Laura asks, “What is the industry trend you’re most excited about in 2020?”
The continued growth and practicality of artificial intelligence, of course! Three specific areas:
– Continued growth in natural language processing
– Exciting new trends in audio, especially music composition
– More pre-trained models for everything
The three areas I’m worried about:
– Interpretability and explainability
– Misuse/malicious use
Can’t see anything? Watch it on YouTube here.
Listen to the audio here:
- Got a question for You Ask, I’ll Answer? Submit it here!
- Subscribe to my weekly newsletter for more useful marketing tips.
- Find older episodes of You Ask, I Answer on my YouTube channel.
- Need help with your company’s data and analytics? Let me know!
- Join my free Slack group for marketers interested in analytics!
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.
In today’s episode, Laura asks, What is the industry trend you’re most excited about in 2020? Well, the continued growth and the practicality of what’s happening in artificial intelligence and machine learning and in marketing data science force.
There’s three specific trends that I think are so exciting and have been exciting for quite some time.
Number one, continued growth in natural language processing.
2018 was sort of a watershed year for a lot of natural language processing in 2019 brought some of the largest models available that we can use to process text, to understand it better to be able to in some cases, generated and that is stuff that I expect to see continue in 2020 not just at the cutting edge of research but at the ability for the average machine learning practitioner to access it and deploy it.
So by that i mean it’s it’s relatively friendly, relatively easy to use it is not the incredibly challenging you know, super high tech stuff although that that stuff is exciting too.
But for the average well versed practitioner in data science and machine learning to be able to pick up something in NLP and use it.
Number two, there are some really amazing things happening in the world of audio with wavenet a few years back and now Mel net bunch of other technologies that are making it easier for machines to understand audio, particularly the spoken word and Then able to replicate it and synthesize it.
And this is some amazing stuff again, there’s if you look at the milnet demo, you can listen to a machine synthetically generate voices, synthetic degenerate music.
I think music composition is has come a long way and continues to accelerate.
It’s not going to replace humans anytime soon.
But it’s getting there.
And it I would say it’s at the point now where AI can generate, like elevator music, right? stuff if you put it on in the elevator like no one would be offended.
But certainly it’s not going to win any Grammys.
But as with all things in, in machine learning, the technology improves, the models get better.
And whereas two years ago, it was it sounded like something like your three year old would You’re hammering pots and pans with spoons.
Now it sounds like elevated music.
So expect in time for it to become a credible alternative for people who want access to music composition and don’t have the skills themselves can’t play an instrument but can direct or conducting an orchestra of machines.
Number three, and this is a bit more esoteric, but there will be more and better pre trained models in 2020.
We saw within the natural language processing field open AI is GPT to released its mega model, the 1.5 billion hyper parameter model in 2019.
And that was a really big deal for that company because that was the model they were so concerned about falling into the wrong hands.
And they said they didn’t see any evidence of its its use and partly because it’s such a beast try and get to us.
But one of the major Overall meta trends and AI in the last couple of years has been, instead of generate your own models, pick up a pre trained model and then fine tune and expect to see more of that happening in music, image recognition, video, natural language processing, good old fashioned statistical computing all these things.
We’re seeing more and better models for better just you pick it up and you work with it.
Now, there are some challenges with that in the chat.
One of the big challenges, of course, is that you are relying on a pre trained model and you you don’t know for sure how that model was trained.
You after take on faith to some degree, the model was trained well, so that’s going to be a consideration.
Which brings me to three trends I’m concerned about number one is interpreter ability and explain ability, the ability for machine learning and AI practitioners to explain what is the models are doing, which is really important and something that The industry is struggling with and will struggle with for some time.
Partly because interpretability is so much more expensive than explain ability, but is the gold standard for being able to say, watch and see what the machine is doing any step of the process.
Number two, the things I’m worried about is, is bias.
And this goes hand in hand with interpretability.
If we don’t know what the machine is doing, we don’t know if it’s doing something wrong.
There have been many, many, many examples of machines making decisions that on the surface seem okay.
But when you get inside the model, or when you see the post hoc results, you’re like, That’s not right.
So bias is a major concern.
And it’s one that the industry is making strides on.
But the industry as a whole is not going fast enough.
Not just fast enough to allay some of the fears that people have.
And, and set aside the misconception that’s important.
And number three, and by far the one I’m most concerned about in 2020, because it is an election year is the misuse or malicious use of artificial intelligence technology for things like deepfakes is the most popular cited example not the one that I think is probably the most prevalent.
I honestly think that bots with minimal natural language processing capabilities are much bigger problem because there’s so much easier to scale deepfakes don’t scale well.
deepfakes require a lot of computational power.
And yes, you can rent it for pennies on the dollar from like a Google Cloud or something.
But to do so, then also makes you not anonymous, right? Because once you sign into something like when a big text clouds every single thing you do is tracked and can be identified but long for So in a lot of cases, if you’re doing something malicious, you need to be doing it in the dark away from the prying eyes of every major tech company ever.
So things like, you know, mass armies of Twitter bots and Facebook bots and things are a much more practical application.
And very easy, very cheap.
And they have the ability thanks to the hyper partisan world that we live in, to really manipulate people.
And it’s not the machines fault that humans are gullible and that humans like to have confirmation bias out the wazoo.
But it’s the machines enable hostile actors to do more faster and better.
And, at least in the context of the United States of America, our defenses have been largely dismantled the last couple years with the abolition of the cyber security Council and things like that.
So we are in a case where AI can be maliciously used.
And that’s very concerning to me as a practitioner because, again, we want people to trust this technology.
If the technology is being used for malicious means really hard to build trust around it.
So that’s sort of the opposite of is exciting trend.
That’s the the most worrying trend but that’s what we have to look forward to in 2020.
For marketing data science for machine learning for artificial intelligence, would love your thoughts, leave them in the comments box below.
Subscribe to the YouTube channel on the newsletter, I’ll talk to you soon take care what help solving your company’s data analytics and digital marketing problems.
This is Trust insights.ai today and let us know how we can help you
You might also enjoy:
- The Biggest Mistake in Marketing Data
- How To Start Your Public Speaking Career
- Is Social Listening Useful?
- Almost Timely News, 17 October 2021: Content Creation Hacks, Vanity Metrics, NFTs
- Marketing Data Science: Introduction to Data Blending
Want to read more like this from Christopher Penn? Get updates here:
Get your copy of AI For Marketers