Is AI Really Everywhere? Scorsese!

I am sure many of you have had the experience of texting someone only to get back a reply that is either not at all related to what you sent or is simply question marks and a confused face emoji. Only then do you realize you have fallen victim to autocorrect. I myself experienced this last fall when, after encountering some car issues, I texted my sister. She offered her condolences and wanted to know if I planned to get another one. Well, yeah, I needed a car—but my phone corrected my message, “the cay died,” to “the cat died.” Fortunately my pets were fine. And I love my cats, as anyone who knows me will tell you, so it is not a surprise that my phone autocorrected to cat. But how would the phone know to make this correction? Autocorrect uses machine learning and is perhaps the simplest example of AI (artificial intelligence). Each phone, tablet, or even program like Microsoft Word is set to study patterns to see commonly used words as well as how we react to suggested changes.

Google’s dictionary defines AI as “the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition,

decision-making, and translation between languages.” Essentially, AI is found on devices that have been programmed to teach themselves rather than to work from a pre-programmed list to reference. Again using autocorrect as an example, a lot has changed over the years. If you recall older versions of Word or Word Perfect, those programs used to have a dictionary function which listed commonly misspelled words to alert the user to a possible mistake. That eventually evolved to allow the user to add in their own words and then to today’s version of autocorrect, which still features a standard dictionary of words at its base, but is also programmed to learn new variations or words based on what the user consistently inputs. The phone is learning a new dictionary on its own by using a set of algorithms behind the scenes. In my example above, my phone knew I used the word cat far more often than I used car, so it guessed that I wanted to say cat. On a new phone, the algorithms are set to train the autocorrect by providing examples to use to see if the user will accept or correct them. An example of this is one my friend encountered when she texted her mom about some overdue DVDs that needed to come back to the library. She meant to type “of course” and autocorrect on her new phone inserted “Scorsese.” Likely the algorithm saw letters that looked a bit like Scorsese and decided a conversation on DVDs must include the director.

Beyond phones, programmers use similar algorithms in other devices and software to do everything from automate machines to offer correct auto responses to basic information requests to run large sets of data analysis for medical or other scientific applications. This form of machine learning tries to drive the software toward correct answers without knowing those answers in the first place. This is different from autocorrect in that the device will try to reason as well as offer suggestions. A chatbot is an example of this type of application, where a user will ask a question and the robot will look up the answer for you. Chatbots can be found in information kiosks at places like airports or as a text messaging system for helplines. Universities, for example, will often have a chatbot set-up for faculty to use to save time on email tasks, like if a student wanted to know what the week’s reading assignment is, the chatbot would check the date and reference a stored version or the course outline. Because the programming on these chatbots is more complex than trying to figure out what you meant when you misspelled a word, they can get into conversations and follow a thread by using the previous questions you asked it to reply and to ask questions of you to clarify what you are asking it to compute. Cornell University ran an experiment on this type of chatbot and got a rather interesting result that they posted to YouTube—two chatbots engage in a conversation that is fairly logical, despite a unicorn reference.

Perhaps one of the most advanced, and recognized, AI applications in use today is IBM’s Watson, which gained fame for being able to beat two of the most successful champions in the history of the game show Jeopardy!. Watson has blossomed into a full-service AI and has applications in a variety of fields, including health, finance, and education. An example of Watson in the medical field is Watson for Oncology, which is an AI that allows doctors to pinpoint certain types of research articles, clinical trials, and suggested courses of treatment based on data sets they enter about a specific patient. Along with health, other sciences use Watson to compute huge sets of data that normally would go untouched or take decades for humans to collect and analyze. Another example of this type of AI is the Folding@Home project run out of Stanford University. Home users can download apps to their devices, usually computers or game consoles like PlayStation and Xbox, to donate computing time to the project, which uses the computers in the network to analyze data on different diseases and run scientific experiments on possible cures.

Lastly, AI is inserted into devices that you are probably already using at home or work. In addition to autocorrect, assistants like Siri, Cortana, and Alexa are all AI-driven programs. Devices themselves can be run by an AI without much input from humans, such as home security systems that send out an alert when there is suspicious activity seen on a camera or automated vacuums that detect carpet thickness, remember where walls are, and know when to go back to their charger to refresh their batteries. Robots that are used in high-risk jobs also have a good deal of AI in their programming to help with split-second decisions when time is essential or an incorrect decision can lead to serious error.

—Laura N.