D-RAFT Corporate Demo Day: Startups Entering The Machine Learning Era
During the last D-RAFT Corporate Demo Day, the ten startups:
2. Cervi Robotics
3. Invoice Sharing
4. Quantum Lab
got the opportunity to present their solutions (including AI & digital disruption, big data & machine learning, human to technology, and exponential technologies) in an 8 min pitch to corporate executives and participate in individual matching sessions. The event took place on Thursday, September 22nd, 2016. We asked startup vendors and representatives from the organization team about these trends.
Kevin Kelly was right when he predicted that the business plans of the next 10,000 startups were easy to forecast: »take X and add AI«. Computers that see and listen, think and predict are already making a difference across industries. Artificial intelligence can automate processes, reduce costs and improve customer experience. Corporations need to leverage those machine learning technologies or risk being replaced by ‘smarter disruptors.’ [Tomasz Rudolf, CEO D-RAFT]
How is AI or RPA different today than 20 years ago?
Tomasz Wesołowski, CEO 2040.io: For sure, we now have the technology (measured in computer power and algorithms) that is able to achieve great progress every year. But the most important difference is that AI started to finance itself. A great recent example is about using DeepMind’s work on reinforcement learning to reduce Google’s Data Center cooling bill by 40%.
Igor Sawczuk, CEO Wandlee: The biggest difference is unsupervised learning and the availability of “cheap” GPU power. That’s why we see so many startups rising in the field of AI. I believe this is the main reason that we are entering the machine learning era.
Marc Torrens, CIO & Co-Founder strands.com: From an academic point of view, there is nothing new or revolutionary that makes AI algorithms better nowadays than 20 years ago. The revolution in AI we are seeing these days comes from the quality and quantity of data that is available in many industries.
Vlad Ionescu, UiPath: The history of computing and smart machines is the history of the ongoing increase in our ability to [invest computers with the capability to] perform computations and manipulate information. And where once computers were massive, slow to process and costly, today they have gotten very powerful, cheap and fast – for example, think of the huge leap in computing capacity since the advent of the microprocessor back in the 60’s. With faster hardware speed and larger software applications, information technology has managed to advance exponentially, and we humans have been able to leverage this advancement for various uses in our lives. Robotic Process Automation is, in its own right, one such result of that ongoing developmental process. Just as is Artificial Intelligence, although on a different scale. There are after all very few, if not none, areas of our daily lives that are nowadays not significantly reliant on, or even fully dependent on the progress of information technology. And what RPA does is it relieves us from doing routine, repetitive work by relegating it to software robots that can process it automatically.
RPA as a standalone solution to automate business processes is not at all old. But the first sparks of this technology can be traced back to the 90’s when ERP (Enterprise Resource Planning) systems dominated the corporate landscape and companies used rudimentary automation tools like Screen Scraping (See more here). Then came business process management systems (BPMS) as a more process-focused alternative to ERPs. And to make things even more efficient, companies then favored a new service: business process outsourcing (BPO) which instituted the labor arbitrage model, where processes were outsourced to other geographies in order to achieve quick savings with little investment.
During all this time, RPA was slowly surfacing out of its own boiling soup as a desktop automation solution, a quick and efficient single instance of automation (like copying and pasting data between two applications) but with very promising features: it worked at the user interface level, emulating an employee’s own actions, required no coding, and didn’t intrude on the legality systems. It was easy to implement, intuitive to the user, and extremely efficient in reducing processing costs. All of this made RPA very attractive to the business world. A few years later, RPA was not only on everybody’s lips but had already made substantial technical advances and managed to move from the desktop to the cloud, on server platforms. This was the tipping point for RPA products like ours, which is now a fully grown, enterprise level automation solution capable of streamlining an entire organization’s processes at any scale. It has the technology, it has the methodology, and it is backed by enterprises as a highly profitable innovation.
As RPA products developed for enterprise level complex process automation, they have also started to incorporate more advanced capabilities reminiscent of Artificial Intelligence: cognitive features (advanced computer vision, increased robotic autonomy, and complex exception processing) built into the RPA software to expand on its capacity to meet increasingly complicated business situations. And this is also where we are today within the domain.
Is Big Data Still a Thing?
Bartłomiej Rozkrut, CTO 2040.io: Big Data helped us achieve big advances in machine learning. However, we must go further. There are many fields where we don’t have a huge amount of data and must make proper decisions. Big data will become a de facto standard whenever possible, but the most exciting advances will be in One-Shot Learning and Zero-Shot Learning. Achieving that type of learning will allow us to solve problems more like a human does.
Igor Sawczuk, CEO Wandlee: Big Data was always a thing. Buzzwords come and go, but understanding that data is money and extracting information from data is power is crucial.
Marc: The Big Data approach is here to stay in the sense that it is a practice that will see higher and wider adoption. The Internet of Things (IoT) will make Big Data a must and not just practice for visionaries.
Vlad: Big Data is definitely still a thing. By estimates, the total amount of data stored globally today is measured in thousands of billions of gigabytes, and it’s increasing as we speak. It also represents one of the main pillars of our current economy, because companies rely on big data to get an unprecedented level of insight into their operations or the buying preferences of their clients.
A strong, up-to-date RPA product should have big data analytics embedded into the product. Ours, for example, is based on the Kibana and ElasticSearch technology, which makes it highly potent. Now, with big data analytics, everything the RPA robots do, every activity and every interaction is captured live in the system. For an organization, this translates into a newfound capability to basically record operational performance and find opportunities for improvement, ways to strategically elevate the business and it’s a huge benefit.
What might the future hold for Big Data Intelligence, the post-convergence of big data and AI?
Bartłomiej: Big data is a natural companion of AI – especially for Artificial Narrow Intelligence problems. Big data allows us to achieve superhuman results in narrow fields. Although to achieve human-like AI, we need to learn how to solve problems using less data or even without data, using only the description of a concept.
Igor: Many scary things and many promising things. The scariest (for me personally) is scoring. Scoring is basically using all big data to extract a single value from your life and your history. The promise, on the other hand, is dealing with all the information we produce. For example, YouTube users now upload more than 400 hours of video to the site every minute. Forecasts say that it will go up. There is not enough man power to watch all this content or even to moderate. Big Data Intelligence already helps to fight copyright infringement, and we will probably see full-blown artificial lawyers.
BDI also helps you communicate better. The best example is the latest Boomerang Extension which analyzes your email and gives you a score of “respondability” of your email.
Marc: Big Data Intelligence is about extracting insights from large and heterogeneous data sets. This can be achieved with different computational approaches, from simple statistics to sophisticated AI techniques.
Vlad: This will probably be a world where robots will possess capabilities that were once exclusive to human beings. Clearly the convergence of RPA, Big Data and AI will have a revolutionary impact on business, and of course, in a wider range of areas as well, including research, medicine, politics, and virtually any field of knowledge and science. Investments will be made from every direction, in varying domains of application, thus advancing the technology even further and generating, in turn, an improved capacity for knowledge manipulation and prediction.
YouTube users now upload more than 400 hours of video to the site every minute. Forecasts say that it will go up.
RPA is already starting to incorporate cognitive tools for more advanced decision making. Robots have an increasing capacity to handle complex business situations, to independently handle exceptions, correct errors and apply limited judgment. A company will be able to operate its business with a high degree of automation that is adaptive and responsive and delivers superior performance. There will be a focus on creating more valuable customer experiences, with robots working in the back and human employees upgrading to more creative, more strategic and client-centric positions.
What seemingly impossible challenges could we tackle that we humans are struggling with understanding?
Bartłomiej: Our brains can’t operate with the huge amount of data that AI uses. AI can find patterns which are too complex or too high level for our memory constraints. One of the best examples is understanding our human DNA – AI is helping a lot in that field, and there will be huge advances a few years from now.
Igor: Bioinformatics and healthcare. IBM’s Watson promises that it will make the best doctor entity available to anyone. The IBM Watson team has been building its own AI for quite some time, and they still haven’t shown us the best examples. This is something I’m looking forward to seeing or using.
Bartosz Rychlicki, CEO Quantum Lab: Understanding ourselves, our psyches, our emotions, thoughts and actions in a cause-and-effect way. So basically: what happened, what I was thinking and feeling, what I did and how the world is supposed to respond to it. That’s exactly what we do at Quantum Lab.
Vlad: The difference between humans and intelligent machines is that while both we and these machines are currently more or less able to process a stream of random information from various sources throughout the environment, the machines will do it on a scale that for a human being would be impossible. A computational power so great that it will unravel many wonders for us. And not just for business use. Think for example of all the benefits that big data have brought to our safety and well-being – real-time data analysis used by aircrafts, police operations, in healthcare, etc., not to mention the research value for domains like natural science, social science and so on.
Please choose a few areas where it’s possible that big corporations haven’t already eaten everybody’s lunch.
Tomasz: Big corporations, due to their size, have common problems with managing innovations. They also cannot take a lot of risk because of the market and shareholders. So we have a lot of startups, that are fast and can adapt all new technologies immediately, and on the other side, corporations which may cooperate with these startups to quickly integrate these new solutions into their product or help in finding product-market fits for new ones.
Igor: IoT security is a growing market which will require big data and algorithms for real-time anomaly detection. If you believe the experts, there will 6.4 billion connected devices in 2017. This is a huge market without any clear winner yet.
Marc: One industry in which large corporations still have a monopoly-like control is in banking. With FinTech, we are seeing how this monopoly is rapidly changing as new initiatives by startups start to eat large corporations’ (banks’) lunch. It has not fully happened yet, but it is definitely starting to.
Vlad: Big corporations are ultimately a reflection of our behavioral and consumptive patterns. We represent them as much as they represent us, and together we live in a happily complicated marriage. But with technologies like RPA and AI, what we need to understand is that the algorithms behind these tools are no corporate secret (unless they are of course) or at least that the tendency is for them to become commoditized. What this means is that the instruments for creativity and development will be open to virtually everybody, and it will only be a matter of putting these instruments to use.
To give an example, the technology platforms behind digital assistants like Apple’s Siri and Cortana from Microsoft have been open-sourced, enabling development of other applications based on speech-recognition technology.
There will 6.4 billion connected devices in 2017. This is a huge market without any clear winner yet.
Now that RPA is starting to become a standard in business process automation, there is an increased need for domain experts to put RPA robots to good use through a variety of applications. The technology, any technology, is nothing without the human expertise needed to lend it to practice. And human creativity is virtually endless, so I strongly doubt that we will hit a brick wall any time soon.
Big corporations will continue to provide for our generalized need for comfort, while smaller organizations and entrepreneurs will have a fervent area of opportunity to draw from with these new technologies, and a correspondingly new (and growing) knowledge base to develop and curate.
What might these technologies – AI, big data & machine learning – do?
Tomasz: We believe that by 2040 we will achieve the level of a human mind in AI. So combined with the advances in robotics and biotechnology, almost anything a human can do will be possible. But there will be a very important cost factor. If those technologies will be cheaper than human work, it could change our landscape dramatically not only in business but in everyday life.
Igor: Definitely disrupt our lives. I hope and work for positive disruption, but many already predict massive job losses. A Davos study says even 5 million by 2020. But I think that we just will work less and have more time for our families. Technology makes everything change faster and with a technology that promises to learn by itself or program itself (like Viv.ai), it will make the change even faster.
Marc: Machine learning algorithms are very powerful tools that can predict certain behaviors based on past events. Artificial intelligence is a much broader term that includes Machine Learning. AI, in general, refers to empowering new ways of interaction with machines that don’t involve programming, such as speech recognition with natural language, just to mention one.
Vlad: Smart machines get smarter over time because they learn from what they do. Just like humans. An RPA robot records its own activity, and as it collects more data, it learns how to process exceptions and handle mission critical issues. So they will do whatever we need them to do, at the speed and scale that only they can achieve.
RPA will converge with AI technology and create a powerful combination of tools that will address new business challenges and consumer realities. Nothing is static, and there is a continuum of machine and human activity evolving into new forms of communication. Businesses have an opportunity for real-time interaction with customers, but there is also the pressure to deliver services with added value. The speed and accuracy of these services will be ensured by automation, but the qualitative part of what is delivered remains the job of humans.
Let’s talk about business now. Is having competitors before launching a bad or good thing?
Tomasz: These days, we must know that probably a lot of people around the world are thinking about similar products. The startup ecosystem is strong, and even if we don’t know it, we probably have competitors. Having a competitor is a good thing if you compete in a market with the same rules. The problem is when you compete with startups that have a lot of funding, so they don’t have to focus on cash flow. That’s why we have to focus on our product to be growing fast and also to be technically capable of achieving such growth.
Igor: Competition is a good running mate. On the other hand, my favorite entrepreneur Peter Thiel says “competition is for losers.” And I can’t agree more. Although with competition it’s always easier to navigate through market changes and market education, but at the end, you will always split the cake, and only a random factor will decide that you are either Intel or AMD.
Competition is a good running mate. On the other hand, my favorite entrepreneur Peter Thiel says “competition is for losers.”
Bartosz: A good thing. It allows benchmarking. Oftentimes competition helps to motivate and also validates the direction in which you’re going.
Vlad: Competition is of course intrinsically propelling. If you don’t pass that test, you are not fit for survival. For us, it was definitely a good thing. Those few established RPA vendors at the time had already made their contribution by pioneering RPA as an efficient, emerging technology solution to automate work. And that is something which we are genuinely grateful for. Also, having competitors before launching gave us the opening to build a correspondingly competitive product, with components and features more up-to-date, more intuitive, more inclusive of current enterprise needs, a lot faster and generally more highly performant. For example, our robots have highly advanced computer vision. They see the user interface like Google cars see traffic. The benefit from this is that you can automate more processes with exceptional flexibility and precision.
What’s your strategy for acquiring enough data to test a machine-learning product?
Tomasz: Our product works with historic data which helps us adapt to the user faster and learn on huge amounts of data. We also focus on an omnichannel philosophy – we acquire data from many points of interaction with the user. We design all interactions with a feedback loop which gives us important, relevant metrics continuously used to improve our machine-learning models. That’s why we can achieve results shortly after the new company’s onboarding process.
Igor: It always starts with the 3 F’s – friends, fools and family. The amount of data is crucial in unsupervised learning. You can always hack your way in by using Amazon Mechanical Turk or partnering with researchers. In our case, we always choose the biggest player in their field and talk to them. In many cases, big corporations don’t have the talent to do big data intelligence on their own and need us – startups :)
Marc: Ideally, you need three sets of data: the training set, cross set and test set. The training set is used to build different models, the cross set is used to select the best model, and the test set is used to generalize and evaluate the prediction error of the selected model. The size of these sets should be determined by iterating through the process until the final prediction error is minimized. The important thing is to take a very methodological approach and avoid any type of bias when creating the data sets.
We believe that by 2040 we will achieve the level of a human mind in AI. So combined with the advances in robotics and biotechnology, almost anything a human can do will be possible.
Vlad: UiPath currently offers a Community Version free UiPath Studio license and training for developers, individual projects, or research and education initiatives. This is a pioneering initiative to encourage not only big companies to be enterprising with automation but anyone who wants to create something and be productive with our tool.
For our company’s innovation roadmap, we are also focused on developing cognitive tools that will enhance the capability of our robots to autonomously “learn” and adapt business process automation rules, analyze large structured/unstructured data sets, and thus extend RPA application from clerical to judgment based activities.
With machine learning, how do I know it works?
Tomasz: That problem depends on the layer we are measuring. Low level – e.g. classifying documents are easy to check. However, on a high level and human interaction, it is all about the feedback loop and business KPI’s. So we need to rely on checking the algorithms using humans to improve their usage.
Igor: The simple answer is statistics. If you are good at math, and you understand the subject of your work, you could easily estimate the quality of machine learning. Still, it is a process of constant improvement. And there are plenty of papers about improving “only” 2% points.
Marc: In general, you know ML works when testing your model on a test set of data, as the last step in an iterative process. This final evaluation gives you the error of prediction with respect to the data in the test set. You need to reach a margin of error that is acceptable for the specific domain and problem you are solving.
Vlad: Machine learning works when you manage to make sense of the unstructured data that is collected from a variety of sources. And although we do not yet implement machine learning per se in our product, we are adding digitization to RPA, which leverages various technologies like OCR & business rules to format unstructured data. The result of this is that it increases the automation scope & analytical capabilities of the robots.
What about machine learning do you feel people just don’t understand?
Bartłomiej: What that machine learning model thing is and how it is possible that it does magic. We must communicate with users in simple words. Computer algorithms become so complicated that we cannot achieve more because of our mind’s limitations. We can’t design and understand such complex algorithms as for example, speech recognition. So we must focus not on understanding how it works, but on preparing data, the training process and monitoring efficiency.
Igor: Machine learning is not an entity. People don’t understand that it’s not magic – it’s algorithms. You just apply it to really large amounts of data.
Nothing is static, and there is a continuum of machine and human activity evolving into new forms of communication.
Marc: I often see too much effort and resources going to building the right algorithm and not enough interest in looking at the actual data, which is usually the most important aspect of any ML project. It’s like if you want to prepare a delicious meal, you should focus more on the quality of ingredients rather than the recipe itself! In other words, always remember that famous saying in computer science: trash in, trash out!
Vlad: Funnily enough, the correlation is this: in order to make sense of the vast data we are generating, in business as well as in our personal daily lives, we have pushed the boundaries and created new tools like machine learning and content analytics which are capable of churning through that data to find patterns and thus simplify things for us. Otherwise, we would be living in a blurry cloud filled with unstructured information. Maybe what people don’t necessarily see is how far these technologies have permeated our lives. Today we don’t consider our mobile phones artificial intelligence, but only a decade ago they were treated as a wondrous innovation. Also, the automation of work has been with us since the first stone tools were crafted by our ancestors.
Code first or design first?
Bartłomiej: We believe in keeping things as simple as possible and self-explanatory, so design first. But sometimes the code is so simple that it is efficient to do code first. There are huge costs of maintaining both design and code in the long run, so you have to focus on the problem you want to solve and pick the proper solution.
IS: Client first. Everything you do – design or code. Sometimes it is easier to give the client a hacked version of the product than just show a visualization.
Vlad: For us, it was both. These things go hand in hand, the vision and the engineering, they are deeply intertwined. But that doesn’t mean that what you get is necessarily a finished product. Our product is constantly evolving. We are dedicated to releasing a new and improved version every year.
Lots of entrepreneurs are scared of launching imperfect products. What would you say to that?
Tomasz: Creating a perfect product is almost impossible (especially if you are a perfectionist). Designing without end is a trap because the technology changes very fast, so you may end up with an outdated product that customers don’t need. A better idea is building simple MVP’s (not only one) and test them with your target audience as soon as possible. Of course, there are some areas where this is impossible (e.g. gathering big data), but sometimes testing only the idea and user experience (UX) can give you a lot of feedback about what to do next.
Igor: Perfection is a delusion. As an engineer, I’m constantly fighting myself to show my unfinished work. In 2013, I was fortunate to be accepted to the MIT Hackathon. This was my biggest challenge because without knowing anyone I’ve succeeded and received some award with a totally imperfect product, hacked in only 48 hours. In today’s perfect world of perfect people, you need to teach yourself how to be imperfect and have the courage to show your work to others.
Marc: This is the most common mistake made by entrepreneurs across all disciplines. The lean approach has to be radically applied. Entrepreneurs should launch their products as soon as possible, and iterate very often to improve them. There are many examples where very successful products have been discovered almost by chance while testing other concepts. Entrepreneurs should be humble and very open to new ideas while developing the initial idea. Pivoting is a necessity. Most of the times, entrepreneurs are overly perfectionist, and whenever the product is finally ready in their eyes, it’s already far too late for the market. I remember the days when even Google was still in beta version even after being on Nasdaq! Innovation is always in beta and never finishes!
Creating a perfect product is almost impossible (especially if you are a perfectionist).
Bartosz: That you have to find the balance between the two extremes, an ideal product and a completely unattractive MVP. In my experience, MVP doesn’t always show if the product is good – maybe it serves its purpose but it’s really ugly and no one wants to use it. On the other hand, trying to create the ideal product through a launch can lead to a situation where we’re doing things that are fundamentally wrong.
Vlad: There is no such thing as a perfect product, to begin with. The history of technology stands as proof to that. Our startup went through more than one iteration in the beginning, and we are still working on developing our technology. The only way to go to market is with an imperfect product that lends itself to transformations. How else could it evolve? And there’s one more secret. Having a strong product is not enough to win the market. You need a good strategy, you need strong marketing, you need to understand the consumer, all of that which nurtures it and helps it grow.
D-RAFT Corporate Demo Days are designed to connect the best ready-to-scale, late stage startups with corporations that want to buy, partner or invest in disruptive solutions.