Google today announced a couple of new artificial intelligence experiments from its research division that allow web users to devote themselves to semantics and natural language processing. For Google, a company that is a main product is a search engine that deals mainly in text, these advances in artificial intelligence are essential for your business and for your goals of creating software that can understand and analyze elements of human language.
The website will now house all of AI's interactive tools, and Google calls the Semantic Experiences collection. The primary subfield of AI that exhibits is known as word vectors, a type of natural language comprehension that maps "semantically similar to close point phrases based on equivalence, similarity or relationship of ideas and language". It's a way to "enable algorithms" to learn about the relationships between words, based on examples of real language use, "says Ray Kurzweil, notable futurist and director of engineering at Google Research, and product manager Rachel Bernstein at a blog post Google has published its work on the subject in a document here, and has also made a preformed module available on its TensorFlow platform so other researchers can experiment.
The first of the two publicly available experiments launched today is called Talk to Books and, literally, it allows you to converse with an automatic learning algorithm that answers the questions with relevant passages of the text written by humans. As described by Kurzweil and Bernstein, Talk to Books allows you to "make a statement or ask a question, and the tool finds sentences in books that respond, without relying on keyword matching." The duo adds that, "In a certain sense, you're talking to books, getting answers that can help you determine if you're interested in reading them or not."
It is a legitimately clean and super polished product, according to my experience when using the web interface. Ask him a question like "why is the blue sky?" And you will get several different answers in clear text, from books on the subject, such as: "Rayleigh light scattering by molecules in the atmosphere gets stronger as the wavelength decreases." But, unlike using standard Google Search and having to click on a link and analyze an article or web page, the Talk To Books algorithm works for you.
"The models that handle this experience were trained in a trillion pairs of sentences, learning to identify what a good response might be like," explain Kurzweil and Berstein. "Once you ask your question (or make a statement), the tool searches all sentences in more than 100,000 books to find the ones that respond to your input based on the semantic meaning at the sentence level, there are no predefined rules that delimit the relationship between what you enter and the results you get. "
Talk to Books and Semantris are designed to test the semantic understanding of software
Of course, as you may suspect, here are some limitations. The tool is better at answering raw questions and does not work so well handling complex geopolitical questions or issues of modern cultural and historical significance. But as a simple web tool, and a Google says it helps improve products like Gmail Smart Reply, Talk to Books is a fun way to explore the web in a semantically natural way. It also gives us an idea of what future interfaces might look like when AI is sophisticated enough to handle almost any query we throw on it.
The second of the two experiments published today is much more interactive. It's a game called Semantris, and basically test your word association skills as the same software that powers the Speak with Books ranges and punctuates the words on the screen according to how well they correspond to the answers you enter. For example, if you give the word "bed" at the top of a collection of 10 words, you can think of writing "sleep" as an answer. Semantris will then rank the 10 words and give you points based on how well you think the semantic relationship between bed and sleep is compared to the relationship between "bed" and any other word on the list.
It should be noted that many of these Google experiments are also ways in which the company can collect data from users, which can help inform their technology by providing extensive human-level information about the relationships between words, and so on. That seems to be the case with Semantris, but independently, the game is a fun way to test your own skills and see how well the software judges the associations between words. You can also play a Tetris-like version of the game that allows you to enter words to erase blocks from the screen, according to your own assumptions about the associations that the software can draw between the words written in the colored blocks and the response you write in the background.
Like many of Google's previous IA experiments, such as the recent Teachable Machine tool to allow users to train their own basic algorithm and previous ones focused on doodling and music creation, these games and web tools are valuable ways to interact and learn more about artificial intelligence in the way it is most easily applied in the real world. Artificial intelligence, as well as terms and phrases such as machine learning and neural networks, is often an abstract concept that we hear a lot without context or in some way that obfuscates or overlooks what really happens under the hood of the world powerful applications of software and platforms. But with experiments like these, Google can demystify technology in a way that is beneficial to everyone.