotated in three dimensional space into unusual positions, like the top right corner of this figure, in which a schoolbus is mistaken for a snowplow: . . . Mistaking an ...


20

In theory, most neural networks can approximate any continuous function on compact subsets of $\mathbb{R}^n$, provided that the activation functions satisfy certain mild conditions. This is known as the universal approximation theorem (UAT), but that should not be called universal, given that there are a lot more discontinuous functions than continuous ones,...


18

Siri and co. are AI to some extent. The usual label is "Weak AI" (also called "narrow" or "soft" AI). It turns out the Wikipedia article on Weak AI explicitly refers to Siri: Siri is a good example of narrow intelligence. Siri operates within a limited pre-defined range, there is no genuine intelligence, no self-awareness, no life despite being a ...


18

tl;dr: None of these algorithms are practical for modern work, but they are good places to start pedagogically. You should always prefer to use Alpha-Beta pruning over bare minimax search. You should prefer to use some form of heuristic guided search if you can come up with a useful heuristic. Coming up with a useful heuristic usually requires a lot of ...


16

In our deep learning lecture, we discussed the following example (from Unmasking Clever Hans predictors and assessing what machines really learn (2019) by Lapuschkin et al.). Here the neural network learned a wrong way to identify a picture, i.E by identifying the wrong "relevant components". In the sensitivity maps next to the pictures, we can see that the ...


13

It's all about Return On Investment. If DL is "worth doing", it's not overkill. If the cost of using DL (computer cycles, storage, training time) is acceptable, and the data available to train it is plentiful, and if the marginal advantage over alternative algorithms is valuable, then DL is a win. But, as you suggest, if your problem is amenable to ...


13

Sophia uses ChatScript. You can read about what ChatScript can do here. ChatScript keeps track of conversations with each user; can record where it is in a conversational flow and what facts it has learned about a user (you have to tell it what facts to try to learn). You can optionally keep logs of the conversations (either on a ChatScript ...


12

Deep learning is powerful but it is not a superior method than bayesian. They work well in what they are designed to do: Use deep learning: Cost for computation is much cheaper than cost of sampling (e.g: natural language processing) If you have highly non-linear problem If you want to simplify feature engineering If you don't have prior distribution (e.g: ...


10

Is it because their listening function reloads in milliseconds or even nanoseconds Yes, it expects the keyword to start every moment of time and it ignores the rest. Overall, the algorithm is described here, you can read for details: https://machinelearning.apple.com/research/hey-siri


9

There are several examples. For example, one instance of using Statistical AI from my workplace is: Analyzing the behavior of the customer and their food-ordering trends, and then trying to upsell by recommending them the dishes which they might like to order/eat. This can be done through the apriori and FP-growth algorithms. We then, automated the ...


9

I agree that this is too broad, but here's a 1 sentence answer for most of them. The ones I left out (from the bottom of the chart) are very modern, and very specialized. I don't know much about them, so perhaps someone who does can improve this answer. Perceptron: Linear or logistic-like regression (and thus, classification). Feed Forward: Usually non-...


8

I would classify both as having / using elements of AI, yes. But I wouldn't say either represents a truly "intelligent" (in the AGI sense) program. But here's the rub... as you'll see in other questions asking about definitions of AI, there's a sort of memetic thing where anything that AI begins to do successfully, immediately stops being considered "AI". ...


7

So far, I have considered only three algorithms, namely, minimax, alpha-beta pruning, and Monte Carlo tree search (MCTS). Apparently, both the alpha-beta pruning and MCTS are extensions of the basic minimax algorithm. Given this context, I would recommend starting out with Minimax. Of the three algorithms, Minimax is the easiest to understand. Alpha-Beta, ...


7

Ontology learning is a relatively new field that aims to automatically (or semi-automatically) learn or create ontologies (using machine learning, text mining, knowledge representation and reasoning, information retrieval and natural language processing techniques) from some text or corpus. Ontology learning can be divided into different phases or tasks ...


6

Adaptive/predictive features are useful in at least some everyday applications. Take text messaging, for instance. All smartphone SMS apps that I know of keep track of the words you use in close proximity and use that information to predict the next word in a message you're typing. (Some are smarter than others. Relevant XKCD.) It can be used to personalize ...


6

Yes, but probably only to a limited degree in the near term. Where people draw the boundaries around 'artificial intelligence' is fuzzy, but if one takes the broad view, where it incorporates any sort of coding of explicitly cognitive functions, then many routine economic tasks can benefit from artificial intelligence. Many search engines, for example, can ...


6

Deep learning allows you to solve complex problems without necessarily being able to specify the important "features" or key input variables for the model in advance. To give an example, a problem that may be easily tackled without deep learning could be predicting the frequency and claim amounts of insurance vehicle claims, given historical claim data ...


6

One application I know of being used in industry is of image classification, by only training the last layer of one of the inception models released by Google, with the desired number of classes. I can't provide specific details. Transfer learning is useful when: You do not have the resources (time, processing power, etc.) to train a DL model from ...


5

Deep Learning these days mean a lot of things to a lot of people, its quickly becoming a buzz-word. But so far it still retains two very important conceptual properties: Does away with most feature engineering work. This was mentioned in the answer above, but this is very important. It really saves a lot of work. Allows you to make maximal use of ...


5

Nim was actually one of the first games ever played by an electronic machine. It was called the Nimatron and was displayed at the 1940 New York World's Fair. It is also well known that neural networks can model the Xor-function, if they have enough layers. Despite that, Marvin Minsky is supposed to have killed neural networks in the sixties, by asserting ...


5

This is fairly boilerplate advice, but, since you're brand new to AI, I'd personally suggest writing a classical Tic-Tac-Toe AI, ideally using minimax. I suggest this because minimax is fundamental to AI, and there are many webpages devoted to this subject, such as How to make your Tic Tac Toe game unbeatable by using the minimax algorithm and Tic Tac Toe: ...


5

I would guess no, because if the language is unknown (no data available on it), then we would not have training data with which the machine learning algorithm could learn from. If it is related to some known language, then some statistical analysis can lead to a guess at decipherment (assuming certain similarities among the two languages). If interested on ...


5

I say yes it definitely could be. But i agree with Skim you need some information as a starting point. Egyptian hieroglyphs were only (recently) understood following the discovery of the Rosetta Stone (https://en.m.wikipedia.org/wiki/Rosetta_Stone). With the same message in both known and unknown language the program could find the/a correlation. Without ...


5

Easy answer: utility. The strength and applicability of "black box" NNs has been regularly validated in the past few years, and business is concerned with results. (i.e. they don't care how the sausage is made, so long as it gets made.)


5

Using a machine learning or AI-powered model once it has been built and tested, is not directly an AI issue, it is just a development issue. As such, you won't find many machine learning tutorials that focus on this part of the work. But they do exist. In essence it is the same as integrating any other function, which might be in a third-party library: ...


5

This is more in the direction of 'what kind of problems can be solved by neural networks'. In order to train a neural network you need a large set of training data which is labelled with correct/ incorrect for the question you are interested in. So for example 'identify all pictures that have a cat on them' is very suitable for neural networks. On the other ...


5

Yes, there are many, actually. A Google search turned this paper Artificial Neural Networks in Medical Diagnosis (2011) by Al-Shayea up. Not only are they used in disease diagnosis, but even with things like prescribing medicines. In fact, the top project for a hackathon at my school analysed thousands of research articles, and took a patient's medication ...


4

I'm not sure about "emulating the brain" per-se, but in a more general sense there has been some thought given to using analog computing for AI/ML. It seems clear that analog computers do have certain advantages over digital computers. For one, they can (depending on the application) be faster, albeit at the cost of some loss of precision. But that's OK, ...


4

Chieko Asakawa (wiki, TED, IBM) is a major researcher in this area, and the linked TED talk is probably a good introduction to the state of the art as of 2015. Here's a link to a 2016 paper on a smartphone navigation system. Guide animals perform manipulation tasks as well as identification tasks, and so it's not clear if those could be replaced well at all....


4

There are many online services that use statistical neural networks for recommendations. For example, we have a well known service here in Russia that could give it's users recommendations for movies and shows to watch and books to read. Its recommendation core is based on many things known about a user: what movies/books he or she loves and what not, ...


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