Building on decades of previous research, Curious AI – a Finnish pioneer in artificial intelligence – is facing an exciting future. Will AI truly work when it receives data direct from the factory floor or data about office routines? Kalle Raita, VP Sales, sketches what that future holds in store.
Curious AI received USD 800,000 in financing from international investors when it was founded in 2015, plus just under USD 4 million more that it raised in 2017. Kalle Raita explains what was achieved with it.
“The first three years were spent on basic academic research. During this phase, the algorithms were fine-tuned so that at the start of this year we were able to start commercialising the technology we developed. We’re looking for applications where commercially attractive, scalable products can be developed from our technology,” he begins.
There are two pilots currently ongoing. In one, Curious AI is applying machine learning to process optimisation. In an industrial plant, the personnel running a process make daily decisions as to how devices are run and what adjustments are made to them. Curious AI develops a digital assistant for them that helps them achieve the desired result, whether that is maximum production, lower costs or faster throughput. For example, the assistant can tell personnel what is worth doing during the next 12 hours in order to achieve the target. Neste Engineering Solutions is a partner in the pilot.
The other pilot is being conducted with Berggren Group and relates to interpersonal communication in an office environment. Curious AI’s goal is to release people from the kind of slavery to IT systems in which working time is spent on entering the contents of order books, invoices and receipts into IT systems via a keyboard. When the digital assistant interprets messaging between people and converts it into a format a computer understands, people’s time can be diverted to more productive use.
“The systems understand their own limits, though, so people will still be involved in making difficult decisions,” Raita says.
The last uphill stretch
Curious AI’s competitive assets are its continuous basic research in AI, its own patents, and the courage to think big and work with listed companies.
From the technology viewpoint, the pilots have progressed well. The first AI program version aimed at process optimisation is ready for trial, while the coding for the office routine optimiser is still in progress. Raita points out, however, that there’s a long road ahead before a shopfloor operator supervising a process would call artificial intelligence his/her favourite workmate. Production processes are large entities, so adopting new methods takes time.
Technology for managing large companies is now expected to make faster progress because the pieces to be controlled are smaller and the companies have IT staff who speak the same language as developers.
“We’re now on the last uphill stretch before the peak, where the landscape will open before us. Soon we’ll see where artificial intelligence will next be of most help,” comments Raita.
Why, then, did Curious AI take on such a big challenge as managing an entire factory’s production process? What need is there to struggle with the administration of personnel and accounting in, specifically, large corporations?
“We want to do something that truly benefits our customers. If the efficiency of an industrial enterprise’s processes can be enhanced by one or two per cent, or if a plant can be run each shift as well as if the best and most experienced operator were in charge, that makes a big difference in productivity and profits,” points out Raita.
Furthermore, only large companies have sufficient volumes of data. A data-driven system will not function without a large pool of data. Curious AI has worked hard to reduce the amount of data needed, but at present artificial intelligence still needs to chew through some 10,000 similar data items to get the same result out before it is viable.
“Usually with R&D you first start small and then scale upwards, but with AI technology it’s better to do it the other way round. First we want to proceed with large customers to a level at which we can address their requirements, and only then will the product be scaled for smaller players.”
Independent learning means the coder does not need to be all-knowing
There are three discernible waves of artificial intelligence. The first wave systems were based on conventional rules-based programming. The applications in this wave do exactly what the programmers told them to do. Second wave AI is data-driven, and this is the type of artificial intelligence seen in marketing at present. Second wave systems do exactly what the teaching data tells them to do. Programming with data is easier than the conventional model in the case of problems to which a human has the correct answers but is unable to justify decisions in detail. Two examples of these types of tasks are identifying objects in a picture and driving a vehicle.
Third wave artificial intelligence simulates a human brain and its conscious, slow thought process that also takes into account the backgrounds to events and that can select the correct path of action based on them. A person often makes decisions intuitively, in which case he/she is unable to list the grounds for a decision but nevertheless knows the decision is right. That is what Curious AI’s technology incorporates.
“We build a system that itself learns what to do from real-world data. That way, coding is not needed for each new aspect or each new situation. We believe that by teaching a machine to learn and to understand its limits, we can open up new areas in which artificial intelligence can be profitably deployed,” explains Raita.
AI is not a mass product but rather an enhancer of operational efficiency
Raita points out that Curious AI’s artificial intelligence is hardly likely to become a consumer product. The company could well, in time, considerably boost GDP – not in itself, but by enhancing efficiency in other companies’ operations and by helping people to work better in different ways. The multiplier effects of harnessing AI can be very large. He points out, however, that technological transitions are only fast in science fiction and movies.
“Transitioning to an era of artificial intelligence will probably be so slow that we don’t even notice it. One day we’ll simply notice that we’ve been using some highly intelligent systems for many years. Artificial intelligence and machine learning make a lot of things possible, but are rarely a product in themselves. Added value is created from the small wonders that artificial intelligence brings about. It’ll be interesting to see how artificial intelligence and humans get along together,” muses Raita.
The artificial intelligence in Curious AI’s sights is an assistant to humans, not an independent actor. It will, without complaining, do the slow and tedious work so disliked by many people. It will also be capable of handling large amounts of data better than a human brain.
“People are good at understanding entities up to a certain point but complex systems, such as finely-tuned industrial and business processes, go beyond that point. When a person can no longer keep a complete entity in mind, a solution based on machine learning is better at predicting and at identifying aberrations,” says Raita.
A person is needed to make decisions and to check the results of work done by artificial intelligence, because artificial intelligence makes mistakes just like humans do. One of the ethical challenges facing the development of artificial intelligence relates exactly to this point: how resilient will we be to AI’s mistakes and what kind of risks will they pose? In many cases the risks are not large, and often the risks can be reduced with human support, but Raita believes they should not be ignored.
In his opinion, there is a wide gap between the expectations placed on artificial intelligence and the current implementations of it. This creates some challenges for sales, marketing and social dialogue.
Global problems will be solved together
Research into artificial intelligence is progressing rapidly, everyone is interested in each other’s results and minor improvements are announced every week. For that reason, Curious AI keeps its whiteboard hidden.
“Our long-term vision and position in the world is to develop artificial general intelligence, AGI. In other words, machine intelligence that performs all the intelligent actions of humans. We surely won’t get there tomorrow, and no-one knows exactly how to get there. A digital co-worker is a step in the right direction. When our artificial intelligence can be used flexibly for different tasks and its training doesn’t require an AI expert, we’ve fulfilled our mission and we can all take a vacation,” says Raita.
Curious AI’s specific goal is not to solve big global problems, such as climate change, with artificial intelligence, but the indirect effects of applying artificial intelligence could encompass that.
“Process optimisation can have an impact in many different ways – in how we use energy, for instance, which in turn affects the climate. No single company can solve such huge problems alone. But together we might be able to do it,” concludes Raita.
Photos: Junnu Lusa
Who he is: A jack-of-all-trades in the software business. Curious AI’s VP Sales. Curious AI’s product team member responsible for sales, customer relationships and product specifications.
Education: M. Sc. in information technology from Helsinki University of Technology, majoring in interactive digital media.
Professional backstory: Started as a programmer in Hybrid Graphics, a developer of graphics software, in 2001. Nvidia bought the company in 2006. In 2010, Raita moved to drawElements, where he was CEO from 2011 to 2014. Then Google bought the company. He worked at Google, California, until 2017.
Curious AI Oy
Established: 2015. Founder Harri Valpola, D. Sci. (Tech.), who at that time had already studied artificial intelligence for over 20 years.
What it does: Develops and deploys self-learning artificial intelligence that models the functioning of the human brain.
Personnel: Some 25 people.
Owners: Balderton Capital, Invus, Lifeline Ventures, Data Collective, Westcott and Jaan Tallinn. Tesi is an indirect owner through Balderton Capital.
Watch the Curious AI’s CEO Harri Valpola’s presentation about artificial intelligence: