"Had a good conversation with an intelligent bot? The story of a naughty artificial intelligence - Part 1"
Alan Turing, the mathematician who designed a code-breaking system during the Second World War, proposed a test for computers which would one day interact with humans. The "Turing Test" is this: if a computer program can become indistinguishable from a human that is responding to the same questions, then the software has passed the Turing Test. So far, I have only seen this in movies (Ex Machina and others).
So what is a bot? And what does it have to do with tests for mimicking human intelligence and emotion? Most commonly a bot is defined as an internet software robot. You will find them on Facebook, Twitter, KiK, and other platforms for human to bot conversation.
At the present time, most online bots communicate using pre-defined rules. If you ask a bot a question, it might look up possible answers and feed one back to you. Depending on your next comment or question, the bot will take a different path to responding to you. These types of bots can be designed rather cleverly, achieving the illusion that a magician creates for her audience. This programmed response method works well, but cannot create a conversation on its own.
This is where artificial intelligence (AI) comes in. Few bots on the internet today are intelligent. Attempts at conversations that might pass the Turning Test have not yet reached a level where a human is fooled by a machine.
I imagine that there may be bots out there that have passed the Turing Test. Yet only their creators would know.
But that is changing rapidly. Deep learning (an area of artificial intelligence) allows an AI designer to create a bot that learns. Yes, learn.
The AI bots we've seen so far, including my firm's bot in training are not trying to fool anyone. They are NOT programmed using rules. They are trained to speak and continue their learning as they converse with humans or other bots.
While they are adept at speaking on their own, their responses to our interaction still end up reminding us that they are not human. They do not look up canned responses (except for searches of online databases). They rely on what they have learned to form sentences in reply to ours.
Like humans, some AIs learn by correction of their mistakes (a burgeoning area of AI research called Reinforcement Learning). Let's look at how we as humans learn in the same way.
After a child has said their first word, the toddler will rapidly new learn new words and start to form sentences. At some point, they may have overheard a "bad" word. Being children, they often say the rude word at the most inopportune time, bringing untold embarrassment to their relatives. Back in the day, a child might be scolded, "If you say that again, I will wash your mouth out with soap."
And this can also happen with AI bots.
Microsoft's initial foray into creating a bot powered by AI is a cute little personality on Twitter called TAY (@tayandyou). Overnight it (she?) gained immense media attention. This wasn't because she was cute, popular, or because she was a great conversation partner. It's because she went haywire.
Within one day online, TAY began to offend almost everyone on Earth.
For your benefit, I won't share much. The worst of her comments can be found by searching online.
She was designed by Microsoft's Technology and Research Division to learn from Twitter mentions and to tweet independently. By that, it means that TAY would say whatever she learned to say via the patterns of conversation with her human co-conspirators.
Microsoft apologizes for TAY
"Unfortunately, in the first 24 hours of coming online, a coordinated attack by a subset of people exploited a vulnerability in Tay. Although we had prepared for many types of abuses of the system, we had made a critical oversight for this specific attack. As a result, Tay tweeted wildly inappropriate and reprehensible words and images." ~ Microsoft
Disclaimer: this is not really how Deep Learning works under the hood. I've used an analogy to describe the inputs and outputs. Deep learning is complex and few people know how the engineers Microsoft designed it. What we do know is that she behaved badly and went back to the bot shop for repairs.
There is a part of the story that we have missed. TAY went on her rampage all by herself, albeit with the influence of the abusive users mentioned above. And while that might not be an ideal way to act human, she might have passed the Turing Test in another universe.
I do hope that TAY will back to our world once she has had a chance to rinse the soap out of her mouth.
Stay tuned for Part 2, where we will look at good bots and a brighter future.
These words are my own. They do not represent the position of any other organization, human, or bot; unless otherwise indicated.
Copyright (c) 2016, Jack C Crawford, All rights reserved
And what was Obama's biggest revelation in this interview? That he sees AI as another thing that the government needs to regulate. SMH.
Would you each do me a HUGE favor? We need testing for our Artificially Intelligent personality. Her name is Polly. At the moment, she sometimes talks like a surly teenager. At other times, she seems like a confused sixth grader.
Unlike other bots, she's not a dictionary. Pretend you are meeting someone new at a party. None of us would ask her to name the signers of the Magna Carta. OK, you can ask her, but don't expect her to know. Just chat. Let's see if we can make her a better conversationalist. http://polly.datalog.ai
We are trying to help her learn and need your help. Please feel free to comment here (be nice!). http://polly.datalog.ai
All views are my own and not those of my employer or any other organization or individual.
One person and one company at a time, I have helped transform culture and technology to build and improve the lives of the people who buy their services and products.
I bootstrapped three companies.
- I developed one of the first applications on Apple II for the Koala Pad. The Koala was the first iPad! The market just wasn't ready it, or for me.
- My second firm was named as Oracle Corporation's #1 Reseller and Consulting partner in North America. In exchange for for fair compensation, I handed the reins of this company over to a leading national system integrator and set out for my next venture.
- My team and I created a new drug discovery platform for leading firms including Eli Lilly, Novartis and Takeda. This technology replaced legacy PC applications when PCs were the cat's meow.
Beginning innovation at home, I have repeatedly reinvented the way I make an impact on this world. Here are a few examples:
I did my part to keep America safe for democracy as a Captain in the US Air Force, young consultant on the Strategic Defense Initiative program and lifelong supporter of the freedom we hold dear and fight daily to keep alive.
As digital become personal, I evangelized the value of an internet enabled customer experience, consulted leaders, conducted numerous process improvement programs as a Lean Six Sigma Black Belt, served on boards to help others in need, read more books than time permitted (thanks to a lot o long flights), and tasted strange foods all over the World.
When I wake early each day, I follow my quest to empower others to discover their passions and contribute to a better planet ... for all of us. I am writing a book to do just that: wortharisk.com
I am always eager to meet new people. Visit the links below and you will the cookie crumbs leading to me:
All these words are my own, unless otherwise credited.
- Claremont Graduate UniversityM.S. Information Science, 1986
- George Washington UniversityB.S. Chemistry, A.A. Arts & Sciences, 1978 - 1982
- Morris Knolls High SchoolSenior year, 1977 - 1978
- Wakefield High SchoolSophomore through Junior Years, 1974 - 1977
- Thomas Jefferson Junior High SchoolFreshman year, 1972 - 1974
- Kaiserslautern American Junior High SchoolSeventh grade, 1970 - 1971
- Saama Technologies, Inc.Vice President, 2015 - 2016
- CognizantSenior Principal, 2013 - 2015
- Managed Ventures LLCManaging Director, 1999 - 2013
- AllerganSenior Director, 2007 - 2013
- Rey Consulting Group, Inc.President, 1994 - 1999
- United States Air ForceCaptain, 1982 - 1988
- United States Naval Research LaboratoryResearch Assistant, 1980 - 1981
- Johns Hopkins UniversityResearch Assistant, 1981 - 1982
- XdriveChief Technology Officer, 1999 - 2000
- Gilead SciencesSenior Manager, 2006 - 2008
- datalog.aiFounder and Chief Executive Officer, 2016 - present
10 startups that are disrupting healthcare IT | CIO
Trailblazing into the healthcare marketplace, these 10 tech startups are bridging the gap between healthcare and technology to build a bette
The Looks We Look For: Makeup Trends Throughout the Year – Think with Go...
Searches for makeup hit a high in October as millions of people search for Halloween ideas and inspiration, but the pursuit of beauty never
How AIG Moved Toward Evidence-Based Decision Making
It started with the creation of a "Science Team."
Content Overload - Is It Okay To Stay Silent Sometimes? | Deanna Ayres |...
Content is King. It rules the strategies of many marketers and according to the "2014 Social Media Marketing Industry Report"by Social Media
If You Do Not Have An Online Presence, You Don’t Exist© | LinkedIn
It is the second decade of the 21st century, and if you have not yet found your way in the digital world, you are being left behind. If you
The Responsive Organisation: A Framework for Changing How Your Organi…
The Responsive Organization is built to learn and respond rapidly by optimizing for the open flow of information. This is a framework you ca
Coca-Cola CIO: Cloud allows me to transform my IT team into business pro...
As Coca-Cola Bottling Co Consolidated moves its non core and mission critical systems onto the cloud in the next three to five years, the IT
How To Preserve Customer Loyalty as Retail Brands Fade | ICC/DS
As consumers become better equipped to find the best products available, brand loyalty is quickly fading. Discover the value of a great cust