Know My Company
What is your experience about the journey into Artificial Intelligence?
As a 14-year-old that could code and hated chores, I dreamt of creating intelligent robots that could do the drudgery for me. I wrote overly simplistic AI brain architectures and dreamt. It was only at university when we had a lecture where they cited Minsky on why Neural Networks didn’t work – which didn’t sit right with me, that I started experimenting with various AI techniques.
Throughout my career, from gaming to banking, I’ve found that using AI / ML allowed us to do things better. So, when I met Alex Allan, my co-founder at Kortical, and discovered that he had a Ph.D. in AI, we immediately hit it off. I had just delivered an AI solution to Barclays, and Alex and I quickly realized that my background of deep tech and enterprise AI, coupled with his deep knowledge of the latest AI, was a powerful combo.
What made you launch an AutoML platform?
Kortical’s first project was a sales optimization product for call centers. By training our models on the call history, we could determine which agents should talk to which customers, based on geographic demographics. Our models increased sales conversions by up to 20%. But because each customer’s data was slightly different, it was taking 3 months to onboard new customers, as we did the data-science for each one. We knew that was untenable, so we started automating.
With each dataset being slightly different, we started embedding more and more machine learning approaches into the platform to get the best results in each case. We soon realized that we had built the core of an engine that could be used to solve almost any AI problem. We started Kortical in late 2016. In the years since the platform has come a long way, but our DNA of enabling data-scientists to get the best results and get them live, delivering value quickly, is still at the heart of the Kortical platform today.
What are the core tenets of Kortical’s platform?
We built Kortical with two main goals in mind. We want to get the absolute best model results we can, and we want to make it easy to get the models live, delivering value quickly.
Better results tie directly to better ROI increased sales conversions, better demand planning, driving efficiency, more automation, etc. While AutoML is better at certain aspects of model building – like parameter tuning; data-scientists still have a lot of value to add. Especially in how to frame a business problem and represent it, so machine learning can perform best. Kortical is uniquely positioned to empower data-scientists. Giving them full control to work with the AutoML and get the best of both man and machine.
We also appreciate that getting models live in the enterprise comes with requirements, such as failover, redundancy, audit, multiple environment support, scalability, and operations support. Kortical is trusted with predicting demand and managing the logistics for the national UK blood supply for hospitals, which is a critical system. Encapsulating an enterprise-grade deployment process in an elegant user interface that allows for simple deployment, online learning and model management is key.
Last, but not least, opening the BlackBox and explaining how the models work is really important. Not only does this help data-scientists to build better models; but it allows you to spot and remove bias, comply with GDPR and the right to an explanation. A clear explanation is also critical in showing senior stakeholders how Machine Learning is making its decisions and building confidence.
How does AI leverage AutoML solutions?
Elon Musk has a great quote “If your competitor is racing to build AI, and you don’t, they will crush you”. Gartner reports that on average it’s taking 58% of companies 2 years to get to pilot with AI solutions, and 4 years to get them live. Kortical customers can get to pilot in 1 month. Our typical, full solution roll-out, even in traditionally risk-averse industries, like healthcare and banking, is 6 months. However, Kortical offers far more than AutoML, we offer full AI as a service – automating the deployment and management aspects, as well as the model creation.
Success rates are also key to AI. Gartner says that 85% of AI projects will fail . Trying to build out these complex AI automation tech stacks, from scratch, is hard. Many companies get lost reinventing the wheel. Often, they don’t have the in-house talent or budget to build it correctly. In much the same way, developers use best in class databases, rather than write their own from scratch. Leveraging tried and tested AI as a service platform, so they can reduce project risk and focus on solving business problems with AI.
What are the key benefits of AutoML? How can data scientists benefit from Kortical?
There are a number of steps in a typical data science project. Some of those steps require a lot of human knowledge and intuition, and some of them are more of a ‘robotic’ search of the solution space, trying to figure out the best tools and parameters for the job. Like much of modern AI, AutoML takes away the more robotic aspects of building a model and makes it possible to rapidly experiment. Almost at the speed of thought.
A data-scientist might be thinking that an NLP task would work really well, if they had word2vec feeding into a Deep Neural Net, with a couple of RELU layers. A decision like that would typically be followed by weeks or months of building and tuning the model. With Kortical, the data scientist can specify that in seconds and have the platform creating models immediately. The data-scientist can then interact with the platform to refine the solution further, fleshing out the bits they’re interested in, letting AutoML take care of the rest.
Kortical is the only platform offering this patented, high level of interactivity and control over the AutoML process, which is key to getting the best results and iterating quickly.
What are your AI research programs and what is the most outstanding project or campaign at Kortical?
Kortical’s research focus is our ‘AI to create AI’. The brain of our AutoML. While we’re pretty tight-lipped about how that works, we love talking about the results researchers and data scientists have been able to achieve by using it.
For example, from August (2019), we’re set to go live with the first live AI project in the NHS. While other projects are still in their infancy, ours will be the first solution to go live.
Kortical’s AI-powered an 8.6% improvement in kidney transplant waiting time prediction, using NHS genetic data. We are currently on track to deliver a 50% improvement in efficiency and cost for blood supply and demand planning across the UK.
Kortical also won the Schroders and FCA Datathons against major industry players and other AutoML solutions, by building highly accurate models that we designed to solve major industry challenges in days.
I don’t want to bore your readers with a massive list of examples, but it’s amazing how many industries teams are using Kortical to achieve industry-leading results, in fields as diverse as reducing carbon emissions to reading emails.
How do your solutions automate the creation of models to process Big data?
Big data is the lifeblood of AI, the more data the better the model. It’s easier to create better models with more data. However, at this point, our findings show that very few business challenges have enough data behind them to be classified as big data. So, while big data does make for better AI models, the real challenge is making sure the AutoML can create good models for all levels of data, from large to small, to get the results that can justify the investment in further data capture.
At the smallest end of the scale, we’ve managed to create a valid model with 85% accuracy from a training set of just 60 emails. At the larger end, we’ve had an 85 terabyte data set to work from. Being a native cloud-scale platform, that uses distributed AI to build models, gives us an inherent advantage when it comes to dealing with big data.
What does your ‘Ideal Customer’ look like? Which new geographies are you currently targeting?
Our ideal customer has a reasonable data set. At least a thousand examples and the clear problem they want to solve or has the ambition to roll AI out across their business, at scale.
Ideally, the customer would have sign-off and be able to move quickly, but this is seldom the case, and we have a lot of experience supporting visionaries within a business to build support for AI projects through quick, low budget proof points, through to low friction pilots and full delivery.
When clients work with Kortical, we invest in them and their success. So far, we have a 100% success track record helping businesses deliver AI at pace. We love hearing from people who have the ambition to make a real impact on their business with AI.
What are the biggest challenges and opportunities for AI companies? How does AutoML help?
There are a lot of smaller AI companies who have a great idea of how AI can solve or improve something in a specific vertical. If they’re not aware of the advances in AutoML and AIAAS, they are probably making inordinate investments in building out the models and technology; when they would much rather be focused on realizing their goals, without that investment. While also doing it with less risk, many times faster and getting a return on their efforts.
This scenario is being played out every day, across the enterprise, but that’s where the opportunity is even greater. Each new custom-built AI system is another operating system that needs a team of engineers, data scientists, and operations to support and maintain it. The big tech players have already invested several billion in AutoML and AI as a service platform, but the chances of industry-focused business investing that much, or even getting something comparable, is minuscule.
The real opportunity is using tried and tested technology to get most of your AI solutions live and maintained by operations in a uniform way. This makes for a very scalable data science process, which can be rolled out and adopted by numerous teams quickly. Giving that organization a real edge on their competitors.
Where do you see AI/Machine Learning and other smart technologies heading beyond 2025?
For all the hype, we’re still just on the cusp of the AI revolution. Even the real industry leaders are just starting to get operational with AI in their businesses. From the companies that we engage with, we are seeing transformative results and proof that businesses are reacting by aligning themselves around AI. While McKinsey has reported that on average strong AI adopters are seeing a +10% difference compared to partial or non-adopters; we are just seeing the tip of the iceberg and these early successes are why mentions of AI on earnings calls are through the roof.
As enterprise AI technology, like Kortical’s, drives down average AI delivery time from 4 years to months. Delivering pilots in weeks. We expect the pace, adoption, and impact of AI to snowball. Businesses that haven’t prepared, or learned how to work effectively with AI, risk being left behind.
Of course, at some point, AI enablement will plateau and become the new norm; but with the constant stream of advances in AI, that seems unlikely by 2025. The other side of the coin is that right now AI is mostly helping us do what we currently do, but better. As the industry becomes more familiar with AI, I think we’ll see more left-field applications and seismic shifts in how business processes are done.
What AI start-ups and labs are you keenly following?
I think what the guys at Waymo are doing is very interesting. Their AI systems have driven 10 million miles, with hardly any incidents. Waymo has a live taxi service on some US states. Tesla tends to grab all the headlines and I like them, but the Waymo guys are already up and running.
Deepmind is working on some interesting problems. Having beaten the world’s best at ‘GO’ and coming close on ‘Starcraft’. Their deep reinforcement learning approaches are very cool and work well with simulations, where you can create billions of examples. They are making progress in reducing that requirement, making those techniques applicable to more real-world problems.
My team and I constantly track all the latest research from the universities and labs. The pace of progress is phenomenal. We build the best into Kortical, enabling us to train up new models that get smarter from added data and better AI techniques.
What technologies within AI and computing are you interested in?
I’m actually a little in love with our core engine ‘the AI to create AI’, it’s such a deep and complex challenge to get your teeth into.
On top of that, there’s NLP, time series, signal processing, deep learning architectures that all play into our work, but are super interesting. As an example, the rate of progress in NLP just this year, first with BERT and now XLNet. I’m also very interested in hyper scalable data and compute cloud architectures. Quantum computing is also a keen interest of mine, though I haven’t quite managed to find a good home for that in Kortical yet.
What’s your smartest work-related shortcut or productivity hack?
I wouldn’t think of myself as a guru on this topic. I love weekends, as a business leader getting a long unbroken run at something during the week is almost impossible, but weekends give me a bit of breathing room to really get my head down. I’m not saying it’s advisable, but it sure is productive.
Apart from that, I’d say focus on delegation and processes. If you can get the right ownership and processes in place, you have parts of your business that tick along and self-correct, which frees up a lot of time.
As an AI leader, what industries you think would be fastest to adopting AI/ML with smooth efficiency? What are the new emerging markets for AI technology markets?
This is a bit of an odd question because the facts on the ground seem to contradict what would seem like the logical order. Finance and healthcare are typically bureaucratic sectors that deal with a lot of highly sensitive customer data, so you might expect them to be the laggards on tech adoption, but they seem to be the strongest adopters.
Manufacturing, supply chain, indeed any industry that doesn’t deal with sensitive data should be easy to transform. Our experience shows that when businesses in these sectors go for it, you can really move quickly and make a big difference, but those sectors can be hard to engage with.
Regarding the emerging markets for AI, I honestly believe it’s everything. In 10 years, AI will be as integrated as part of our lives and business, as the internet is now.
Tag the one person in the industry whose answers to these questions you would love to read:
Mustafa Suleyman is head of applied AI and a co-founder of Deepmind might have an interesting perspective. Or Simon James, who is a friend and Group VP of Data-science for Sapient Publicis. He always has a unique perspective and engaging way of getting his points across.
Thank you, Andy! That was fun and hope to see you back on AiThority soon.
Andy Gray is a Co-founder and CEO at Kortical
Kortical uses AI to build AI, which enables companies to build, explain and deploy world class, enterprise grade machine learning and artificial intelligence models. We accelerate the process of delivering AI solutions from years to weeks. Kortical not only has a world class platform, but also offers expertise from ideation to strategy, delivery to bespoke app creation to ensure that you from data to business value from AI, fast. NHS, Deloitte, BT/MBNL are all benefiting from Kortical’s platform and services to deliver AI innovation at scale and at speed.