The Inside Intelligence on Artificial Intelligence: Q&A With Mike Tamir
- Updated on: 2019-04-25
- Read original article here
The demand for skills in artificial intelligence (AI) and specifically machine learning has been growing exponentially over the past five years, as businesses from online entertainment to eCommerce scramble for new ways to utilize data to improve customer experience and realize new features.
Simplilearn recently appointed Mike Tamir, Ph.D., as the Advisor for Simplilearn's Artificial Intelligence and Machine Learning curricula. He has been instrumental in developing the course structure and incorporating advanced programs on AI Engineering, Machine Learning and Deep Learning with TensorFlow. Dr. Tamir is ranked number one globally as an influencer for Machine Learning and AI by Onalytica and currently serves as Head of Data Science at Uber ATG (self-driving vehicles) and is a lecturer for the University of California, Berkeley - iSchool Data Science Master’s Program.
Recently, Simplilearn spoke with Mike Tamir about his insights, predictions, and recommendations about machine learning and how both businesses and career-seekers could prepare themselves for the future. Regarding the views expressed in this interview, it’s important to note that Mike Tamir is speaking solely for himself and not as a representative of his employers, particularly Uber.
Q: Hello Mike, welcome to Simplilearn. What does “Artificial Intelligence” mean in the real world today?
A: AI, in general, has been a moving target as a concept. Modern usage of the term AI has become synonymous with machine learning applications for the purpose of automating certain tasks that humans can’t do at scale. Between Big Data, Data Science and AI, there’s a common thread of utilizing huge volumes of data to accomplish real-world tasks and solve real-world questions at scale.
The key to this has often been machine learning. We are currently capable of accomplishing a whole host of simple tasks with machine learning, probably far better than we thought we would be able to do; and in several cases we’ve exceeded our most optimistic expectations.
Q: What’s behind the explosive evolution in machine learning?
A: Over the past several years, we’ve seen a lot of success, thanks to the rise of open source tools. Our ability to use deep learning at scale in production now is possible because of TensorFlow. The competition that TensorFlow has gotten from Pytorch, MXNet and a host of other competitors made TensorFlow really step up its game. We also have uniform exchangeability with ONNX [ Open Neural Network Exchange] that lets us transfer deep learning models between frameworks. This flexibility gives us many powerful ways to put these algorithms into production.
Q: What are some of the various applications of AI for business?
A: There are a lot of really fascinating use cases, for example, dealing with different kinds of equities and being able to forecast the value of those equities under different contexts. Certainly, finance has really started to take advantage of machine learning and deep learning applications.
There are use cases for natural language processing, which is one of the most exciting applications for me. With modern neural net supported Natural Language Processing, we can represent language in semantically rich ways. Words can now be represented mathematically in ways that capture how different concepts relate to one another.
A: Back when these technologies first came out, we would do things that are simple, like just add up all the vectors and take the average. You would get a rough vector that kind of points in the direction of what that paragraph means. But the problem with taking averages is that addition is commutative. When something’s commutative, you can switch things like, “Jack fell down but did not break his crown,” or “Jack did not break his crown but fell down.” They’re all the same words but very different meanings, in fact kind of the opposite meaning.
Now, with recurrent neural networks like “Long Short-Term Memory” or LSTMs, we have a deep learning algorithm that takes things in sequence. It remembers, forgets and keeps track of different things as it’s going through a sequence, enabling it to answer certain questions. Other tasks include getting these algorithms to rearticulate the meaning of a sentence in another language, or with fewer words. What we’ve found is that you can get much better at neural machine translation and summarization, which is why Google Translate got so much better in the last couple of years. Lastly, you can use these techniques to help solve problems like labeling text. For example, what we did at FakerFact.org was that we built a similar algorithm that reads through the text and then gives an assessment as to whether the text is “real journalism” or not and why.
Q: What are some of your favorite machine learning developments?
A: I’m really passionate about the FakerFact research project. Using these modern techniques for fake news detection is really exciting, because we can now detect if an article is trying to present us with just the facts or if it is more focused on being sensational or clickbait, trying to get an emotional reaction from a hot-button issue. We created algorithms that are pretty decent at this and put them up on the FakerFact site.
One of my all time favorites is “wisdom of the crowds”, where if you have a lot of independent sources giving you feedback, but they are not experts, you can aggregate their feedback in order to get to the truth. For example, think of jelly bean guessing contests at say a fair, where whoever guesses the closest wins. Once you get a few hundred guessers, if you average all of their guesses together, you will get remarkably close to the true answer. Data Scientists use this trick in developing machine learning algorithms often, but instead of using lots of human guessers, we use ensembles of machine guessers to get more accurate results.
Q: A lot of people are afraid of AI. A report by Jobseeker Nation said that about 55 percent of candidates are worried that job automation is going to eliminate the jobs that they're preparing for. Is this justified?
A: There is an old trope, “Technology changes jobs, but it doesn’t eliminate them.” It changes what jobs are but it’s actually better for the economy. I’m not sure we have a reason to believe that it’s going to be any different this time, but there is, of course, a risk that there will be a lot of friction and individuals might not be better off in the short term with the coming technology changes.
I spend a lot of time in education, in particular, the task of trying to provide skills at all different levels - from entry-level, to immersive boot camps, to Master’s programs, and through courses like those offered by Simplilearn. I spend a lot of my free time thinking about what we can do to minimize the friction that comes before everything turns out alright in the end economically. It doesn’t mean that the coming technological changes are not going to hurt, but it will hurt less if people start getting the skills that they need.
Q: When you started out a career in AI, what did you wish you had known beforehand? What would have made it easier for you?
A: I got a mathematics education, so I never took a Data Structures and Algorithms course, which is an introductory level class that a Computer Science major should take. When I first started, I hit the books and read through some of the classic texts and watched online tutorials. I wasn’t an expert on algorithms by any means, but at least I had the basic concepts. The years of practice was what made me better.
Q: Would you say then that courses like the ones you’ve developed for Simplilearn would make a good supplement to a solid math education? Wouldn’t such courses be helpful for both people starting out as well as mathematics graduates?
A: Yes, I think that online services including the ones at Simplilearn are one of the best use cases for filling those gaps, because people that come from a math background will probably need to fill in gaps on all sorts of aspects of coding and engineering as well as working as a developer. People who come from more of an engineering background and are very good as developers may have other gaps to fill in, say, with mathematics.
I see this all the time in my classes at Berkeley. The students who have engineering backgrounds can usually get through the machine learning classes more easily instead of beating their head against the wall with debugging. The mathematics students kind of struggle. Often that eventually flips. When you run into tough problems, not having a mathematical literacy makes it challenging to get unstuck when you are solving hard problems on the job, even though you could probably debug more quickly. So you really need both.
Q: Other than AI engineers, who would benefit from doing a machine learning course? Is it just for the developers or could other people like marketers and finance professionals or executives benefit from AI?
A: That’s another great use case for these more a la carte education opportunities like Simplilearn provides, because someone who is working as an engineer or product manager might better understand what other parts of their business are going to be using machine learning for and why.
You can create these very potent partnerships if you have a data science team that has a combination of engineers and machine learning experts. That would mean figuring out ways for people who aren’t engineers to speak the same language and be comfortable discussing the same machine learning topics without necessarily needing more grad school, which is a great use case for a la carte courses.
Q: How can businesses prepare themselves for bringing machine learning technologies and skills in-house?
A: One option is to start building from scratch, but often it might be better to prove out the values first. That way you have momentum before making a huge investment for very high-value resources. Coming up with some smaller proof of concept opportunities to actually see how machine learning can deliver for particular use cases is probably a more successful pattern compared to building an entire organization from the ground up.
Something that is also a common mistake is hiring a bunch of machine learning experts who are very expensive before getting the data infrastructure and the data accessibility in their organizations in order. That’s something that is often a big challenge and then what ends up happening is a low or delayed return on investment, or a hacked together system which leads to bad engineering for years.
Q: How can AI be used to improve the performance of existing business processes?
A: There are always recommenders and search relevance. There are also algorithms that we’ve built for predictive maintenance, fraud, and anomaly detection. All of those are good examples of the kinds of use cases that you might leverage.
Machine learning and deep learning are now ubiquitous, and consumers use and benefit from these algorithms every day without even realizing it. For example, every time you use Google search or search for products on eCommerce sites or any situation where you are exposed to a lot of products and a lot of users of those products, you have an opportunity to use machine learning to effectively match those up in the right way, immediately or over time.
Advertising is a great example. Information about users and products can be leveraged in order to suggest products that they might be interested in. Nobody likes advertisements, but at least if you’re only served things that you’re interested in, it’s less frustrating.
Q: Tell us about the talent shortage in machine learning? Is the industry hurting for people that are skilled in machine learning?
A: This wasn’t the case five years ago, but for now there are plenty of different options for skilling up in machine learning. Almost every university in the world has a Master’s program in Data Science. So today there’s actually not much of a shortage for people who are just training. For people with several years of experience, that’s still a shortage. If you’re coming out of a program like Simplilearn, then figuring out ways of building things and showcasing those skills is probably the best way to get to the top of the pile.
Q: What AI courses could someone take to really boost their skills even if their job is not directly related to machine learning?
A: An engineer might learn quite a bit of machine learning to work well with a machine learning developer. Similarly, such courses could be great for a marketer or a product manager who wants to get a little bit more integrated or know a little bit more about what could be done with their product or what could be detected by their analysts.
Q: Great, well thank you again for your time Mike, and for all your advisory help with the Simplilearn AI and Machine Learning courses. We look forward to hearing your fireside chat with Simplilearn for our learners and other curious listeners from around the world.
It’s clear from this interview with Mike Tamir that regardless of what you do—as a business, jobseeker or consumer—AI will play a big part in your future. Whether you’re seeking to start a career in AI, or you want to gain machine learning skills to apply to your current job, or you just want familiarity so you can better interface with developers for bringing machine learning applications into your organization, Simplilearn can help. Our AI, Machine Learning and Deep Learning courses, developed under the guidance of Mike Tamir himself, can prepare you to take an active role in using AI to set yourself or your company apart from the competition. To get more intelligence about our AI curriculum, visit our website, or call us anytime at 844-LEARN-88 (844-532-7688).
