After seeing a bunch of online quizzes that could guess what kind of year I will have in 2018 based off of strange things such as my preferred pizza toppings (mushrooms and green peppers) or my favorite Beatles songs (Penny Lane and Come Together FYI), I felt inspired to put together a list of AI predictions for 2018.

As with all things AI, I reached out to my fellow ROSS cofounder, Jimoh Ovbiagele, the brains and the beauty of the operation, for his take on where he sees deep learning going in 2018, what deep learning improvements he sees are on the horizon, challenges the AI community will face this coming year and lastly, to get his take on what his AI wish for 2018 is.

First things first Jimoh, I’d like your prediction on where you think deep learning will make the biggest impact in 2018?

Natural language processing. Amazon, Google, and Apple sold millions of devices with voice interfaces in 2017. To deliver on their promises, they are betting heavily on deep learning to process and fulfill the voice commands given by their users. 2018 is going to be the most significant investment yet in deep learning and natural language processing. I expect to see incredible improvements in natural language processing, similar to what occurred in computer vision in 2017.

Makes sense to me. It seems that Google Home and Amazon Echoes were among the “it” gifts this holiday season and with that must come steady improvements in the natural language processing field which obviously bodes well for ROSS Intelligence as well as we continue to develop on the bleeding edge.

I’d like to get your prediction on what you think will improve deep learning the most in 2018?

More compute power. Intel, NVIDIA, and other prominent chip manufacturers have been developing chips specifically for deep learning. These chips are promised to be orders of magnitude times faster than current chips running deep learning algorithms. Faster chips mean deep learning engineers/scientists will be able to conduct experiments quicker — decreasing the time it takes from months to weeks to days to hours to minutes — leading to faster results about what works and doesn’t work. Also, because we can run deep neural networks through more training iterations in the same amount of time, we increase the likelihood they learn the most optimal features.

Agreed here, NVIDIA’S recent release of Titan V had me picking my jaw off the floor — 21.1 billion transistors and 110 teraflops of horsepower — wow. Looking forward to where we will be at in 2019 and beyond.

Changing gears here with this question, what is the biggest challenge in 2018?

Backlash. The AI hype of 2017 has set buyers up for disappointment. Real AI companies will need to distinguish the actual usability, performance, and reliability of their systems compared with AI wannabes.

I hear that. I see 2018 being a great year for separating the wheat from the chaff when it comes to those who slapped AI into their company’s marketing materials and those who have been patiently building great AI-enabled software.

Alright, last question, if you could get a wish granted for 2018 in AI, what would it be?

The time it takes deep neural networks to learn drops by 100X!

I should have known that would be your wish — reminds me of the comic below of a whale saying their New Years resolution is to lose thirty-eight thousand pounds — steady progress will win the AI race Jimoh!

‍