I do research & development in machine learning and natural language generation at Phrasee.
Getting computers to do “smart” things. By “smart,” we mean the sorts of things that humans and other animals do all the time. This includes vision, communication, reasoning, learning, etc.
This is the hype machine at full throttle.
I did a PhD in a related field. At the time, I would have said that I worked in “statistical pattern recognition” or “machine learning.” It is only in the past few years that the term “artificial intelligence” has gained popularity and is now used as a catch-all term for a number of areas.
"I was blindsided by the power of deep learning. It has solved problems that I never thought would be solved in my lifetime."
Optical Character Recognition (OCR) is a classic example of AI. I’m still surprised at how well computers can read my messy handwriting.
Postal services have been using OCR to help sort mail since the 1960’s. I love this example because it shows that AI has been around, and part of our lives, for much longer than most people realize!
The unknown. I was blindsided by the power of deep learning. It has solved problems that I never thought would be solved in my lifetime. I have no idea what the next breakthrough will bring. And that’s what excites me!
Since the 1950’s the tech world has gone through a number of AI “hype cycles.” We are currently near one of the peaks.
The technology is being over-hyped by many companies, and I fear a backlash against the field as a whole when those companies fail to deliver on their promises.
There are some deep and profound mysteries about the human cognition (e.g. consciousness and free will). However, these mysteries don’t yet apply to machine intelligence. AI is not mysterious and magical. Anybody with an interest in the topic can understand the basic concepts underlying the technology.
I have trouble imagining an advanced AI creating long-form fiction that humans enjoy. The stories we find compelling are intimately intertwined with the experience of being human.
A Google AI/deep learning researcher named François Chollet. On one hand, he avoids hype and favors intuitive explanations over unnecessarily complex ones. On the other hand, he often shares his excitement about the tremendous untapped potential of the deep learning.
I am fascinated by the technology behind deep fakes (but terrified by the implications).
"Postal services have been using OCR to help sort mail since the 1960’s. I love this example because it shows that AI has been around, and part of our lives, for much longer than most people realize!"
I wouldn’t change a thing. When I did my PhD in AI in the mid-2000’s I wasn’t sure it would lead to job opportunities. AI was a rather obscure field with limited commercial applications. However, several years after I graduated my skill set was suddenly in demand.
I wish I could say I planned it this way, but it was all luck!
The field of AI is evolving extremely quickly. The biggest pitfall would be teaching students about specific technologies that are obsolete by the time they graduate.
However, some topics are timeless, such as ethics and how to thinking critically about technology. I’d make course in these mandatory.
Focus on the value it will bring. It may be advanced technology, but is it solving a problem that is important to you?
Some vendors won’t deliver on their promises. Don’t let this taint your experience with the field as a whole.
Before worrying about what is going on under the hood, ask how performance/ROI is measured. Ideally, improvements over existing solutions can be quantified.
There are some tasks, like bid management, that are numerical optimization problems. People should not be doing this by hand.
Ask vendors difficult questions. Who do they work with, and what safeguards do they have in place to make sure the technology is not misused?
People differ on what information they are comfortable sharing. Therefore, companies and brands working on personalization can’t have a “one-size-fits-all” solution. They must allow people to set their own boundaries.
As some tasks get automated, resources should be reinvested into the human and creative elements of marketing.
"AI is not mysterious and magical. Anybody with an interest in the topic can understand the basic concepts underlying the technology."
It can get a bit heavy and technical at time, but there is a wonderful podcast called “Artificial Intelligence with Lex Fridman.” Each episode is an interview with one of the world’s top AI researchers or thinkers.
You can check out the podcast here.
Her (2013, Joaquin Phoenix).
Deep Learning with Python by François Chollet.
There are lots of great AI products on the market these days, serving a range of marketing use cases like personalization and customer intelligence. It’s now finally possible to use AI to optimize one crucial element in the marketing mix: language.
Obviously, I would recommend Phrasee’s AI-Powered Copywriting. By generating and optimizing short-form marketing copy like email subject lines and push messages, Phrasee’s tech is freeing up creative teams to be more creative. It also drives engagement and conversion results. We work with some of the world’s biggest brands, and some are making millions in incremental revenue with our product.
I know it is kind of obvious that I would be an advocate of Phrasee, but I really believe that the work we’ve been doing is having a powerful commercial impact on marketing across the board, and everyone should check it out.
"Neural text generation: How to generate text using conditional language models."