Improspectives

Improv skills lead to success

Posts Tagged ‘Microsoft

Tay, Improv, and Artificial Intelligence

leave a comment »

Kristian Hammond, a professor of computer science at Northwestern University, wrote a guest article for MIT Technology Review that offered his perspective on how the spectacular public failure of Microsoft’s Tay chatbot could have been avoided. Hammond brings up some good points, but I believe his analysis is incomplete.

What…Happened???

Many of you have probably heard about Tay, the youthful-seeming chatbot Microsoft released into the wilds of Twitter. Within a very short time, malicious users took advantage of the bot’s learning algorithms and caused it to create homophobic, racist, and anti-Semitic tweets. In a press release, Microsoft noted that they hadn’t had a problem when they tested a version of Tay in China, but I argue the team should have suspected the cadre of trolls on Twitter would take shots at the bot and try to make it produce offensive tweets. At the very least, the team should have built in a list of banned terms rather than use a strictly naive learning procedure.

Hammond, who is both a computer scientist and improv comedian, argues that using a combination of techniques inspired by Marvin Minsky and improv comedy could have helped avoid the worst effects of malicious targeting. I agree, and believe his notes assessing the difficulty of the AI problem Microsoft tackled are spot on. I do have some significant disagreements with his suggestions for how using improv techniques, specifically regarding show management, would help.

Improvised Doesn’t Mean Unstructured

Improv comedy groups, which rely on audience suggestions to make the show run, must determine how much control they want to grant the audience. Some groups are open to any and all suggestions, regardless of how offensive, and build the best scene they can given the subject matter. Other groups control their subject matter more closely. The trick is finding the right balance to do a show you’re comfortable with and that will attract an audience. But beyond attracting an audience, you must attract the audience you want.

Much, if not most, improv is done in bars. This consideration is especially true in the Chicago area, where Hammond works. That consideration means at least a portion of your audience didn’t know they were going to see an improv show, doesn’t want to see an improv show, and are drinking. A lot. Hammond notes that groups can manage their show by choosing which suggestions to ignore (perfectly acceptable) or by pointing out that it’s ridiculously obvious someone making deliberately offensive suggestions just wants to manipulate the show. He further states that a similar technique could work on Twitter:

Nothing neutralizes a bully as well as being called out. My guess is that if Tay pointed out that it knew it was being played in one-on-one interactions and provided attribution for newly learned “facts” when using them in public tweets, the shaming effect would have been enough to shut down even the nastiest attacks.

I believe Hammond is just plain wrong on this point. As Whitney Phillips, now a professor at Penfield College of Mercer University, discussed at length in her book This is Why We Can’t Have Nice Things, Internet anonymity shields trolls from the consequences of their actions. Trolls do what they do for lulz, laughs at someone else’s expense, and either don’t care or get lulz when their unattributed Twitter posts provoke someone enough to warrant a counterattack. Alcohol provides a similar shield for audience members watching improv shows in bars. The bar makes money off drinks…entertainers are just there to attract audiences and help maintain a steady flow of orders. Many talented performers, whether improv comedians or musicians, have lost gigs because they couldn’t get enough friends to show up and spend money each week.

I also disagree with Hammond’s depiction of the consequences for a drunk twenty-something audience member who “scream[s] out obscene suggestions that she will regret for the next two years”. First: been there, feel your pain. Second: she probably won’t regret what she said because she won’t remember what she said. For individuals such as her (or him), this incident is just one of many similar nights on the town. You just happened to be there when it went down.

Conclusion

The team behind Tay failed to accurately assess the environment into which they released their bot. That said, Microsoft can move forward by using another time-honored improv technique: the Failure Bow. When a scene, song, or on-the-spot pun goes poorly, the performer steps downstage center, faces the audience, says “I failed. Thank you.” and bows. Acknowledging the moment helps everyone move on, most of all the person who failed.

Dr. Adam Gazzaley: Gaming and Brain (OHSU Brain Awareness Series)

leave a comment »

Oregon Health Sciences University (OHSU) is a leading research and teaching institution located in Portland. They run a speaker series every year, with this year’s presentations centered on the brain. I’d forgotten to put the talk in my Outlook calendar, so I was very happy to receive a reminder message earlier Monday.

Our Speaker

The speaker, Dr. Adam Gazzaley, is a member of the faculty at the University of California San Francisco (UCSF) and, as required by California state law, a co-founder of and adviser to a start-up company. That company, Akili Interactive Labs, produces games designed to provide cognitive benefits to players who play the games for 30-minute sessions three times a week over the course of a month or two. Gazzaley is an engaging speaker, though it seems like he’s spent a lot of his recent appearances speaking more about ideas and vision than hard science. As another attendee remarked in the lobby after the talk, “His presentation was more like a performance than a science lecture.” That approach isn’t surprising, given that he co-founded a company and has to compete for investment funds, so I didn’t think it was a problem.

I have to admit that my heart sank a bit when I realized the presentation focused on brain training using games. Lumosity recently agreed to pay a $2 million fine to settle a U.S. Federal Trade Commission deceptive advertising action that claimed the company’s brain games could help students and seniors improve mental performance. Gazzaley was quick to point out that Akili’s games are going through the U.S. Food and Drug Administration (FDA) approval process, with their first game ready to move from a pilot program step to wider public testing. Unlike Lumosity’s claims and those of “nutraceutical” supplements (available in stores throughout the U.S. or your favorite multi-level marketing scheme) which make claims that have not been evaluated by the FDA, Akili is going through the rigorous, some would say onerous, testing required by the FDA before it will back up claims of a product’s efficacy. Gazzaley emphasized that if the games don’t pass FDA trials the first time around, they’ll revise them and try again.

Works in Progress

Gazzaley showed several games that combine motor skills and recognition tasks. The game that’s farthest along in testing is a driving game where the player attempts to keep a vehicle centered on a winding road using a joystick. The player must also press the joystick’s button when the vehicle passes under a sign that displays a designated symbol from a set of possible designs. The game’s adaptive algorithms increase or decrease the difficulty level based on the player’s performance, with the goal of encouraging an upward path. The player must switch between the two tasks — as switching becomes more fluid, their score increases. Gazzaley noted that “humans love to level up” regardless of age, a finding borne out by gamification research, so games are a natural mode for brain training.

He also noted that there is the possibility games can be used to treat Alzheimer’s Disease, Parkinson’s Disease, ADHD, and other conditions. While his work on games as treatment is still in the very early stages, there is the potential for significant advances. This research program will benefit from new devices such as Oculus Rift or Microsoft’s HoloLens, which generate virtual reality spaces in which a player may act. There are also interesting new neurological measurement devices, such as a whole-head electroencephalogram (EEG) helmet that transmits data to a waiting computer, that can provide researchers additional flexibility in measuring brain activity. I also see great potential for other serious game applications involving teams, such as squad-level military exercises or firefighting, though Dr. Gazzaley said that he’s not aware of any work yet in that area.

Conclusion

I enjoyed Dr. Gazzaley’s presentation. Yes, it was a little flashier and a little less sciencey than other talks I’ve attended, but he provided a lot of useful information and compelling demonstrations that made the hour fly by. I will definitely follow his work in the coming years.

To Be a Beginner Again

leave a comment »

When I think of what it means to “be a beginner” at something, I think of learning a new language or trying a new sport. I didn’t expect to rediscover the joys and struggles of being a beginner as a writer.

I had just finished Microsoft Excel 2016 Step by Step for Microsoft Press, my sixth Excel Step by Step book, when the publisher approached me to take on Microsoft OneNote Step by Step. I looked at my schedule, swallowed hard, and agreed to do it. I’d worked a bit with OneNote as part of my Office Online Essential Training course for lynda.com, so, while I wasn’t an expert, I wasn’t coming in completely cold. Besides, I’m a writer and course developer—my job is to tease out a program’s intricacies and make them clear to the reader or viewer. How hard could it be?

I’ll pause until you stop laughing.

You can write a book about anything if you do enough research, develop a few ideas of your own, and quote liberally from other sources. I just read a business book, Everything Connects, that did exactly that. The main author probably wrote a great proposal based on his experiences as a serial entrepreneur and meditation practitioner, took his advance, and wrote down everything he could about those subjects. My guess is that he produced about 150 pages for a planned 250-page book, so the publisher brought in (or had already hired) a professional writer as co-author.

I also turned to the supporting literature on OneNote for guidance, but there’s not a lot out there compared to the vast, rich resources on Excel. That said, I wrote what I could and discovered a lot as I went along, but I didn’t have an experienced user’s feel for the program. I’m fortunate Microsoft convinced Ed Price, formerly a member of the OneNote product team, to be the book’s technical editor. Ed knows the software in depth, both as a user and someone familiar with the broader customer base’s needs and desires. He added a lot of material I’d considered not important enough to include, changed the emphasis of certain sections of the book to improve its usefulness, and became, in all but name, a full co-author.

I’m grateful for the substantial help Ed provided and hope to make him a full co-author, with cover credit, when it comes time to refresh the book. As a writer it was good, though incredibly frustrating, to write about a program I hadn’t worked with extensively. I had 20 years of skill and discipline to power through a first draft we could use as a basis for critique, but I had several flashbacks to when I was just starting out and lacked the tools I have now.

I survived (with the help of Ed and others), the book will provide good value to OneNote users, and I was reminded how difficult a thing it is to produce a manuscript from whole cloth. I’m glad I agreed to help out, but I’m sure glad we’re done.

Microsoft and Minecraft

leave a comment »

It appears likely that Microsoft will announce it has finalized a deal to acquire Mojang, the Swedish company that created Minecraft. I think it’s a great idea.

Minecraft is huge–players of all ages build creations that demonstrate their creativity and test the limits of the platform. Business analysts have asked why Microsoft would want to buy Mojang. The game is available on the PC but not Windows Phone because of the latter’s 2.5% market share, but it doesn’t seem likely that a single game, even a tremendous hit, would bring $2.5 billion in added revenue to Microsoft.

It’s true that Minecraft won’t make Microsoft a mobile front-runner, but Mojang’s insights into the Minecraft community’s wants, needs, and emergent behavior could be worth much more than the hefty purchase price. Microsoft has made strides toward listening to and acting on user input, such as by altering Windows 8.1 and, if leaked screen shots are to be believed, Windows 9/Threshold to be more in line with user preferences. Bringing in the Mojang team could provide similar insights to teams across Microsoft, allowing the folks in Redmond to incorporate what they learn into future products and giving them a fighting chance of thriving in the mobile world.