Gadget Blog

Tweetear

Mobile devices, including tablet computers and smartphones are rapidly becoming the computing and communications device platform of choice for consumers and business users. Messaging, gaming, engaging friends on social media, web surfing, reading e-books, and making payments (m-commerce) are just a few of the activities that consumers are doing on their mobile devices, for work and leisure. Just as the world wide web transformed travel distribution forever back in the late 1990s by enabling consumer-direct shopping and booking, mobile apps too will again alter how consumers plan their trips to achieve their travel dreams and experiences, and shop for and book travel.

Mobility devices

Home PC purchases and usage peaked in 2008 and have been on the decline ever since. And the companies that missed out on the mobile computing wave such as Dell and Intel are now feeling the pain. It is more likely to find a traveller in a café or at home on their couch using their Apple iPad, connected to the internet via Wi-Fi, dreaming about their next trip and researching new destinations and the sites and activitiesthat are available once they get there. It took 10 years for smartphones to reach 40 million unit sales, but tablet computers only required 4 years to reach the same level of unit sales. Analysts predict tablet unit sales will reach 100 million by 2015, so the race is now on to build the next generation of travel apps for mobile devices.

The mobile entrants are both large incumbents and new upstarts. Expedia launched their newly redesigned mobile app for iPhone at the 2012 travel industry expo PhoCusWright in Scottsdale, AZ. Critics hailed it as the best new mobile app for booking travel from any of the large OTAs. Also highlighted there was WeHostels, a Mobile Travel Agency (MTA) app that enables youth travellers to find a hostel to lodge for the night that is within their price range, has availability and is located via their mobile phone GPS geo-location feature. WeHostels is 100 per cent mobile -- no call centers, not even a web site -- WeHostels is the perfect offering for the wired up, young traveller on the go seeking shelter! HotelTonight is another immensely popular mobile phone app for the adult traveller who may find themselves stranded away from home overnight on business, or wanting to stay an extra day on a leisure trip for that one more day of fun, and needs a place to stay -- find it and book it in a few easy steps right on your smartphone!

According to research from Expedia and comScore, 58 per cent of mobile users have already booked air travel on a mobile device, and 56 per cent of mobile users have booked a hotel. By the end of next year, mobile bookings will be higher than “classic” desktop bookings, and not only for last minute reservations. But what about before making the booking? Travel begins with an experience in mind that the consumer wants to create with whom, when and where they travel. Two thirds of mobile users surveyed spend time dreaming about travel at least once per month. Forty eight per cent of travellers use their mobile device to research their dream vacations to far flung destinations. The majority of travellers, 49 per cent in developed markets and 66 per cent in emerging markets, are both undecided and budget-focused when they start planning discretionary travel. Seventy eight per cent of travellers planning a quick get away don’t know where they even want to go! Fifty per cent of travellers surveyed refuse to stay in a hotel that has no favorable reviews.

Travellers are first and foremost looking for credible sources of information and curated content (photos, videos, video tours, ratings, reviews and recommendations) from trusted advisors, including family and friends, travel bloggers and experienced travel agents.

Mobile Vacation Planner

Mobile Vacation Planner (MVP) is a tablet computer-based app by Blue Star Infotech (initially available on the iPad), that will enable travellers to plan their itinerary for a trip or vacation to any destination of their choice. MVP enables consumers to search a wide range of curated content and enjoy a highly-interactive user experience for travel planning using destination photos, videos, maps, and detailed descriptions of hotels, activities, site-seeing opportunities, and ratings, reviews and recommendations of travel content.

Mobile Vacation Planner (MVP) expedites travel planning and makes the process an enjoyable and hassle-free experience.

From the pre-booking stage to pre-travel itinerary management, as well as intra-trip and post-travel sharing, MVP addresses all of the aspects of planning a trip and telling your friends about it afterward. Simply enter the date and the travel destination, along with your desired activity timing and preferences, and MVP will offer a variety of suggestions for hotels, dining, cultural events, and activities suitable for different types of trips, including family, adventure or romantic getaways. MVP also offers an automated itinerary generation feature complete and the freedom to manually select activities to plan, re-plan, add, rearrange, and delete activities until the customer arrives at his or her perfect vacation travel plan.

Mobile Vacation Planner (MVP) is designed to be highly interactive and travel content-rich. MVP includes features for seasonal, context & time-sensitive, geo-location-based, pre-and intra-trip planning and post-trip sharing, and offers APIs to integrate with Google Maps & Places, TripAdvisor, Bing Attractions, a  Travel Planning Calendar, including export  to Apple iCal. MVP also offers Travel Booking Engine & Commerce Integration, as well as  the capability to integrate third party travel  content sources. MVP also includes the capability to upload trip details to Social Media, like Facebook, Twitter, and instagram, and the  facility to use the app both online and offline, has enhanced user experience, flexibility, and  ease of use.

Private labelling of MVP is available to companies seeking to engage and add more value to the consumer in their experiential travel planning, including tour operators, traditional & online travel agencies, youth and adult travel management companies, hotels, casinos & resorts, vacation clubs, event management companies, and convention & visitors bureaus. The business benefits of Mobile Vacation Planner (MVP) include powerful consumer bonding with your brand and increased customer loyalty, increased long-term revenues, increased travel agent productivity by offloading some of the travel planning activity to the consumer, and reduced computing infrastructure costs by leveraging both open source technologies and cloud computing-based hosting services.

Blue Star Infotech also offers other traveller value-added mobile apps including two for mapping routes to destination locations and providing turn-by-turn directions, one at street level (iRoadGenie) and the other inside large facilities for large resorts, convention centers, casinos and other venues with navigationalchallenges (iMapGenie). The company has also developed mobile native and web apps for booking air, car and hotel services, travel itinerary management, and weather and flight delay information.

The new age of mobile travel apps that increase traveller productivity, enhance their travel and user experiences, and help plan and  realize their dream vacations is upon us, and the forecast looks bright!

Tweetear

Natural Language Processing lies at the  intersection of linguistics  and computer science,  and promises to change  the way we interact with  computers. Forever.

What is Natural Language Processing?

Natural language, or the language spoken by humans the world over, is  a fascinating construct. Not only does  each language have its own set of rules  concerning grammar and semantics,  but over time these languages spawn  dialects that modify these rules in subtle  ways. Computers, on the other hand are  programmed using a set of instructions that are collectively called a programming  language. These languages have their  own ‘grammar’ (called syntax) and their  own ‘vocabulary’ (called keywords) and  bear little or no resemblance with natural  language. While these programming  languages operate on a very strict set of  rules governed by logic, spoken language  is often far more fexible, and any attempt  to quantify these rules would fall under  the domain of ‘fuzzy logic’.

Natural Language Processing (NLP)  is an attempt to use the concepts of  linguistics and machine learning to  allow computers to understand and  communicate using natural languages.

The earliest attempts at getting computers to understand human languages  were in the 1950s, where machines were  programmed to translate texts between  different languages using a hard-coded  set of rules. However, such attempts failed quickly owing to the imprecise and ever-changing nature of language. Recent developments in the felds of machine  learning using artifcial neural networks  have used statistical models that ‘train’ the software to understand the subtleties  of language. Such systems can interpret  natural languages more accurately, owing  to their probabilistic approach, where  they are trained using a set of known  rules or words, and use this data to inter- pret new, raw data.

This form of NLP that uses machine  learning models has proven far more  successful than earlier approaches which  modelled themselves after computer  languages rather than human languages.

The problem in this approach was that  programming languages have fewer rules  which are enforced more strictly than  natural languages, which follow complex  and ever changing rules of grammar.

Language-specifc semantics is another  hurdle, as the literal meaning of words  may not always be accurate.

Machine learning models the problem from a human perspective. Humans tend  to learn a new language faster using  examples as opposed to strict rules, and  NLP algorithms try to use this very fact to  improve the accuracy of their systems.

Algorithms used in Natural Language Processing:

The Markov Model and Parts of  Speech

Most algorithms that classify text use  complex learning algorithms that involve  statistical models to extract meaning from text.  An older, less accurate but considerably simpler technique used is the Markov  model, which is used in classifers such as  Parts Of Speech (POS)  tagging or sentiment analysis. POS is a system which tries  to assign a tag to each word in a sentence,  identify it as an adjective, verb, noun,  article, singular, plural etc. Sentiment  analysis is a technique that looks for certain words or ‘identifers’ that determine  the overall ‘mood’ or sentiment of the sentence. Such analysis is useful analyzing  social media to gauge group sentiment.

One company putting sentiment  analysis to good use on social media data  is Paraktweet, which has launched two  products. The frst is called Bookvibe  which gives book recommendations based  on positive tweets about a tome, and the  second is TrendFinder which helps enter- tainment companies gauge the audience’s  response to a movie by following the con- versations around a movie or TV series.

Parsing systems have also been developed that analyse online reviews on sites like  Amazon to gauge the overall response to a  product, helping prospective buyers make  an informed choice.

Markov models are a class of probabilistic learning algorithms, where the  sentence is broken down into its individual constituent words, and represented  in the form of a tree or a chain. The system  contains a corpus of pre-classifed words which are used to perform a preliminary POS classifcation. However, one  word may belong to more than one part  of speech, and the classifcation of such  a word can never be completely unambiguous. For e.g. the word ‘train’ can be  used as a noun (a local train) or a verb (to train as a professional athlete). Which of the two is correct in a particular context, is  determined by using probabilities that are  assigned on the words based on historical  data. The words that immediately follow  or precede an ambiguous word may also  be analysed to assign such probabilities.

By repeating this process, the system  ‘learns’ over time; and it’s known as  Machine Learning.

One excellent application of a Markov  chain is in text-prediction, specifcally in  predictive keyboards in smartphones such as the Android 4.1+ keyboard, or Swiftkey. 

These employ a similar Markov model,  where the next word is predicted based on  a probabilistic model, taking into account  previously typed pairs of words as well  as a corpus of word-pairs found in the  training data for accurately predicting the next word that the user might type. Recent  updates to Swiftkey have even allowed  for the prediction of whole sentences  by extrapolating the same algorithm to  groups of words. Swiftkey’s predictions  tend to be eerily accurate, and the app literally ‘gets to know you’ with time.

Swiftkey learns your typing habits over time, which can get slightly... embarrassing

A modifcation of a  Markov chain is called  Tree-traversal, where indi- vidual tokens (letters in a word, words in a sentence  or sentences in a block  of text) act as nodes of a  tree, and each parent node  has a child node for each  allowed permutation. As  an example, a spell-checker  would work by traversing  a tree with 26 parent nodes  (one for each letter of the  alphabet) and upto 26 child  nodes, who in turn will  have at most 26 children  of their own. Each level  of the tree represents the next letter in a  word (so a four letter word would be four  levels long), and the algorithm traverses  the nodes as the letters are entered in the  text document. The number of levels in  the tree would be the size of the biggest  word in the language. Every time the user  enters a combination of letters that is not  a valid path in the tree, the program will fag the word as being incorrectly spelled.

Suggesting an alternate spelling is far more diffcult though, as there is no way  of knowing which letter of the word has  been incorrectly entered, and will require  traversing each letter in the word-tree,  trying to fnd a word that has a spelling  similar to the mistyped word.

LexRank and Automatic Summarization

Thanks to the ubiquity of the Web, people  are fnding themselves at the receiving end of a frehose of new information  everyday, from  email and social  media updates to  news stories or  feeds from blogs  and websites. It is  practically impos- sible to consume and digest all of  this information  in a sane manner,  which has given rise to the need for automatic summarization of texts.

Summly provided summarized versions of news stories in a gorgeous interface. No wonder Yahoo bought it!

Popular iPhone app Summly provided  summaries of top news stories in an elegant, well designed interface until it was  acquired by Yahoo, making the 17 year  old prodigy behind the app richer by $30  million.  Summify, a similar service that  aggregated content from all your social  feeds and delivered it in via email, web or  email was acquired by Twitter last year.

Not to be left behind, Google acquired  Wavii, a startup that offered a personalized, summarized news feed to users by combining stories from social networks and other web content.

One of the most popular algorithms  for performing automatic summariza-tion is the LexRank algorithm which  gets its name from Google PageRank that powers its searchengine. It works by  using a graph-like structure to represent  the sentences and their relationships  with each other. Each node of the graph is represented by a sentence, and the edges  connecting these nodes indicate the similarity between the sentences. The shorter  this edge, (or the closer two nodes are) the  more similar in meaning they are. The  similarity of the sentences is calculated by measuring the frequency of the words occurring in the sentences.

This leads  to low frequency words being assigned  a higher priority in the summary. The  importance of a sentence is determined by the relative importance of the adjacent sentences, similar to how PageRank  calculates the relative importance of a web page. A similarity matrix is then con-structed using this LexRank data, and all sentences with a score below a threshold  value are rejected.

Speech recognition –  Apple’s Siri and Google’s  voice search

While the process of recognising spoken  word and transcribing doesn’t exactly  fall under NLP, training computer systems to respond to the transcribed text is a very exciting application. The process of extracting meaningful and unambiguous text from an audio clip is done using an algorithm similar to the Hidden Markov Models described before.

The audio clip is sent through a Fourier Transform, which splits up the stream  into different frequencies (similar to an  equalizer), and a correlation algorithm  forms an association between the shape  of the curveand the (known) word being  spoken. A corpus of such training data  is formed, which allows the system to assign probabilities to each word spoken by the user. Based on these probabilities,  a Markov chain is constructed which gives a reasonable transcription of the  spoken word into text.

Once the text has been extracted,  virtual assistants like Siri, Google Voice
Search and Siri-like alternatives on  Android such as Iris, Vlingo and Robin  parse the text and break it into parts of speech. The second part of the task  involves extracting useful information  from the text, such as the nature of the query and the best way to respond to it.

These systems use a pre-indexed database  that is used for replying to queries, which can include both offine data as well as online search engines to deliver the most relevant response.

Another age-old problem that is seeing a new lease of life is that of machine translation. Google Translate is a great example of applying the modern techniques of machine learning to automatic translations, as earlier attempts based on hard-coded rules failed while translating beyond simple phrases. This is still an open problem, however, as  translating large chunks of texts (such as  scientifc papers, journals or novels) is far  from perfect.

Google debuted its updated Voice Search in Android 4.1 Jelly Bean

Not to be dissuaded however, researchers at Carnegie  Mellon University have come  up with a unique crowd-sourcing model of translating  documents on the Internet  using an app called Duolingo.

The website (along with mobile  apps for iOS and Android)  gamifes the process of learning  a new language, where users  are rewarded with points for  completing new lessons. The  service, which is completely  free of cost, tests these newly  acquired language skills of its users by asking them to translate documents from/to  the language they are learning.

A statistical model is employed  that correlates the translated  documents submitted by thou- sands of its users to weed out  any mistranslations.

Facebook’s newest feature, Graph Search employs Markov  Models to break down the search string into a hierarchy  of components that are then  used as commands to query  the database. For e.g., “Friends  who live in Mumbai” will look for a union  of all users who satisfy both conditions  simultaneously. Permutations of this  string such as “Friends in Mumbai” or  “My Mumbai friends” also point to the  same dataset, as the parsing engine is  intelligent enough to distinguish between  the words in the string that are essential  (friends, Mumbai) versus those that are non-essential (living, my). This allows for  highly specifc searches using data your  friends have entered.

As statistical models for processing  natural language data get more effective, the divide between programming  languages and natural languages seems  to be closing. Data scientists are fnding  new ways to parse human languages into  tokens of data, leading to new high level  programming languages being developed that lie very close to to spoken languages on the scale, making programming accessible to a wider audience. Intelligent voice- activated assistants like Siri are indicative of the massive progress that NLP has  made over the years, but localising these  systems into multiple languages, each  with their own idiosyncrasies is another  challenge that engineers are working hard to tackle. 

Apple’s Siri employs highly sophisticated classifers to parse speech input

Facebook’s Graph Search uses NLP to fnd the most relevant search results

Tweetear

Agent 001 goes on a mission to uncover the truth  behind brain training. Is it  truly effective?

I landed myself a rather cushy task this month which involved  spending most of my time in the  offce with a steaming mug of coffee  rather than hunting for a very specifc  motherboard on Lamington Road in a heavy downpour. It all started when my Digit friends came across this brain  training website/app called Lumosity.

Lumosity’s gamifed training program  promises a substantial boost in your  cognitive and reasoning abilities, in  other words it makes you smarter and  more intelligent. These claims kicked off  a huge debate regarding the effcacy of  such brain training programs in actually enhancing your brain’s abilities.

Now, a lot of research is going on in this  particular direction with some of the  world’s best psychologists and neuroscientists working to fnd conclusive  evidence backed by statistics. But unlike  peer reviewed research papers we just  took on to test these claims ourselves,  we decided to use this service and see for ourselves if it enhanced our brains  capabilities. I was to use Lumosity for  a month and get a frst hand encounter  with serious brain training. The next  step was determining how would we  measure my progress – using Lumos-ity’s own numbers didn’t make much  sense here. Hence, I decided to settle on a  standardized battery of tests created by the Medical Research Council at Cambridge University. The results would  give me my position on a bell curve of the  population’s performance in the areas of  memory, reasoning and verbal ability. I  would take the test again after a month  and measure my progress and verify  the validity of claims discussed earlier.

A little background information on  Lumosity before proceeding further: It  was co-founded by one Michael Scanlon,  who abandoned his PhD in neuroscience at Stanford to convert his research  into a fedgling business. Lumosity had  launched in 2005, it has grown 150% year-on-year and as of April 2013 it has  40 million members. Lumosity’s iOS  app has been downloaded more than 10  million times with the app frequently  fnding it’s way to the top positions in its  category. The science behind Lumosity  is what they call ‘neuroplasticity’, which  is defned as the brain’s ability to physically reshape itself when faced with new,  challenging problems. Lumosity also  carries out research by collaborating  with researchers from prestigious universities across the world, which  it calls the Human Cognition Project.

Lumosity

Lumosity’s website and app allow these researchers to conduct experiments over the Internet obviating the need to recruit participants and arranging lab visits.

I took my initial Cambridge Brain Sciences test on 25th of May, it turned out  that I wasn’t as intelligent as I thought.

I was in the 34th percentile for short-term memory, 71st percentile for  reasoning ability and the 46th percentile  for verbal intelligence. My regime was to train with Lumosity in the morning after  or accompanied by my morning cuppa everyday for one month. I would take the Cambridge Brain Science IQ challenge once again at the end of the month to see if I’ve actually made progress or not. Another thing to add here is that I used the free version of the program, which according to Lumosity was not giving me the beneft of a comprehensive training program. The paid subscription service would have given me personalized training,access to all the games that

Lumosity has to offer, anytime and compare my scores with others. A monthly subscription costs $14.95 per month, which according to the current exchange would convert to `887 per month. The charge comes down to $6.7, which is equivalent to `397 per month if you commit for a year or $4.99 that is about `295 per month if you commit for two years. There is an option to purchase the subscription for a lifetime that costs $299.95 that comes to around `17,779.

The app’s interface is incredibly simple and dead easy to follow when you register for the program you  are greeted with a wizard which helps  you build your personalized training  program. The areas where Lumosity  focuses are memory, attention, speed,  fexibility and training. Lumosity offers  different games to improve specifc  aspects in these areas viz., for memory, the aspects on which training is offered are  recalling the location of objects, remembering names after the frst introduction, learning new subjects quickly and  accurately and keeping track of several  ideas at the same time. I wanted to check all the tests that they had to offer thus

I selected every aspect of focus areas  on offer for my personal training.

After a day of training on Lumosity

Alternative Brain training programs/Games

Brain Spade Games
www.brainspade.com
AARP Brain Games
games.aarp.org
Brain Metrix
www.brainmetrix.com
Brain Age Games
www.freebrainagegames.com
Mind Games
www.mindgames.com/brain-games.php
Miniclip
http://www.miniclip.com/games/
genre-476/brain-training/en/

Speed Match is good to sharpen the brain

The games themselves are incredibly polished and well designed. I played ‘Speed Match’, which improves my speed  and memory. I played ‘Eagle Eye’ and  ‘Lost in Migration’ which worked on my attention skills. I played ‘Speed Match’  and ‘Memory Map’ enhancing my short- term memory and ‘Word Bubbles’ and  ‘Rain Drops’ for fexibility and problem  solving skills respectively. In the free  version you can play these games only  once per day. The games are not acces-sible after you’ve completed your daily  dose of brain boosters. Another thing to  note is that these games are assigned to  you on a random basis, thus everyday  is like a lottery. Lumosity’s own scoring  index is called BPI (Brain Power Index),  which it derives by comparing your  game scores with results from other  players, and assigning you an index  based on your relative performance.

Now talking about the results, I tried  to stick to the regime that I came up with,  but quite frankly I was not able to do it  everyday. I stuck to the schedule for the frst two weeks but then the practice session kept on becoming more sporadic. I  found the games slightly juvenile - stars  dancing around the screen, submarines,  monsters and penguins, which made it  quite boring to trudge through the games  everday. Though this is what I felt, and  other people who are playing these games  might not necessarily feel the same. I took  the fnal Cambridge Brain Sciences test  at the same time and same conditions as  the frst one, so there is no discrepancy in  the result. I scored in the 38th percentile  in short-term memory, 78th percentile  in reasoning ability and 49th percentile  in verbal ability, a modicum of progress  in all areas. According to Lumosity itself 

I have been able to bump my BPI from  485 to 672 in the span of one month, with  the largest gains in memory, speed and problem solving. The free version does not  let you see your historical performance  and the detailed breakdown of your  performance, but if you keep the track of  your scores then you can make out your  strengths and weaknesses easily enough.

The most important question, is it  actually effective? I am now supposedly more intelligent than I started out,  but the skeptic in me refuses to believe.

Does intelligence in its truest sense only  mean getting better in these sections? I  couldn’t actually see my performance  on each and every game that I played  but I could make out from the general  progress in scores that the more I played  a particular game, the more I scored.

I played these games repeatedly  for quite some time and the final test  panned out in the same vein, and thus
I believe that I’ve gotten better only at  these particular games and tests. None- theless, it does feel like I have achieved  something even if it’s a few notches  higher in the BPI or hitting a higher  percentile on the tests. Not necessarily  boosting your IQ and making you a  genius, these games in my opinion are  just a form of mental exercise. They just  keep your brain in shape like any other  muscle in the body, the choice of training  is entirely your. It can be Sudoku or a  crossword in your daily paper, riddles  and puzzle, math-based exercises or even brain games such as these.

Tweetear

The fear that our devices are somehow altering our brains might seem exclusively modern. But in 1931, Technology Review published  “Machine-Made Minds: the Psychological effects of Modern technology,” in which John Bakeless explored how machines had transformed the very nature of human thought. here’s what he had to say:

It is a curious fact that the writers who have dealt with the social, economic, and political effects of the machine have neglected the most important efect of all—its profound infuence on the modern mind. Anything that shapes our thoughts shapes society also; and the efects of the machine on contemporary thought must, therefore, be at least as signifcant as its effects on contemporary economics or industry, or the life of society in general.



Even our republican form of government is possible only because a few machines—mainly vehicles (railroads, airplanes, and motor cars) and means of communication (mails, telephone, tele- graph, radio, wireless, and machine-made  newspapers)—bring the minds of a continent sufciently close so that we can live  and work together. In fact, if we may trust Shakespeare, who certainly was not a product of the Machine Age, “there is nothing  either good or bad, but thinking makes  it so.” If the machine really controls our thoughts, no wonder it controls all else.

Consider the mental equipment of  the average modern man. Most of the raw  material of his thought enters his mind  by way of a machine of some kind—often  through the agency of several machines.

Newspapers, magazines, moving and talking pictures are the clearest examples.

All this creates an almost incalculable  diference between the modern mind—the  scholar’s in his study, the technologist’s in  his laboratory, the engineer’s in the feld,  as well as the giggling, gum-chewing shop- girl’s on her way down town in the sub- way—and the mind of the Eighteenth or  early Nineteenth Centuries. For the frst  time, thanks to machinery, such a thing as  a world-wide public opinion is possible.

Quite as significant as the machine- made power of the press and of mechanically reproduced art upon our minds, are  the various mechanical devices developed  during the last two decades for pouring ideas into our eyes and ears—movies,  talkies, radio, and television. Some of these  mechanical devices probably have more  efect upon the less literate levels of modern society than the printed word could ever  hope to have.

The danger is that our minds may be tied down to the machine. Our art may  some day be restricted (as advertising art  always has been) to that capable of mechanical reproduction, our music to the requirements of radio, talkie, and phonograph ...

All because we have misused the machine,  or allowed it to misuse us.

If the world ever realizes that hitherto  Utopian vision of a general difusion of the  good things of life—an ample assurance of  food, clothing, and shelter for everyone, to  which is added leisure for art, letters, pure  science, and philosophy, the gorgeous play- things of the mind—it will have to look for  them to the machine.

That is, it will have to  look to the machine for the economic basis  on which these things must inevitably rest. Strangely enough, we have hitherto  been willing to enslave ourselves to the  machine instead of enslaving it. Most of our contemporary troubles arise from that  odd willingness to allow the machine to be  master instead of slave. If we are to build  a great civilization in America, if we are to  win leisure for cultivating the choice things  of the mind and spirit, we must put the  machine in its place.

Tweetear

A venture capital frm throws out intuition and uses computer models to determine investments.

Aldea Pharmaceuticals, a startup developing an emergency treatment for alcohol poisoning, seemed like an attractive investment to venture capitalist David Coats. But he didn’t rely on a hunch—he consulted the computer model he’d built.

Wenjin yang is research vice president at aldea pharmaceuticals, which got funding thanks to software suggesting that its method for speeding up alcohol metabolism was a good investment.

Two weeks and a few phone calls later, he cut the company a $1.25 million check. “A decision like that would have normally taken a minimum of three months,” says Tim Shannon, who is Aldea’s CEO and a partner with the frm that had led Aldea’s $7 million fund-raising round.

The $1.25 million was a follow-on investment from Correlation Ventures, which calls itself a “new breed of venture capital firm”—one driven by predictive analytics software built over the last six years by founder Coats and his partner Trevor Kienzle.

Correlation Ventures asks startups to submit fve basic planning, fnancial, and legal documents. It enters these into a program similar to credit rating software.

Entrepreneurs with low scores can get their rejections in as little time as two days.
High scores lead Correlation to a 30-  minute interview with both the startup CEO and the outside venture frm leading the investment, plus a quick legal review and background check.

Once it makes an investment, Correlation backs of and doesn’t take a board seat.
That policy is itself data driven: the frm’s analytics show that companies with more than two VCs on the board are less likely to be successful.

What’s not yet clear is whether this system works. Correlation Ventures has so far invested in 26 companies in diverse sectors but says it is too early to grade its success.

None of this might have been possible a decade ago. Harvard Business School professor Matthew Rhodes-Kropf, who advises

Correlation Ventures and is an investor in the fund, says the venture capital industry has only recently worked through enough business cycles to look for subtle trends.

There was also no complete, accurate, public set of venture capital data, so Cor-relation Ventures hustled for it. To build and maintain its database, it partnered with Dow Jones, scoured the Internet, signed nondisclosure agreements with more than 20 venture funds to see their internal statistics, and called hundreds of companies.

While so-called Big Data companies have attracted plenty of investors, the reputation-  driven venture capital industry itself has yet to embrace their tools. (There are exceptions, such as Google Ventures, which uses quantitative analysis to help guide decisions.)

One fnding from Coats’s research is that while top-tier frms invest in a disproportionate share of “winning” companies, the majority of successful investments are led by venture frms that don’t even crack the top 50. So it makes logical sense for Cor-relation Ventures to focus equal time and energy on many companies and co-invest with a diverse set of venture capital frms, he says.

To explain his project, Coats cites Money  ball, the book and movie about how Oakland Athletics general manager Billy Beane rejected the conventional wisdom on evaluating baseball players and built a winning franchise by letting a computer tease out variables that others overlooked. He believes the averages will work out. “We’re not claiming to have a magic crystal ball,” he says. “We’re tilting the odds a little in our favor with each investment.”

Tweetear

Successful businesses will be those that optimize the mix of  humans, robots, and algorithms.

In his new book Automate This, Christopher Steiner tells the story of stockbroker Thomas Peterfy, the creator of the frst automated Wall Street trading system. Using a computer to execute trades, without humans entering them manually on a keyboard, was controversial in 1987—so controversial that Nasdaq pressured him to unplug from its network. Then, with a wink, Peterfy built an automated machine that could tap out the trades on a traditional keyboard—technically obeying Nasdaq rules. Peterfy made $25 million in 1987 and is now a billionaire. Today, automated trading bots account for nearly three-quarters of U.S. equity trading by volume. Trading houses plow millions into fber optics and microwave dishes so their algorithms can send trades a millisecond faster than the next guys’. And although the frst trading robot was built 25  years ago, most of the change on Wall Street  has occurred during just the last few years.

Robots made by Kiva Systems move product shelves on a warehouse foor. Amazon bought the company earlier this year in a step toward automating its distribution system and reducing labor costs.

When it comes to automation, we may be in  the elbow of an exponential curve.

In this business report, we look at this  cutting edge of automation. Consider Amazon. The company not only automated  book buying but also turned the computer  systems it built to do so into a service called  Amazon Web Services, making them avail- able for anyone wanting to repeat the feat.

And now Amazon’s founder, Jef Bezos,  is placing new bets on automation. In  March, Amazon paid $775 million for Kiva  Systems, a company that makes robotic dollies that zip across warehouse foors car- rying shelves full of goods. Kiva found it  was more productive to have the humans who “pick, pack, and stow” items stay in  one place and let intelligent shelves come to  them. Among other reasons, Amazon said,  it bought the robotics frm because the technology ofered the chance to reduce labor  requirements at its dozens of warehouses.


Any work that is repetitive or  fairly well structured can be fully  or partially automated. this may  explain why economic output  has risen while the number of  jobs has fallen.

This is an example of what is going on  in the economy more broadly. As the MIT  economist David Autor has argued, the  job market is being “hollowed out.” High- wage, high-skill employment is still being  created—and so are many poorly compensated service industry jobs for food preparers, home care aides, and others. It’s the  jobs in the middle that are disappearing:  certain clerical, sales, and administrative  jobs and some on factory foors.

Now a combination of growing computing power and advances in data crunching  means automation is primed to threaten  not just tax preparers and travel agents  but higher-rung jobs such as those in the  medical and legal professions, where soft- ware can increasingly do things like analyze images and understand speech more  accurately and in more contexts than ever  before. Any work that is repetitive or fairly  well structured is open to full or partial  automation. Being human confers less and  less of an advantage these days.

Some economists believe automation  may explain why U.S. economic output  has grown since 2007 while the number of jobs has fallen. That kind of dislocation is unusual. The U.S. economy has  evolved from agriculture to manufacturing  to service industries. Each time jobs were  destroyed in one sector, they were replaced  elsewhere. Data from the Bureau of Labor  Statistics provide some clues to what the next economy will look like. Seven of  the 10 fastest-growing new job cat- egories between 2009 and 2011 have  the word “computer” or “software” in  them, according to an analysis by Matt  Beane, a doctoral student at MIT’s  Sloan School of Management.

Some say what’s taking shape is a more productive symbiosis between man and machine—and successful
businesses will be the ones that optimize it. Rodney Brooks, founder of ReThink Robotics in Boston, believes that a new type of general-purpose robot could reinvigorate manufacturing. The machines he’s building aren’t hardwired for any one job; they’re fexible, so many types of businesses could use them for a variety of production tasks. The company aims to democratize automation the way the PC did for computing, spurring similar efciency gains.

There’s defnitely good news here: more people than ever have access to afordable, powerful tools that can help them and their businesses become more productive. Take Todd Ruback, a privacy lawyer in Warren, New Jersey, who handles legal paperwork for companies that have lost sensitive data like credit card numbers. The job involves fling forms and notifying consumers in dozens of states, each with slightly diferent laws and deadlines. He’s been testing software made by a company called Co3 Systems that automates much of the process. It walks attorneys through what they need to do and prints out the right form letters for each state.

Ruback estimates that the software cuts the time it takes him to handle a case by 10 to 20 percent. But lawyers bill by the hour, so why would Ruback want something that makes it all go faster?

It’s pretty simple, he says. The software makes him more efficient. And if he doesn’t automate, the other guy will.

Tweetear

A biotech company is pursuing an approach that could redefne infectious medicine.

For some infectious diseases, traditional  vaccines just don’t cut it. Microbes that  hide inside human cells and cause chronic  illness aren’t stymied by the antibody response  generated by the kinds of vaccines available at  the doctor’s ofce. T-cell vaccines, which acti- vate a diferent type of immune response, could  in theory ofer a better way to prevent or control such infections, but so far nobody has been  successful at bringing T-cell vaccines from the lab bench to the clinic.

A colored scanning electron micrograph depicts a t cell.

Now Genocea, a biotech company in Cam- bridge, Massachusetts, thinks it can do it. It will  test the claim this fall with its frst clinical trial,  on an experimental herpes vaccine.

All existing vaccines rouse the body into  creating antibodies that attach to the sur- face of infecting microbes and fag them for  destruction. But pathogens that live inside our  cells, such as the viruses, bacteria, and other microbes that cause AIDS,  malaria, herpes, and chla- mydia, can evade this sur- veillance. “In order to deal  with those types of pathogens, oftentimes we  have to stimulate what we call cellular immunity,” says Genocea cofounder Darren Higgins, a  Harvard biologist. “Unlike antibody immunity,  which recognizes pathogens directly, cellular  immunity has to recognize the infected cell and  get rid of your own infected cells.”

It’s challenging to activate cellular immunity and the family of infection-fghting cells,  known as T cells, that drive it. The trial-and-error method used to develop antibody-based  vaccines has not worked for T-cell vaccines.

Despite years of academic and industry work,  and even clinical trials, there are no T-cell vaccines for infectious disease on the market. “We  don’t know all of the rules yet—if it’s possible  to make a T-cell vaccine, [or] how efective it  would be,” says Robert Brunham of the University of British Columbia, who is developing  a T-cell vaccine against chlamydia.

Indeed, our understanding of how T cells  control infection is still developing. The challenge is to identify the pathogen protein that  will grab a T cell’s attention and signal that a  human cell harbors an infectious agent. “If you can fgure out what those protein pieces are,  then you can use those proteins as a vaccine to  sort of educate your immune system on what  to respond to,” says Higgins.

The challenge gets tougher with pathogens whose genomes encode more proteins. 
There are 80 or so proteins in the herpes simplex 2 genome, about 1,000 in chlamydia, and  5,000 or so in malaria. Genocea has a high- throughput screening method in which it col- lects as many of a pathogen’s proteins as can  reasonably be produced in a lab and then monitors how human immune cells respond to each.

Although Genocea’s herpes vaccine is still  unproven, the work is moving faster than typical vaccine research, which can take 10 years  to go from discovery to proof of concept and  20 years to reach the market.