James, who are you?
I'm a medical student in England, almost finished now, got about a year left. A few years ago, a colleague and I started looking for more interesting ways to study because, typically, the way most students study is with really outdated methods. They're not really taking advantage of technology. We know from research that there's far better ways to be doing it, so we developed a platform called Synap which was made to help ourselves and our classmates to study in a more effective way. We built the basic website, and mobile app, but basically got students writing their own test questions, publishing them online for people to share, and practicing them in short bursts over an extended period of time rather than trying to fit all their studying into one go.
This ended up getting really popular in our year and with other medical students, so we opened this up to other people and it ended up getting really popular. What we've done now, over the last 12 months, we've actually built that from just some sort of idea into a start up business. We've gone and secured funding for it, we've got a team of developers and other people behind it, and we've started to incorporate AI elements into that. What we do now with Synap is, in addition to getting with students to engage in their learning, we use elements of AI, machine learning, big data analytics, whatever you'd want to call it, to look at the way students are revising, identify what their strong areas are and their weak areas, and suggest new content for them to study, basically offering a personalized, algorithm based, studying service to any student who signs up on the site.
Can you say more about Synap’s AI?
We didn't start Synap as an AI company at all. It was only fairly late into its development that we realized that there was an AI element to this, I think. The thing with AI is that, because it's such a broad definition generally, artificial intelligence, when you mention it to people on the street they don't think it's really cool, sort of a Skynet, take over the world, sort of thing, which is one end of the spectrum I'm sure we'll end up talking about. An AI could just be something that learns what your name or what time of day you're most likely to respond to a notification. You've got a whole range of different sort of, I guess you might call, IQs, within there when you say AI. We just started looking at the data that we've got and started realizing that certain students were very strong on, in the case of medics, the anatomy, but very weak on, perhaps, their prescribing. That meant that we could select specific tests for them to help them improve and then we could build that into algorithms and that's when the AI element started.
We started using the term AI to describe Synap around the time that we were doing our investment rounds and really thinking about the future of the business. What we do at
the moment is, there's two sides to this, firstly, we've got algorithms that are based on research into how the brain and memory works.
Spaced repetition is a formula for how you can space out your learning over time, so you can learn in different chunks. The idea is, if I read this significantly different piece of information multiple times throughout a couple of weeks, you're sort of hacking your brain into recognizing this as important information. Part of the AI element focuses on internalizing these algorithms and building it with technology so our algorithms will use the basic spaced repetition research to find out when we should be sending a student a specific piece of content and what content we're actually sending. The second part it using AI at an individual student's or the peer level, to find out X for information about them and use that to further personalize their learning. This is where learning the student's own revision cycles comes in, for example, what times they are most likely to engage in tests, which subjects they are struggling with, what sort of level a question they're good with, and gaining that big data in our analytics site.
How did your company obtain the expertise to do AI? Where did you find your AI hackers and AI algorithms?
Omair and and I have been developing for a while. I was doing software development in various forms for about 7 years including a bit of game programming which, obviously, introduces you to AI. Omair's been somewhat developing for about 3 years now. Until very recently, we've handled all of the development. We've now got some external developers on board, but most of the actual AI stuff is still done by us. That's because compared to what some companies are doing in AI space, what Synap is doing right now is, in some ways, quite simple. Again, we're just using this data, building these algorithms based on existing research.
In the future, what we'd like to do is hire AI specialists. That's when we'll go into further investment rounds. We want to really squeeze these algorithms for all that their worth and start using technologies like IBM Watson and Facebook's API platform. When we do that, we'll be looking for someone who's done an actual degree in artificial intelligence or something along those lines. As to how we actually get them, it's not a bridge that we've had to cross yet.
How do you measure the quality and performance of your AI?
The way we can measure it is on the results it's having on the students. It's difficult, in a way, because obviously you can't effectively measure how good the code is other than basic checks on syntax and stuff like that. You've got to put all the code in design systems up front, and you're measuring it right at the other end. Obviously, the ultimate measurement you'd want to go for would be how well the people using Synap do in their end of year exams or in their careers further down the line. A lot of the time, obviously, that takes way too long to do. You're introducing a whole bunch of confounding variables, meaning that it will be hard to do accurately. It would be hard to find the students and say, "This student's got 39% better results than these other students. We're attributing all of this success to Synap." There's lots of other differences.
The way we're going to do it is, we're going to run a randomized controlled trial with a number of universities in the UK. We're basically going to offer a question bank to, let's say, 100 students studying a particular course, and we'll randomly allocate them so that 50% of them will use Synap with all the personalization and spaced repetition and the algorithms that come with that. 50% will still have access to the questions, they just won't have access to the sort of intelligent algorithms that Synap provides. That way, we've got two groups who are as similar as possible at the starting point and we're going to track them over the course of a month or two, give them a pre and post studying exam and compare the results that way. That will give a good indication, hopefully, of what sort of impact the algorithms, if any, are having on students' education.
Can you share something awesome that your AI has done or learned that surprised you or you think would surprise most people?
In the early days, Synap focused on medical students and there's a few reasons for that. Partially, it's because that's our own background, so we understood that market. We knew what the needs were and we had routes getting in there, and partially because medical students were a really good testing ground for any educational platform anyway, because they're a tightly knit community, they've got lots and lots to learn in a very short period of time, and there's quite a few of them in most countries. One of the things we picked up early on, there are students in medical school who do
really well in years 1, 2, and 3. Then suddenly they sort of drop off the end of the cliff when it comes to different years, so in year 3 or 4 when the content changes slightly or when it gets harder, they'll go from getting A grades to getting D or E grades, which isn't what you'd expect. You'd expect students getting better and getting worse, even, but you'd usually, in an A student, you'd usually see them getting a bit of a drop off first. You'd see A, A, B, your B would be a bit of a warning sign that something's not right, and you could maybe target that student with if you're in university, extra support, mentoring, making sure they're okay, and then from the B you could either correct them back up to an A, stabilize them on the B, or they might go down to a C. That's quite unusual to see people going, as a cohort, from As to Cs.
What we found when we were looking into this that was really interesting is that the reason a lot of these students were going directly from an A to a C is that they had never been good at answering a certain type of question that comes up more frequently in real life and in the later years of the course. In years 1 and 2, you get a cohort of students who manage to get A grades by answering the very factual, simple, recall-based questions, but they never really engage with the ones that require higher level thinking, critical analysis, and that kind of thing.
We managed to pick that up in our data, and that's something that is a part of our algorithms now in terms of figuring out where the student's at and where they'd like to progress to, what they might benefit from, and that was just one thing that we picked up in the early days before we could really call ourselves an AI company. Now we've got all these analytics in place, we're really looking forward to what else we can find out about medical students, what we can find out about math students, or physics students, or whoever it is, and hopefully identifying these trends early so you can correct them.
When you talk to investors, do they care about your AI or do they care that you're solving a customer problem?
Investors are all different, obviously. We went through the crowdfunding route. There's a lot of interest in it, because I think everyone sees the potential of AI in a whole range of different industries, including education. If you brand yourselves an AI company, that's definitely going to get some interest there. I think investors are interested in startups with something they can protect, because it something that's missing from a lot of them today, and it's something that, obviously, if you have the best idea in the world and you've got 1,000,000 people using it but you've not got anything that you can protect, then you've just told 1,000,000 other people about your idea. All someone needs to do is slightly improve your product or make it look slightly better, do it slightly cheaper, or offer it to a different market, and suddenly you're going to have a major competitor on your hands that may run you into the ground. That's probably the biggest thing that investors are worried about these days.
What AI does, is it gives you some degree of protection because AI is incredibly difficult code to write, especially when you're getting into the high levels. The stuff that Facebook, Google, and IBM is doing now, they're literally hiring all the AI programmers that they can get their hands on so they've got almost an exclusive access to this new resource, because it's not something that's as widespread a skill as non-AI programming is right now. It's harder to do, it requires a lot more in depth insight into maths, and that kind of thing and. That's one reason investors are interested in AI, because it's something that they can protect.
In our case, it's partially the AI and it's partially the community that we're building behind it. The larger Synap is, the more useful it is to other people, because the more access to data we'll be able to get, the more tests we'll have in our system, the more communities we'll have of students in different universities across the world, so it gets easier and easier for us to acquire people and it gets harder and harder for competitors to build something. Also, it creates a positive feedback loop with our AI. When we've got 1,000,000 people using Synap, when we've got 5% of medical students across the world using Synap, it gives us unparalleled insights into how these students perform, what they struggle with, what they're good at, what sort of things they respond to, and how you can improve them and it's a load of data that, once you've got it, it's hard for someone else to compete because you're having these insights come out all the time.
Why is there so much excitement about AI now?
It's got that Hollywood dimension to it. You've got the Terminator sort of movies and all the other things we love to watch are about AIs, and I think it's something that fascinates us because it gets down to a really interesting question of just how intelligent can computers get? It comes on to the sort of Turing test and the consciousness question of whether that means that they could ever think. I think, no matter what job you're in, people should be interested in AI because the question on everyone's minds got to be, "How long until a computer can do my job? How long until it can do it far better and cheaper than me? What does that mean for my job?"
AI means huge things in terms of how we're going to have to structure society in the future. Elon Musk, Stephen Hawking, and all of these guys who are in positions where they should know a lot about AI are very worried about it. They think that, without the right legislations in place, this could be something that is far more catastrophic than nuclear weapons or any other weapon of mass destruction that we could think of.
On the other end of the spectrum, it could totally almost save the planet. AI could could deliver exponential increases in medicine and other industries. It could alleviate the need for sort of dangerous or tedious work from humans, and it could put the right structure in place to make sure that that isn't abused, either by people or the machines, themselves. We're going to be living in a very different, and hopefully more pleasurable, world, for lack of a better word.
Why are people getting excited about it right now is a good question. Either it's because people are getting excited about it because they sense something happening, at the moment, that means that maybe now's the time that AI is actually going to take off. People have been wrong about this before. I think the technologies that are about now, things like the computing power, cloud computing, the things that IBM's already done with Watson and other developments happening in the field are really interesting. It probably means that within the next 5 to 10 years the major advances in computing will be major advances in AI and it's going to totally transform industries.
We've already seen various industries totally transformed or even decimated by just raw computing power. You've seen the taxi industry totally changed because of cloud computing and because just the ability to track locations, which is fairly algorithmic. Obviously, music, books, all the rest of it's changing. Assembly line work, factory work, and that sort of thing is changing due to computers, and that's all changed because of the computing power dime that is not all that intelligent, it's just telling the computer to do process A to B under these conditions, keep repeating, if this happens then do this. It's sort of linear, A to B traditional computing.
With AI, I think we're going to see the same changes happen in more professional and creative jobs. They'll see the same changes that happened to factory workers happen to teachers, doctors, lawyers, all of these jobs that we, at one point, thought were protected from the computer revolution that are now going to change significantly with AI. I think that's exciting for people. I think it's scary for people, but either way I think the fact that people are taking an interest in this now is interesting.
Where do you suggest that people go to find more information about AI and AI startups?
There's quite a lot of lists being compiled. We were featured in one Business Insider. Other than that, I like to read the blog called Wait But Why. The guy on that writes really long form, very well researched, articles on a whole range of things. He did a series on AI and the coming "AI revolution." as he called it.
Other than that, IBM's got a guide that I sort of follow in AI field with what they're doing with Watson. IBM's got a new processor that's totally different to any CPU that's been made before because it's based on a totally different architecture that is based on how the human brain sort of connects and synapses with itself.
Make a prediction about the future of AI startups.
It's really difficult to do because the only thing that you can predict is that things are going to change drastically in ways that we couldn't really predict right now.
It's going to change more drastically than the sort of computing revolutions of the past and in totally different ways. Compare what we're doing with AI now to the early stages of the internet, when people were messing around with HTML and posting university websites and that sort of thing. You could sort of predict that, in the future, websites would get better looking, faster, and more in depth. What you probably wouldn't have predicted is Uber, Skype, the fact that we're having this conversation now, the fact that 1 billion people in the world were sharing information through Facebook, that there was this thing called Twitter that's being used to coordinate revolutions in the Middle Eastern countries. I think the predictions in AI are going to be even further afield than that.
The guy that writes the Wait But Why blog made this great point. He said, "We're standing on the edge of a change in human history that's unlike any that we've seen before." I think that's true. We're just going to have to go along for the ride and see what happens.
If you're interested in Synap and what we're doing, which is basically building an online platform for students and professionals, using AI to help them learn more in less time, then just check out our website. It's totally free. It's for anyone who needs to learn a lot in a short period of time and we'd really love your feedback on it.