As AI models become increasingly refined, organisations are starting to notice the diverse range of solutions they can provide across not just data science teams but all departments of a company.
Despite this, scaling AI solutions across a company is an extremely expensive and complex venture that most firms would struggle to see a return on investment from. In a market where AI has so much to offer but with such large costs, Peltarion is striving to bridge the gap enterprises face when realising AI solutions in terms of resources and knowledge.
AI News joined Peltarion CEO Luka Crnkovic-Friis and operational AI expert Johan Hartikainen to discuss how the company’s cloud-based software platform is helping to solve this catch-22 situation affecting the industry.
AI News: In another interview you described Peltarion as doing for AI what WordPress did for HTML coding, would you still consider this an accurate analogy?
Luka Crnkovic-Friis: Yes, but with some reservations. The analogy is correct as to what we are trying to accomplish – simplifying a complex system to allow for democratised usage – but on a much higher technical level.
AN: Aside from through that analogy, how would you describe Peltarion’s goal within the AI industry?
LC: If you look at the landscape of AI tools today, you have the super simple tools at one end where you click a button and get results and on the other hand you have tools like TensorFlow that are very advanced and require expertise. Although the simple tools are easy to use they are not very applicable to real world problems and the complex tools are super powerful but need a team of experts and lots of resources to provide solutions. Peltarion’s goal is to position itself in the middle of this chasm and expand to both sides, bringing as much simplicity to powerful AI solutions as possible.
AN: How is this goal realised in practicality?
LC: Our customers mainly fall into two categories. Either they’re larger enterprises who are on a digitalisation journey but lack scalability or it’s smaller companies and startups who don’t have the resources to build an AI department to begin with.
Even today, operational use of modern AI and deep learning at a large scale is limited to the big tech companies. This is because the tools available today are primarily for research and building proof of concept, which cost a lot in terms of infrastructure and the type of talent required. Our platform combines ease of use – so you don’t have to be an expert – with the operational aspect – so you can actually put stuff into production on a commercial and industrial scale.
AN: What is one of your favourite examples of a company you have helped?
LC: The cases I enjoy most are when we enable domain experts to do something that we as AI experts could not have done but, similarly, that they could not have done without our platform. That’s when it feels like we are hitting the sweet spot.
Last year we worked with a medtech company called SciBase. They designed a probe pen that can detect melanoma when put against a birthmark. They then had the idea that, in addition to the probe data based on electrical measurements, they could add images via deep learning. However, when they tried to do it the results weren’t any better. When we stepped into help, the results were slightly better but still somewhat mediocre.
However, using the data from our platform they were able to find another use case. So there is this barrier in the skin that prevents viruses and bacteria from getting into the body and people with allergies have a weakened barrier. When it’s in an especially poor state they get eczema rashes and the only way to measure that has been to do a biopsy. What’s more, this only works on skin where there is an ongoing rash. Using Peltarion’s probe data, SciBase were able to predict weeks in advance on clear skin when this membrane will be weakened. Then it is as simple as slapping on some moisturising cream and you have no problem. In fact, there are now some promising clinical studies being done that show if you test new-borns with this and use the right cortisone treatment they don’t develop allergies at all.
Why I really like this use case is because they’re not data scientists, and we are not medical experts, but together we can realise fantastic things with AI through collaboration.
AN: Looking beyond specific use cases, what trends are Peltarion seeing across the AI industry?
LC: Bigger models. Every time the industry takes a step in order of magnitude we get these emergent properties where the models we are working with can do completely new things. Nothing except the size of the model has changed and yet new capabilities emerge. I think we are now at a stage where the models have become so huge that only the tech giants can build things from scratch.
Johan Hartikainen: We’re also seeing a lot of enterprises who started using centralised data science teams to solve their biggest problems a few years ago now realising that this doesn’t translate well into the rest of the company understanding AI. The last year or two has seen huge shifts in the market as companies seek to democratise AI across their departments and put it in the hands of their employees.
AN: Considering this trend, what will be a key challenge in seeing through its completion?
LC: I think general understanding of AI still has a long way to go. When clients know what they want to solve, have an idea of the use case, and have the data, they can have a model in production within an hour. While if you have a client with less understanding, they need much more guidance on use cases and what data to use. So there is this huge gap between those that are familiar with what AI can do and have realisable ideas and those still learning.
Johan will be delving deeper into how to democratise AI for usage across an entire organisation on day one of AI and Big Data Expo Global, which runs from 6-7 September 2021. Find out more about the event and how to attend here.