David Yakobovitch

Working together to solve some of the biggest challenges as humans that we’re facing, which will create brand new industries and augment our daily life and the relationships and interactions we have with technology.

David Yakobovitch

This is HumAIn, a weekly podcast focused on bridging the gap between humans and machines in this age of acceleration. My name is David Yakobovitch, and on this podcast I interview experts in sociology, psychology, artificial intelligence, researchers on consumer facing products and consumer facing companies to help audiences to better understand AI and its many capabilities. If you liked the show, remember to subscribe and leave a review.

Hello everyone, and welcome back to the HumAIn podcast. My name is David Yakobovitch, your host, and with us today is Sameer Maskey, a good friend and colleague of mine in New York city. He’s the founder of Fusemachines and the new Fuse.AI Fellowship, as well has been teaching with Columbia University for over the past six years.

Welcome to the podcast, Sameer. 

Sameer Maskey

Thank you for having me, David. Glad to be here. 

David Yakobovitch

Thank you so much. And as a fellow colleague in New York, we’re both involved very much in AI and research and education. I know this year that your organization’s been launching this new Fuse.AI Fellowship, and I was part of one of the first cohorts with the Fusemachines Fellowship. Why don’t you share with our listeners a little bit about what this is about? 

Sameer Maskey

Sure. So we recently launched the Fuse.Ai scholarship program, where we announced colleges for 10,000 students from underserved communities around the world, and here in the United States. The motivation is we believe talent is everywhere and there are equally talented people everywhere who may not get opportunities to get into the really good schools and get into really good PR training.

So, for a lot of students who have not gotten an opportunity, we have created a scholarship program where we provide the scholarship to get trained up in AI here in the US and from developing countries. And as a target, we have announced a scholarship for 10,000 students to be able to get trained in AI.

David Yakobovitch

It’s so amazing what your organization is doing to, in essence, bridge the gap between humans and machines. Can you tell me more about some of the countries that these students are from? And the reason I ask that is everyone today thinks: Oh, AI, it’s a race between the United States and China, but is it going on elsewhere?

Sameer Maskey

At the highest level, from the density of AI engineers, I guess the United States and China are the ones who have a lot of AI engineers. For us what we are focused on is more than any particular country, we are more focused on where are the underserved communities who could, where if we are able to chain even 10 engineers, it would fundamentally make a difference in that community.

So we are running these scholarship programs in Nepal, In the United States, Dominican Republic and Rwanda and we plan to add more countries in the coming months and years. 

David Yakobovitch

It’s great that you’re adding more countries, and part of it’s been the mission that I’m smearing this team’s work on is democratizing AI. I’ve actually attended the democratize AI meetup in New York. And it’s fantastic to see that AI should be a global opportunity for all humans to participate in. Particularly with the new Fuse.AI scholarship, the material is very in-depth. The focus is on fancy algorithms in different Python packages, very technical, so that someone could be hands-on and do damage with this.

They could actually create solutions, create production and solve problems. And another question. I’m interested in, on one hand, the Fuse.AI scholarship is teaching all the code and the fundamentals for engineers to be involved in the industry. And on the other hand, the products offered by your startup, Fusemachines, is very much about how can your organization bridge the gap by building AI solutions?

So the question really is: where do you see the industry going? Is it a code based where everyone codes philosophy, or also companies bridging that gap for those who may not have those technical skills today? 

Sameer Maskey

There are a lot of companies that are trying to build AI systems and there’s not enough AI talent. So, as AI gets commoditized more and we build platforms, there will be companies that provide solutions with software and services such as us, where we come in and say: This is what you need. This is what you are looking for. This is the problem that you’re trying to solve. So here is, a few pieces of models that we already have that you can license, and here’s a set of talented engineers that have a control scholarship program that can work with you on top of those tools to build whatever you need. 

So there’s definitely a need for that, and I see a lot of companies using our service accordingly. There’s also a trend of more and more AI engineers trying to learn AI tools and the fundamentals, to be able to provide solutions to the companies or build products for the company. So, in general, companies could look in two ways of building AI features into the product. One is be able to use some of the platforms and, also for service providers, to build on top of those, or they could hire AI team members to build everything from scratch. 

David Yakobovitch

So it sounds like there’s many directions that you can go if you’re a non-technical person today, you could learn the codes, you could learn the systems, you get higher in the organization, like Fusemachines, to help you create that solution. And the industry is becoming more these productized services, these service offerings as if you will. What are some other trends that you’re seeing going on in industry today?

Sameer Maskey

There are pretty great trends on the use of particular kinds of algorithms, I guess. So I’ve been in machine learning for 20 years seeing ups and downs of main different algorithms, being popular and then not being so popular. So, there were times when support vector machines were on the rise and then there were conditional random fields, X amount of models, neural networks at some point. And now neural networks have come back in full force with deep learning. But one thing that I see is deep learning systems being applied across many different problem sets within data has been able to win over most of the other sets of algorithms. So one of the trends is that more and more people are using deep learning solutions for their problems in computer vision, NLP, or any other kinds of prediction.

David Yakobovitch

If I’m a consumer and I work in a non data science, non AI industry. Let’s run a quick exercise, I want you to teach me like I’m five. What’s the difference between machine learning, ML; artificial intelligence, AI; and deep learning?

Sameer Maskey

So, that’s a common question. The way I would describe it is, if you are building a machine  that can take data and reason on it and it does some sort of decision as a whole, that’s an AI. Particularly if the data is related to perception data such as vision, such as self-driving cars, are talking machines like Alexa and what not, they are very visible from the perspective of what people think about AI and when we talk about the phrase Artificial Intelligence. Most of the AI systems these days are statistical models, which are essentially driven by data and building statistical models and doubles as data to do predictions, and that is primarily governed with machine learning. 

So in that sense the foundations of AI or the algorithms, the main sets of algorithms are the foundations that are inside the AI systems and you call it the brains of the AI systems. Those are all machine learning algorithms, and there are many different kinds of machine learning algorithms, as I just had mentioned before, max mentoring models, decision trees, and you name it. There are many, many different ways to build prediction models, one of them that’s very popular now is neural nets, especially with many layers. And different types of neural nets that are being used to solve many problems, which is essentially deep learning. 

David Yakobovitch

That is a great explanation. Like I’m five, although I’m not five, so this is fun. So I really appreciate it and to have a caveat there. So the term neural networks, in the deep learning space, it’s talked about all the time. But most consumers don’t know much about this. If I hear neurons, in the brain, and I’ve heard some miss firings about this neural network. Is it really like the human brain or is it something artificial? Can we debunk it once and for all? How do neurons work? Are they like humans or are they not?

Sameer Maskey

You mean neuron nets? 

David Yakobovitch

That’s right. 

Sameer Maskey

Well, that’s a hard question to answer. I’m not from neuroscience. My wife’s a neuroscientist, so probably she is able to give a good answer to that.

David Yakobovitch

We’ll have her in the future podcast . 

Sameer Maskey

Even though you will net in the sense is trying to simulate how neurons turn on and turn off. I don’t think we, as humans, understand how brains work well enough to be able to really be able to say that neural nets replicate how the human brain functions. Neural nets is just essentially a network of nodes that you have different parameters that demands pass through it. So I’m cautious in being able to try to equate how neural nets work closely with how human brains work, because we, as humans, actually have a long ways to go before we actually understand how cognition works in humans. 

David Yakobovitch

We see new shows coming on Netflix like Altered Carbon and these other futuristic AI shows that are showing cognition and human memory, but it’s still maybe yet to be determined. Let’s give it another, I don’t even know if I want to put a timeline on it. Should it be 5 years, 10 years, 20 years, what does that look like? 

Sameer Maskey

For a small child…. A small child can within three years with a decent amount of data, but not a lot of data, is able to fully converse with humans. I have a four and a half year old daughter and she understands everything and understands all the context, is able to make right decisions and be able to talk  very intelligently. While we are yet to have a machine that can sit down with you and  talk for half an hour. We’re far from it.

David Yakobovitch

I know that in 2018, Amazon had a challenge with universities where it was their grand touring test, if you will. Can they break the machine, break this chatbot, from different universities for a task to: can I make 8 to 10 minute chatbot that, as a human, could say commands to this Alexa device and keep that dialogue going where it made sense.

What I thought was most fascinating about it, I looked at some of these algorithms and the conversations they used, a lot of it was still human train. There were sentences, and scenarios, and decisions and rules that all these researchers and their teams were creating. It wasn’t like the machine was automatically creating these from the get-go.

Sameer Maskey

Amazon does this yearly challenge on this. The challenges on dialogue systems have been going on for many years at this point, before all the AI hype, like more than 10 years. It’s a difficult system to build. Right now it’s a lot of data systems that are pre-trained on,  all the data that’s collected from scenarios, like you mentioned. It’s not like the machine is reading five books and starting to converse about it. 

David Yakobovitch

That would be something, I can not wait until I can give all the books that I read each year to a machine and then they could tell me book reports and summaries, so I don’t have to read anymore, that would be so interesting. But, oh my goodness, how is that going to help our world? So I digress. But for things that are helping the world, I know in our past conversations you’ve talked about some of the exciting projects that Fusemachines has been a part of. One of them, including  drones that are delivering medicines and experimental capacities in Nepal and other very advanced projects. So is there some of the research or projects that you’re working on today or your teams are working on that you’d like to share with our viewers? 

Sameer Maskey

One of the systems that we were building, as you had mentioned for medical dental medicine delivery in Nepal. Nepal is very mountainous and doesn’t have enough roads. So a lot of people who live on top of the mountains and villages don’t get quick access to medicine, so we had built drones for delivering medicine. We are also working on various language recognition systems, speech recognition systems for other languages. We are doing a lot of advanced research on building dialogue systems and question answering systems.

David Yakobovitch

On the languages, most of us around the world take a second language. For me, my second language, I was learning in high school and college was Spanish, and then I picked up a little bit of a French and some Mandarin, dabbling in different languages. But, as a human, one of the challenges is to take so much discipline and commitment to pick up languages, and is it worth the effort. So maybe a question I have for you is: how good is that technology happening here for recognizing speech and real time or translating that in conversations? 

Sameer Maskey

I actually worked on the speech research translation for almost five years in the IBM Watson Research Center before. It’s a hard problem, but it has come a long way. And these days, especially new systems work quite well, at least in a non noisy environment, and with single speakers. And machine translation is starting to work pretty decently as well. So there are switches to switch translation systems on your phones that can let you get by when you traveled to a different country.

There’s still not a full system that could translate as good as humans, but some of these translation apps, at least, allow you to travel to a country where you may not, where the country’s main language may not be English. 

David Yakobovitch

So if I’m a high school student or college student today, should I just give up on learning a second language? Should they be like: Oh, the technology is going to catch up. So, I don’t need to learn Spanish.

Sameer Maskey

No, I don’t think so. I don’t think it’s going to catch up fast enough that it would make language learning of a foreign language redundant in the next 5, 10 years. So if you’re listening right now, maybe if you’re listening in 20 years from now, maybe it would be a different answer, but at the current moment in history you’d still get a lot out of learning a second language or third language.

David Yakobovitch

So I guess to take that even further to jobs in the future of work and how industries are changing, a lot of diplomats work in being foreign translators and so forth, and there’s companies that do transcription and language translation. So, since you’ve done a lot of work in the speech recognition space: Is there any new breakthroughs that you’re seeing that maybe consumers could expect in the next couple of years to hit the market?

Sameer Maskey

The language translation systems are switched with translation systems. I mean, it is continuously improving. I don’t think there’ll be like an aha  moment where it suddenly starts to translate everything perfectly tomorrow, but it is working  in progress and it is improving quite rapidly. 

David Yakobovitch

That continuous improvement that we could think about is for those of us who’ve owned Android or Apple devices. You can think back to when you own the device in 2008 and how good was the audio for doing tasks versus 2012 versus 2015 versus 2018 versus today. And even though Apple and Google and these companies don’t tell us, It’s improving, every time that there’s a software update. Perhaps some of that is this improved speech.

Sameer Maskey

Yeah, it is. And there’s so much data that some of these companies have collected  with all the Alexa CD and Google home devices all over the world, everybody talking to it all the time. You could imagine the amount of data that is being collected that could potentially improve the system  more and more as the years go by. 

David Yakobovitch

So what I’m hearing from our conversation today is it sounds like humans and machines are going to work together more. And yes, technology is changing, it’s more gradual over time and it could augment human capabilities. But the question would be if I’m starting to learn today, what’s right, should I be getting this college degree? Should I go through a bootcamp? I’ll do Fuse.AI and just jump into the workforce, or currently I’m working today. How concerned should I be about my job?

Sameer Maskey

Some jobs would be automated, that’s just going to happen. If somebody is doing a very repetitive task every day, then you should worry a little bit, I guess that repetition could be automated by machines. But there are also a lot of other tasks that are going to take quite a while before machines can be automated.

A lot of people, at least, should think about if they should  learn some level of AI, machine learning and computer science technologies, as computer science is set into all different industries and verticals and business applications. And now, on top of computer science, the AI layer is also being added to automate more stuff.

So people should think about how they get more familiar with the new different technologies in computer science, particularly in  AI. But some also should think about if their jobs could be automated,  but I don’t think everybody should get scared that the job’s going to get replaced tomorrow.

David Yakobovitch

Little too much hype, It’s a little overblown.  But the classic case we could look at, for example, is the new Amazon Go technology, which is creating cashier less stores. There was one launched in Seattle, Washington. There’s one being launched in Manhattan, in New York state. The goal is that there’s going to be just in time inventory and a store manager or two, or maybe a clerk, but you can just pick up and go.

So, these are individuals, who is your standard cashier? We’d have to profile that person and see how they transition them. But then the next question is: Okay, well, if it’s going to be automation, how can you be part of the automation? Can you build the next AI startup? Why not be part of that bridging that gap? 

Moving forward, being someone who’s very involved in applications and research in the industry. What do you see as any of the new breakthroughs or something that’s on your radar that you think is a really interesting technology ,that over the next couple of years, it’s kind of like the underdog or being underrated right now, but you think people should pay attention to.

Sameer Maskey

There’s a lot of computer vision algorithms that are making large strides on what you can do and you could find some of these machine learning conferences like ICML and NeurIPS and whatnot. But one of the things that would be transformational, and it’s not necessarily an underdog, but it gets talked about quite a bit is self-driving cars. From the perspective of the impact of AI, it’s becoming more and more real. And when we have more, a lot of self-driving cars on the road, it would be transformational for how we travel. And that’s one of the things that I’m quite excited about from the perspective of the impact of AI on our lives.

David Yakobovitch

So that’s a signal, to our viewers, that Samir’s looking forward to in the next couple of years, and I’m looking forward to seeing self-driving vehicles in New York city. It would be so interesting to hail a self-driving Uber or Lyft, and it should be such a different dynamic. Like  I can’t even imagine yet today, like what would it be like? There’s no human. Would there be more space in the vehicle? Could I conduct meetings? It’s going to be a whole different type of dynamic that I would imagine. 

Sameer Maskey

Yeah. It’s exciting to think about that future. 

David Yakobovitch

That future is becoming reality. Self-driving cars are being piloted on the road in several states: California, Arizona and Pennsylvania. Depending on legislation in many factors more will happen in 2019 and beyond. Thinking of beyond and future predictions, everyone has something that they believe in that others don’t, it’s being contrarian, against the grain. This is something that is true in AI, but most people don’t. Is there something that you’ve looked at and explored and you also think, people should be considering this because if not it could become more important over time.

Sameer Maskey

The question is: if there anything that I believe that could be potentially game changing, but a lot of people have thought about, or it’s the other way?

David Yakobovitch

Yeah, it could be. Do you have a moonshot idea? Don’t get me your billion dollar idea. I don’t want to hear, don’t want to let the cat out of the bag here today. But yeah, anything contrarian, that you think other people would not agree with you or it might be split. Like half of  the AI world says this, half say that, and you’re on one of those fences. 

Sameer Maskey

What I would have to say is the usefulness of chatbots in their current form. A lot of people say that chatbots are going to transform how, for example, customer service could be served and what not. There’s too much hype. And like I said, dialogue systems are extremely hard to build. We still don’t have systems that can really understand humans from the level of being able to just converse and understand everything. I would say though, some limited versions of chatbots are useful for reducing the customer service load, it is over-hyped,  and it’s not at a point where chatbots can completely replace customer service reps. 

David Yakobovitch

I agree as well. We’re even seeing this year at CES 2019, there are now examples of video chatbots that are having decent dialogues. But the question is when can a bank teller be replaced? When can a customer service agent on the phone for your utility company be replaced? and we’re not there yet, and maybe we never will be. Maybe it will just be an augmented report. The chatbox facilitates gathering the information and does 80% of the process, and then the human comes in for those advanced queries to help solve your problem more efficiently. 

Definitely very interesting and only time will tell. I’m excited to see the changes in chatbots.  Again, we’ve talked so far on today’s episode about how Amazon’s done work there, even IBM’s done a lot of the work there with Watson, the new data platform. It’s very exciting to see the neural networks and the software. There’s so many entrances in the space of AI today. In one of our other episodes we’ve featured the founder of Welcome.AI talking about how just the industry is growing so fast. Is it going to continue? We’re in 2019 now and there’s a lot of growth. Some economic experts are calling for a slowdown for certain segments of the economy. My question, which I’ve been thinking about a lot recently: is that impacting tech or are we isolated? Are we living in our own bubble? 

Sameer Maskey

That’s a hard question to answer. If there’s a global slowdown on the economy it will obviously translate into a slowdown on the total number of investments that will go into AI related businesses and efforts. So they’ll probably be a little bit of slow down on AI as well. 

But if the global economy keeps trending in the way it’s trending and keeps growing, then probably AI will keep pace with it, or actually outpace it on that is on the growth year, over growth percentage. There’s been tremendous growth over the last couple of years, there’s been a lot of hype. But you should remember the concept of AI winter does exist, It has been paceed twice before. When I was in a lot of hype and it did not deliver what it promised to deliver and then the funding dried up. It’s probably a little different today because there’s a lot of data, the algorithms seem to be able to do well in many different tasks and compute power is also at a point where you can crunch a lot of data. So maybe we do not come into the same kind of AI winter, but if the global economy slows down, I’m pretty sure the level of investments in AI will also slow down a little bit.

David Yakobovitch

Of course, this is all hypothetical. So for our consumers, depending on when you’re watching and listening to this podcast, we’ll see how those predictions do hold true or not. What’s interesting, and one of our key takeaways, is that AI can be applied anywhere to any team, in any country, in your app. And that, just having one or two individuals in your organization that know this technology can change the course of your company and potentially your local economy.

And it sounds like Fuse.AI is part of that dialogue. There’ll be thousands of individuals over the next couple of years going through the program, helping democratize that process. What does it mean to you Sameer, to be part of that dialogue where you are having an impact on countries all over the world.

Sameer Maskey

We are humbled by it and we are glad to be in this juncture of being able to contribute to this transformation that AI is bringing to too many different businesses, industries and countries. We are  excited about it to play a small part in this transformation, in being able to bring about AI talent in all these different places where it could have its own significant impact locally and all over the world.

David Yakobovitch

You’re playing a small part, but it’s a critical part. It’s a critical part to bridge the gap between humans and machines in this age of acceleration. And before we know it, there will be, dare I say, hundreds of thousands or millions of AI practitioners who will be working together to solve some of the biggest challenges as humans that we’re facing. Which will create brand new industries and augment our daily life and the relationships and interactions we have with technology. Yeah, what an exciting time this is.

Sameer Maskey

It is, It is. 

David Yakobovitch

It is fantastic. 

Well, this has been the HumAIn podcast  with Sameer, from Fusemachines. Thanks for tuning in this time, and we’ll catch you on another episode where we’ll bridge the gap between humans and machines. Thank you, Sameer. 

Sameer Maskey

Thank you David

David Yakobovitch

That’s it all for this episode of HumAIn. I’m David Yakobovitch, and if you enjoyed the show, don’t forget to click subscribe on Apple podcasts or wherever you are listening to this.Thanks so much for listening and I’ll talk to you in the next one.