The True Dangers Of AI Are Closer Than We Think
Eye-opening discussion on bias in AI
The five-part series 50 States of Voice concludes Monday with the release of Part 5, covering voice/AI initiatives from South Dakota to Wyoming.
The previous four parts can be viewed here: Part 1, Part 2, Part 3, Part 4.
One of the voice/AI articles discussed on today’s This Week In Voice (Season 5, Episode 2) was The True Dangers Of AI Are Closer Than We Think, published by Karen Hao of MIT Technology Review.
The article itself is paywalled, and well worth reading, but you don’t need to see it to get a lot out of the discussion that took place that is excerpted below.
This prompted an excellent discussion from the panel which is transcribed, in part, below. (This story discussion begins within the video at 33:06.)
Kat Zdan (Conversation Designer, Google):
Sure. I actually take exception with the premise of the article, and where it comes from, in two ways.
There's a problem that we all have, when we speak about AI, and that is that the primary metaphor we have to understand what AI is, is the human mind. That's a flawed premise. The mental model of AI being like a human mind, or a human mind being like AI, is fundamentally flawed.
The way I think about it is a human mind does have a Turing machine - in other words, we can do math, like an AI. But the human mind has so much more than that.
If you think about it from narrow to broad scale, the human mind is connected to emotions...is corporeal - is housed in a body...exists in relationships, in family, in community, in society...all of these things input the way that the human mind works, which is so much more than a Turing machine.
So that's my first exception: it's a very human mistake to make, is we think about AI being something other than it is.
The other large exception I take is the article starts with the thought: "is AI going to take over the world?"
In order for anything, including AI, to take over society, or take over the world, it has to interrupt and disrupt the power structures in place.
What the article points to makes that impossible, because the article points to the fact that AI is upholding, and reinforcing, the status quo, with the biases that the human beings which have programmed it have put into it.
Lisa Michaud (Senior Product Manager, Interactions):
I did feel, for me, that the article shifted really quickly to the danger isn't that we're headed for Terminator real soon - it's this perpetuation of bias. This is something that's a really, really large topic of conversation in the AI and NLP communities right now.
When we first started moving toward machine learning as an approach to AI, the idea was to move away from the problems of rule-based AI.
In top-down AI, where they were trying to explicitly take the things that we use to judge, and make decisions, and make choices, and explicitly encode that so that the AI could then make decisions in the same way...we found that that was really flawed, because first of all, it was very, very difficult to really get to the nuggets of what it was that people were using to make decisions.
But also, people have all of these biases, and when they presented their own decision-making processes, they presented those through the filter of those biases.
The idea of machine learning was, let's step away from that. Instead of trying to explicitly teach the AI "this is how you make these choices," we will give the AI an enormous amount of data from which it will learn how to make these choices, without the human beings putting their own bias in there...not realizing that the data itself was going to teach those biases, because of the way we collect the data.
Caitlin Gutekunst (Senior Director, Marketing & Development, Creativity Inc.):
Looking at the article itself, the last question that was posed to the main subject was "what do you dream about with the future of AI?" It was a very hopeful way to end the article.
The author said it could be a great equalizer.
There's great things that AI can drive - AI being conversational tools, as well as facial recognition, and all the other types of manifestations of what machine learning is contributing.
Is it the great equalizer right now? Probably not. Access is an issue. It's maybe not as meaningful as it could be for people's lives, and a part of their lives, in a way that would really enhance standard of living, in a way that's not just a convenience factor.
I always look at voice as having two main purposes. The first is as a marketing channel, so for that, you think of commercialization, consumers, and a 'have' sort of bucket.
The other one is an interaction modality - being able to converse with something. The kids and subjects that we design for are young children, so they may not be able to read.
We've also built an interaction for UNO Braille, their latest game which was built for people with visual impairments. It was a tutorial for people who were learning how to play that game, and have an AI game master for Amazon Alexa and for Google.
So there's great ways that AI can afford us with accessibility, for people who may not otherwise be able to interact with technology in the same way.
Can it improve those standards of living? Right now, absolutely yes.
The Project Voice Advisory Group assists companies looking to add voice / conversational AI, reducing cost while supporting growth. We’re the map and the compass, helping you navigate what can be tricky and complicated waters. We’re working with multiple Fortune 1000 companies already in this capacity - reach out and let us help you create your technical roadmap and manage the RFP process.
Project Voice: Catalyst assists companies operating in the voice/AI space to generate marketing and business development wins, as well as assist in exploring M&A and venture capital as required.
Register for Project Voice: Coast to Coast, our upcoming 34-city US roadshow leading up to Project Voice 2021, using any of the links below for our initial dates:
The Voice of Money examines the rapidly growing intersection of voice, AI, and modern banking and finance.
The Voice of Customer Service is a special two-day event all about how conversational AI is being deployed within modern customer service operations, including contact centers around the world.
Project Voice is the #1 event for voice and AI in America, fully curated and moderated by Bradley Metrock of This Week In Voice VIP.
In-person attendance will be limited but registration is ongoing (register by clicking on the banner below). Feel like going to your first post-pandemic in-person conference next April? Don’t wait to register then - will be sold out well before.
