<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Dan Lee on Medium]]></title>
        <description><![CDATA[Stories by Dan Lee on Medium]]></description>
        <link>https://medium.com/@danieljlee1990?source=rss-e73baa91fc71------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*ABeMqDYeH9NvPFyZ.jpg</url>
            <title>Stories by Dan Lee on Medium</title>
            <link>https://medium.com/@danieljlee1990?source=rss-e73baa91fc71------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 02:53:59 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@danieljlee1990/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[What are you looking for?]]></title>
            <link>https://medium.com/delvify/what-are-you-looking-for-94f1ff88ea7d?source=rss-e73baa91fc71------2</link>
            <guid isPermaLink="false">https://medium.com/p/94f1ff88ea7d</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[search-engines]]></category>
            <category><![CDATA[nlp]]></category>
            <dc:creator><![CDATA[Dan Lee]]></dc:creator>
            <pubDate>Mon, 27 Jul 2020 03:01:39 GMT</pubDate>
            <atom:updated>2020-07-27T03:01:39.431Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/560/1*TXD0cs5aSotPCa5U4AmERw.png" /></figure><p>Ecommerce search sounds so elementary you might not be aware of how crucial this is to your customer experience. A user types a word in your site’s search bar and like magic all your tagged images are displayed. But wait, are you using keywords to search and match products to customer queries? Why are you using a 21 year-old technology when there are much better options available?</p><p>Many studies cited by CMSwire and other publications have shown that poor search is driving customers away. You might think your search engine is free of charge, but the costs area hidden in lost customers. Over 30% of customers are turned off by annoying and fruitless searches. A whopping 70% of ecommerce search engines surveyed by Baymard couldn’t return relevant results for simple product synonyms, requiring users to type the <em>exact</em> keyword tagged to find the product. And this is not limited to smaller retailers. Up to one-third of the top 50 ecommerce sites don’t allow users to search via model number or allow brand name search queries.</p><p>With these types of stats, we know that many customers are interested in your products and are leaving because of poor CX. But, how do we know what they would do once they find the product they are looking for? Here we have some encouraging results. Salesforce found that visits where the shopper clicked a recommendation comprise just 7% of visits but drive an astounding 24% of orders and 26% of revenue. If your shoppers get to the products they want, they will buy. And this effect lasts. 37% of shoppers that clicked a recommendation during their first visit returned, compared to just 19% for shoppers that didn’t click a recommendation. Highly engaged customers are loyal.</p><p>If your user types a search query like, “I want something like that cute sparkly dress with without the stripes that I can wear to a formal event.” Your customer is likely to be shown no relevant results. You might think it is a categorization problem. But most e-tailers spend enormous amounts of time, and money which are valuable resources, to tag and retag and cross tag every possible combination of words. Yet the problems remain. It is as if you used a scooter to get across town. You can add a motor and maybe a small basket, but there are better ways to get to where you are going. This better way is called Natural Language Processing and descriptions of this can be found at delvify.io.</p><p>But if we look further at search query problems, we can turn to the Baymard Institute, which provides independent research on many web related topics recently held a large-scale study of search habits and found many sites committing the same errors. They tested many common types of behavioural searches and found problems in all of them. All brands want to provide helpful, positive search and shopping experience to improve conversions and increase units per transaction, but few do.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/436/1*C79SQH5FRpmyBZu2Ctj_sw.png" /></figure><p>We can see that the beyond the three most common search behaviours, the support fell off dramatically.</p><p>What are these searches and how can we improve the results?</p><p><strong><em>Non product</em>:</strong> Users search for help pages and shipping information.</p><p><strong><em>Exact</em></strong>: Often users copy/paste product name such as “HT-S100F <strong>2.0</strong> soundbar with Bluetooth “</p><p><strong><em>Compatibility</em>:</strong> Customers search for accessories and spare parts for products. This is related to the “Exact” search problem.</p><p><strong><em>Feature</em></strong>: Users submit searches with multiple attributes hoping to narrow their search quickly.</p><p><strong><em>Thematic</em></strong>: Users may be thinking of themes such as “party” or “outdoors”.</p><p><strong><em>Slang abbreviation and symbol</em></strong>: Users have their own linguistic quirks.</p><p><strong><em>Relational</em></strong>: Users search for similar or “looks like” products.</p><p><strong><em>Product Type</em></strong>: Users often look at narrow categories like couches or dresses.</p><p><strong><em>Symptom</em></strong>: User look for answer to question such as, “How to clean an oven.”</p><p><strong><em>Subjective</em></strong>: User have emotional queries such beautiful and quality.</p><p><strong><em>Implicit</em></strong>: (not in graph) Some users submit partial searches.</p><p>The most advanced and forward thinking clients are using AI to help their customers. Natural Language Processing (NLP) in particular has made all these problems easier to solve. In an NLP search a user types their search queries in full sentences, partial phrases or lists of attributes. Any normal search engines cannot parse these advanced queries and return results. NLP handles it with ease.</p><p>With Natural Language search, the search engine understands the semantic content of the user’s query and can return more relevant results beyond simple keyword matching. If done well, applying simple question filters much more quickly than dropdown boxes, sliders and checkboxes. NLP can also help sites match iOS and Android phone speech inputs, enabling users to</p><p>literally speak their queries aloud. NLP is an advanced technology, but once implemented, users simply type phrases and questions and can remove one of onsite’ s weaknesses compared with a physical store experience.</p><p>Search remains a hot area for advances and an important way to help your customers find what they need. Now the tools are available for optimal onsite search.</p><h3>Ready To Learn More?</h3><p><a href="https://delvify.io/contact/">Get in touch with our team</a> of experts to find out how we can help you build predictive AI solutions for your business.</p><p><a href="http://www.delvify.io/">www.delvify.io</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*kTz8YvTWv-HXz3-nMTYjWA.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=94f1ff88ea7d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/delvify/what-are-you-looking-for-94f1ff88ea7d">What are you looking for?</a> was originally published in <a href="https://medium.com/delvify">delvify</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Natural Language Processing — Where are we now?]]></title>
            <link>https://medium.com/delvify/natural-language-processing-where-are-we-now-e71680b6dbcf?source=rss-e73baa91fc71------2</link>
            <guid isPermaLink="false">https://medium.com/p/e71680b6dbcf</guid>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Dan Lee]]></dc:creator>
            <pubDate>Tue, 16 Jun 2020 05:11:00 GMT</pubDate>
            <atom:updated>2020-07-27T03:01:14.431Z</atom:updated>
            <content:encoded><![CDATA[<h3>Natural Language Processing — Where are we now?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/560/1*q4CGqIo7nB9EK_DvZnkvsQ.png" /></figure><p><strong>Introduction to the Introduction</strong></p><p>When the conversation turns to <strong>Artificial Intelligence</strong> (AI), Big Data and analytics many exciting fantasies of the future (Robots replacing people! Martian colonies!) are passed around with little thought to the myriad possibilities already available and underutilized. Artificial Intelligence is a branch of computer science and Mathematics that attempts to solve discrete problems using pre-defined rules or routines (algorithms). Leaving aside a deeper discussion of what AI may bring, the problems that AI users have already tackled range from how to read your license plate from a distance to issue a traffic violation (optical character recognition), to language translation (Natural Language Processing) and autonomously operating vehicles (perception, motion and action). One subfield that is exciting to us is Natural Language Processing (NLP). We can trace the very beginnings of NLP and AI back to the genius insights of the great Alan Turing who proposed a linguistic test, the Turing Test, to determine if computers were sufficiently “intelligent”. NLP is intimately intertwined with AI.</p><p><strong>Where are we now?</strong></p><p>Although we might be tempted to predict what the next five or ten years may hold for us, because it is exciting to make bold predictions that no one will remember in five years (Flying cars! Tang every day!), a few researchers (Joseph Mariani , Gil Francopoulo , Patrick Paroubek and Frédéric Vernier) went through roughly 55 years of papers concerned with research and development in speech and language processing to uncover patterns in the theory and the analytical tools used to tackle the many problems in NLP. Their work, <em>The NLP4NLP Corpus (I, II, III): 50 Years of Publication, Collaboration and Citation in Speech and Language Processing </em>(which in of itself is a fascinating NLP project)<em> </em>gives us estimates of the depth of this field with a study of over 65,000 documents. It is useful to look at academic papers because scientists and academics lead to the development of new ideas. Tech start-ups may develop very good and innovative uses from the academic’s work, but without the hard, academic work there is nothing to innovate from. The recent academic work will give us clues to what is going on right now in the development of NLP linked products and services.</p><p>The authors wisely identified three sub-fields to concentrate on: Speech, NLP, and Information Retrieval (IR). Much of the research concerns acoustics and signal processing as can be seen in the below graph from the same report which charts the rise in Speech Recognition (SR) as a subject over the years. We will not discuss this further, but you are welcome to ask your nearest, speech recognition device for more information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/307/1*sjHApdOyTGd_-IAUMHqt2Q.png" /></figure><p>If we instead, look at the word rankings from the same report we find some interesting insights.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/289/1*rkW88cOeLh-Vl6rnQHjFCw.png" /></figure><p>As much as the statistician in me wants to digress on the use of Gaussian Distributions and Fourier Transformations we will stick with the NLP list. We will briefly look at each term and give a few passing thoughts on uses this may indicate.</p><p>1. <strong>Semantic</strong> — At the top of the list we find the term that refers to the process of matching words or phrases that are semantically similar or related. Shoes are semantically <em>similar</em> to footwear. And shoes are semantically <em>related</em> to walking. Leaving aside talk of vectors and the advances in the contextual processing of data, this tells us that academics are very, very interested in how words are related to each other.</p><p>2. <strong>Syntactic</strong> — Syntactic analysis refers to the rules, principles and processes that govern the structure of a sentence. Roughly speaking, it is grammar. Thanks to pioneering insights of Noam Chomsky we understand how complex this really is. There is any number of algorithms developed by researchers, but the main methods involve looking at the sentence and figuring out which Parts of Speech (nouns, adverbs, etc., i.e., the morphology) each word is and rearranging these according to grammatical rules, the Syntax. If you know which phrases constitute distinct sentence units you can work bidirectionally (forward and backwards). Working forward might involve capturing the key phrases in a series of financial statements and running them through your semantic processing software to determine sentiment. Running backwards will help you determine where a word should be placed in a sentence. A Syntactic Tree (sometimes called a Parse Tree) is illustrated below and will help with understanding the next few terms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/346/0*3wjInAfkwTlQQr3s" /></figure><p>3. <strong>NP</strong> — This refers to a Noun Phrase. We can break up the English language (and many others) into two main components, Noun and Verb phrases. This is related to (2) above.</p><p>4. <strong>POS</strong> — Part of Speech Tagger is a piece of software that reads text in some language and assigns parts of speech to each word (and other token), such as noun, verb, adjective, etc.</p><p>5. <strong>Parser/Parsing</strong> — Although you may remember from your literature class, parsing poems and breaking them into their different parts, here the term is used to refer to the formal analysis by a computer of a sentence or other group of words into its different parts. This results in a “parse tree” showing the syntactic relation of each word.</p><p>6. <strong>Subset/Lexical</strong> — Any software needs to create well-ordered sets (and subsets) of data. Lexical units are the basic elements of vocabulary. It is important to help your software differentiate between plurals and singular, for example.</p><p>7. <strong>Machine Translation </strong>— The task of translating one natural language into another is extremely difficult due to the nuances in usage and continual evolution of meaning in all languages. Machine language uses many of the tools above to translate from one language to another.</p><p>8. <strong>Predicate</strong> — Predicate Logic and Discourse Representation Theory. First-order logic — also known as predicate logic and first-order predicate calculus — is a collection of formal systems used in linguistics, and computer science. Roughly stated it allows you to understand the relations between phrases and not just the declarative power inherent in a proposition.</p><p><strong>Where does this leave us in understanding the trends and focus of innovation?</strong></p><p>Firstly, the list above gives us a clearer understanding that much of the work in NLP is concerned with identifying and classifying words and phrases. This can help us focus our attention on the tasks that are involved in identifying words and phrases. Matching similar or related phrases, perhaps to find sentiment in financial reports, or to search for specific words in large reams of documents such as a medical library is where NLP excels. But we should not get carried away. Because we know a machine works better when there is a human attached, uses of machines to help us will be immensely more satisfying. For example, with a POS tagger, an editor could highlight all the adverbs in a sentence and gain some insight if a journalist was being sufficiently fair in her writing. It is not the machine that decides or asks the question, but a machine helps make this decision faster and more accurate.</p><p>Businesses have much data locked up in documents, documents that are largely mute and of such volume as to be inaccessible. NLP unlocks the value in this data. One industry likely to use these tools to great effect will be the industry that relies primarily on words, the legal profession. Inundated with reams of impenetrable documentation stretching back years, the legal profession has formed a verbal wall that has protected the inhabitants from the pressures in many industries. But that will change soon. when the financial pressure builds significantly enough, the legal profession will be transformed in a matter of months to a more outward, customer-facing businesses, putting many tools and processes at the disposal of clients, removing all the carefully constructed barriers to understanding and democratizing one of the last standing guilds.</p><p>The only question is who will be first to build this tool?</p><h3>Ready To Learn More?</h3><p><a href="https://delvify.io/contact/">Get in touch with our team</a> of experts to find out how we can help you build predictive AI solutions for your business.</p><p><a href="http://www.delvify.io/">www.delvify.io</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*kTz8YvTWv-HXz3-nMTYjWA.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e71680b6dbcf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/delvify/natural-language-processing-where-are-we-now-e71680b6dbcf">Natural Language Processing — Where are we now?</a> was originally published in <a href="https://medium.com/delvify">delvify</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GLUE]]></title>
            <link>https://medium.com/delvify/glue-f97119fa604c?source=rss-e73baa91fc71------2</link>
            <guid isPermaLink="false">https://medium.com/p/f97119fa604c</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Dan Lee]]></dc:creator>
            <pubDate>Tue, 09 Jun 2020 04:00:11 GMT</pubDate>
            <atom:updated>2020-06-09T04:00:11.358Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/560/1*Yvv2dKE36NUpmlJfgw0Fvw.png" /></figure><p>Since the beginning of time, humans and some animals have been able to communicate using several abstraction techniques. The communication exists in different forms such as gestures, languages, emotions, etc. Next to abstract mathematics, both spoken and written language is perhaps the most complicated and most interesting communication tool humans use. In the general field of Artificial Intelligence (AI) using complicated techniques such as machine learning and deep learning is a sub-field that is interested in understanding how humans communicate. This is commonly referred to as Natural Language Understanding or Natural Language Processing (NLP).</p><p>When we talk about NLP, we are concerned with a narrow subset of tasks. These tasks are typically connected with designing a computer or machine to understand human language through learning. The machine uses data such as the spoken word or written symbols that are pre-processed by statistical and mathematical models and made usable by computers. The topic under the discussion for today data and how we use and measure this data to gain insight into the efficacy of NLP models.</p><p>We believe that humans understand and communicate using tasks gained from past experience. And every experience is stored in the human brain as a complex structure. So, when a human communicates, they decode past data from their brain cells for use in the language. The field of neurolinguistic biology is enormous and beyond the scope of this post, but NLP is not the same as a brain. It is inspired by and uses similar concepts of how we believe a brain uses data, but in the end it NLP models are constructed using one of the human kind’s most powerful concepts, statistics. By using statistics to search and analyse amounts of data we can train computers to do stunningly complex tasks quickly and with fewer errors than a human. But what do we mean by “data” and how do we know the data is useful? How do we judge one computer system versus another? Luckily a few researchers, Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman have tackled the task of pre-processing the data and making it useful. This is summarized in an academic paper called, <em>GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding</em>. I am indebted to their work in the field and draw on their work for this post.</p><p>Which benchmark and why?</p><p>Benchmarks are an impassionate method to discover the relative performance of similar actions. Benchmarks are diverse and can be a stock index or the error rate in a manufacturing line. By measuring performance versus a benchmark one can identify gaps in an organization’s processes so as to improve and gain a competitive advantage. For our purposes, we will look at a benchmark that serves as a measurement of how the machine learning community including University and Industry researchers have performed on a common set of tasks. To measure the performance of several NLP tasks the GLUE benchmark is used. We will describe GLUE in the following sections.</p><p>GLUE?</p><p>The creators of the GLUE benchmark state it is “several hundred sentence pairs labelled with their entailment relations (entailment, contradiction, or neutral) in both directions and tagged with a set of linguistic phenomena involved in justifying the entailment labels.” In other words, it is a large data set designed to measure the outcomes of specific tasks. The GLUE dataset includes: i) the Corpus of Linguistic Acceptability; ii) the Stanford Sentiment Treebank; iii) the Microsoft Research Paraphrase Corpus; iv) the Semantic Textual Similarity Benchmark; v) the Quora Question Pairs; vi) the MultiNLI Matched; vii) the MultiNLI Mismatched; viii) the Question NLI; ix) the Recognizing Textual Entailment; x) the Winograd NLI; and, xi) the Diagnostics Main datasets.</p><p>What are the tasks used to test versus this data?</p><p>Researchers can access these datasets to test the accuracy of their computers on understanding specific tasks such as if the sentences are grammatically correct or if they contradict each other.</p><p>There are four primary NLP tasks that are currently tested, but the data is flexible and one can create as many task as needed in order to deepen our knowledge. However for now, the main tasks are as follows.</p><p>a) Sentence Pair Classification Task:</p><p>The GLUE data sets used for this challenge are MNLI, QQP, QNLI, STS-B, MRPC, RTE and SWAG. This task involves identifying weather the two given input sentences are a pair or not.</p><p>b) Single Sentence Classification Task:</p><p>The GLUE data sets. used for this challenge are SST-2 and CoLA. This is a classification task. Given an sentence input, the job is to find of the label of the class.</p><p>c) Question Answering Task:</p><p>The GLUE data set used for this challenge is SQUAD. This problem is a relatively difficult task as compared to the previous two tasks. This task is to input a passage of text, pose a question about the text and return an answer based on the given passage.</p><p>d) Single Sentence Tagging Task:</p><p>The GLUE data set used for this challenge is CoNLL-NER. The job of this task is to find the entity in a given sentence. The researcher would input a sentence or sentences, and the output would the classes identifying the entity. This is similar to a classification task.</p><p>There have been other attempts at benchmarks such as SentEval which also relies on a variety of existing classiﬁcation tasks involving either one or two sentences as inputs. SentEval is well-suited for evaluating general-purpose sentence representations in isolation. But something better is needed like GLUE to test cross-sentence contextualization and alignment, so we can drive new research and create a state-of-the-art performance on tasks.</p><p>What are we really testing?</p><p>There are four broad categories of phenomena we are looking at:</p><ul><li>Lexical Semantics (word meaning)</li><li>Predicate-Argument Structure (how parts are composed into a whole)</li><li>Logic (conclusions based on logic)</li><li>Knowledge (common sense word and sentence meaning)</li></ul><p>This is by no means a complete set of phenomena in a language. For example, meiosis, or the use of understatement to enhance an impression on the hearer is not tested. Remember we are testing a specific set of phenomena allowing us to create better tools, not creating a speaking android. As mentions, the creators of GLUE have set sentence pairs to each of these tasks and labelled the relationship between the pairs as entailment, contradiction, or neutral. Each of these in turn are:</p><p>Entailment: the hypothesis states something that is definitely correct about the situation or event in the premise.</p><p>Neutral: the hypothesis states something that might be correct about the situation or event in the premise.</p><p>Contradiction: the hypothesis states something that is definitely incorrect about the situation or event in the premise.</p><p>Let’s give an example of an entailment.</p><p>If I want to test entailment in lexical statement, the statement might be, “I climbed a mountain.” This naturally entails that “I can climb a mountain”. But if I say, “I want to climb a mountain”, it does not entail that “I can climb a mountain.” The computer needs to successfully identify if there is an entailment between the sentences.</p><p>How does good are human beings versus the benchmark?</p><p>Based on the current Leader Board ratings on the gluebenchmark.com website we can see the following.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/451/1*GlOnGayyFVxJ-CYX9r7lWQ.png" /></figure><p>It is important to ignore the precise numbers given versus a human being. Why? Because one should never let perfect be the enemy of good. The productivity of humans vary depending on any number of physical and emotional factors. Computers have their own challenges, but emotional factors are not one of them. Sometimes the steady performance on a task may be preferable. Further computers can execute a task on an enormous corpus of work which a human cannot do in a reasonable period of time.</p><p>So why do we not see more adoption of these models? We reach the barrier of the limits of human imagination. You may have created a tagger/ parser/ summarizer/ ASR system/ NLG module, but so what? What you do with it is what is important. Computers can not ask the question, why and what for? That is what humans are for, to pose the questions. Delvify helps you pose the right questions so your business is able to use the tools of NLP more effectively.</p><p>*Note on evaluation. The GLUE testing uses a generalized (averaging the square error from the mean value in each of 3 dimensions) version of the Matthews correlation coefficient to measure results. This measure takes into account true and false positives and negatives and is generally regarded as a balanced measure which can be used even if the classes are of very different sizes. In essence, it is a correlation coefficient between the observed and predicted binary classifications. It is better than an F Score measure because the F measure has little to say on Specificity/True Negatives and doesn’t really measure the proportion of actual negatives.</p><h3>Ready To Learn More?</h3><p><a href="https://delvify.io/contact/">Get in touch with our team</a> of experts to find out how we can help you build predictive AI solutions for your business.</p><p><a href="http://www.delvify.io/">www.delvify.io</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*btdqZ3-IQUv-nP3c.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f97119fa604c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/delvify/glue-f97119fa604c">GLUE</a> was originally published in <a href="https://medium.com/delvify">delvify</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[2020 AI Predictions]]></title>
            <link>https://medium.com/delvify/2020-ai-predictions-e3a6ce8e0324?source=rss-e73baa91fc71------2</link>
            <guid isPermaLink="false">https://medium.com/p/e3a6ce8e0324</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Dan Lee]]></dc:creator>
            <pubDate>Mon, 10 Feb 2020 09:15:53 GMT</pubDate>
            <atom:updated>2020-02-10T09:15:53.376Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0viQDNpsbfbzUIDNFVi3Sw.png" /></figure><p>We enter this decade with some very exciting momentum (and a few flops) in tech. Some of our readers may well remember Galaxy Phones bursting into flames, Google Hangouts with no friends, a confusing Windows 8 and the Facebook Phone that made the Blackberry look good. But behind all of the not-very-good-ideas there have been some real advances in technology that we are only just starting to understand. Aside from the very real advances into developing engineering solutions to physical systems, we can look to some the newer AI and machine learning solutions to see what the near future may hold in store. So with the caveat that I don’t have a crystal ball and hoping that our readers only remember our success here are a few ideas to consider for 2020.</p><h4>Death by Smart Car</h4><p>The leaps and bounds in autonomous vehicle hype have reached new bounds. We recently saw a pitch four. We might expect dark curtain manufacturers to have a boom as these are developed. However, given the real problems in sensor technology and the amount of test data available for regulators to parse, we think that the next death of someone hit by an autonomous vehicle will lead to a big reversal of momentum as the public grows wary the regulators come under pressure to produce cautiously. Please don’t write to us about how much safer autonomous vehicles may be. We are dealing with human emotion.</p><h4>Deep Fakes Funnel Money to Disney</h4><p>The recent advances in neural network architecture and the superb GPUs like NVIDIA GTX 1080, have meant relatively fast training to create good-enough deep fakes. One only need to watch the most recent Star Wars movie to see this in action. The hardware and architecture will continue to evolve as the economic demand increases resulting in the realization by many executives that they can create perfect actors and actresses who never age and never need to be paid a dime. Added bonus — no scandals! This will lead to competition among the content creators for ever-greater access to deep fakes. Will there be bad actors? As long as there are human beings this is a certainty.</p><h4>Facebook Failure (again)</h4><p>As the audience for Facebook continues to age, the pressure for Facebook and other social media plays will be to diversify overseas and to reach for younger customers. Younger consumers have mostly decided that social media is best consumed as either a passive activity such as watching silly cat videos or a community event such as playing a video game. No more curated home pages. Video games are also taking advantage of the newer GPUs and the advances in personalized content and faster</p><h4>Assistants will be Fired. Then rehired.</h4><p>The hype around chatbots and virtual assistants is charging ahead due to the advances in NLP. We are big fans of BERT in particular, but really any of the newer architectures are fantastic. But the reality of what can be accomplished is yet to be clearly defined in new products. The use of NLP will not be like conversing with a human being. It is doubtful machines will every learn to converse. As disappointment sets in the use of assistants and chatbots will stagnate. Until, miraculously, the chatbots will be rehired to do the one function they are good at — directing your query to the appropriate channel. NLP chatbots are excellent traffic cops, directing your queries to where they can be most efficiently answered.</p><h4>Lives Saved</h4><p>The stealth (well for most of the world) application of AI has been in visual matching. Despite the attempts to use visual matching to help assist armies and to deliver better shopping experiences, the real stunning discovery has been the ability of AI to take a relatively homogenous set of images (such as X-rays of lungs) and find the minor discrepancies. This has led to stunning advances in the detection of disease in human beings. We believe there will be greater and louder announcements of lives saved due to AI.</p><h4>China will help Advance the World</h4><p>Chinas use of AI in security and biometric identification has but many people on notice. Because of China’s focus on safety and its harnessing of huge computing power to look through large data sets to identify what is undesirable, brands and will begin to demand greater safety of ad delivery. The mantra of if-China-can-do-it will lead to greater calls for detection of “deep fakes” and the use of AI to discover counterfeit products and fake reviews online.</p><h4>Summary</h4><p>Robots are probably not going to take over the world this year, but there are some exciting trends and developments in the AI.</p><h3>Ready To Learn More?</h3><p><a href="https://delvify.io/contact/">Get in touch with our team</a> of experts to find out how we can help you build predictive AI solutions for your business.</p><p><a href="http://www.delvify.io/">www.delvify.io</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*kTz8YvTWv-HXz3-nMTYjWA.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e3a6ce8e0324" width="1" height="1" alt=""><hr><p><a href="https://medium.com/delvify/2020-ai-predictions-e3a6ce8e0324">2020 AI Predictions</a> was originally published in <a href="https://medium.com/delvify">delvify</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>