Artificial intelligence applications are not created in managerial suites – the entire organisation must understand the basics of artificial intelligence
Payment transactions based on face recognition, artificial intelligence physicians and members of boards of directors, personal assistants, better forecasts and decision-making, automated factories and transport.
Artificial intelligence holds a great deal of promise and a study* has shown that 71 per cent of the executive management level managers in companies consider artificial intelligence an important topic. When it comes to artificial intelligence, Finland is in a good position by international standards – now all that remains is to work out how artificial intelligence should be applied to individual organisations. We enlisted Maria Ritola’s guidance to find out.
Ritola is one of the founders of Iris.ai, an artificial intelligence company. Iris.ai could be described as a research assistant based on artificial intelligence who helps researchers and students find the right content more efficiently and more quickly. The application reduces the amount of manual work while also proposing content that data seekers may not even know to search for. Iris.ai can reduce the time taken for literature reviews by up to 80 per cent.
Iris.ai is a good example of how artificial intelligence can semi-automate work and help people with routine work. Typical neural-network artificial intelligence applications are based on image, speech and text recognition and solve various optimisation problems. “For example, the Google Smart Compose application, which is based on neural networks, suggests how Gmail users could finish their sentences while they are writing. Artificial intelligence can also be trained to become skilful at recognising images – artificial intelligence can use pictures taken using hyperspectral cameras to identify things such as spoiled food more accurately than the human eye,” says Ritola.
In addition to work tasks, the impact of artificial intelligence will also stretch to processes and business models. Ritola mentions processes such as design work, where artificial intelligence could help to build new models: “For example, a designer may specify that a chair must be able to withstand 300 kilograms or that the seat must be 50 centimetres high. Based on the criteria entered, the artificial intelligence application can create thousands of alternatives for the designer.”
Ritola expects business models to be based increasingly on forecasts, which are nowadays often based on user data. “One example is the Netflix recommendation system, which, incidentally, is surprisingly good. Similar services based on recommendations will become widespread.”
Throwing data into neural networks is not the solution
When a company or organisation is thinking about using artificial intelligence, it is essential that people understand the basic concepts of artificial intelligence.
“The people who are not working in IT should also be given training,” Ritola says. “They should be told which things neural networks, for example, can be used for. When people have a stronger grasp of the logic and applications of artificial intelligence, they begin to understand which things in their unit could be resolved using artificial intelligence. People know their own work best.”
When ideas begin to emerge, IT is able to enhance them and think about which data and models are needed to deploy artificial intelligence. According to Ritola, expertise from outside the organisation is often needed. However, this first requires an understanding of the tasks and processes where artificial intelligence can be put to use.
Data is naturally an essential part of artificial intelligence applications. However, throwing data into neural networks is no panacea.
“This could be compared with the way the internet was used 20 years ago. The companies that pioneered digitalisation were not the ones that set up websites. They were the ones that did everything, from A/B testing of products to shortening product cycles and identifying new markets.”
“Similarly, data should now be processed at a strategic level. Organisations should think about how data can be collected as a by-product of existing operations, via partnerships or by combining other things. Many may be surprised by the fact that the quantity of data need not be great. Rather small amounts will get you started if the quality is good. In the future, as new learning methods become established, small data solutions will most likely become commonplace particularly on the B2B side,” Ritola says.
It is easy to reach the crest of a wave when you do not expect a flawless end product
Expectations may be too high on artificial intelligence pilot projects. A more realistic approach will help you get on the right track.
“The assumption may be that the pilot project includes all of the material that can be found on the internet, everything that can be accessed via a smart application and all manner of analysis can be performed. When testing, expectations should be set realistically, and the critical factors that must be satisfied should be defined. If expertise from outside the organisation is used, make sure that all parties know what will arise as a result of the collaboration and what will not,” Ritola says.
“After the pilot, the most challenging work involves scaling the experiment, i.e., building new operating models at the organisational level.”
Ultimately, people are responsible for decisions
One challenge facing artificial intelligence applications is the question of ethics, which has also been a topic of public debate.
“It is good that the ethics of artificial intelligence are being discussed and that frameworks for perceiving questions of ethics are being formulated. To avoid bias, companies and organisations must understand which types of data were used to train the models and which types of data may be lacking. It may be impossible to obtain fully comprehensive data but, when parties are aware of the deficiencies, they can be taken into consideration when decisions are made. Naturally, the concept of objectivity is a philosophical one, as we may already have numerous biases between our ears which we are not even aware of,” Ritola states.
The problems with artificial intelligence are related to factors such as centralisation of power and the risk of increasing discrimination: “China is gradually introducing a point-scoring system that rewards or punishes people based on the things they say or do. A social media company has applied for a patent on technology that would make loan decisions based on the credit ratings of the applicant’s Facebook friends. HR departments have been using an application that compares the facial movements of the best employees with the facial movements of job applicants.”
Ritola emphasis that when people make decisions, they must always be able to apply the collected data, understand what it is about and personally evaluate the situation. “Responsibility for decision-making cannot be outsourced to a machine.”
“In the best case, algorithms will help to make us aware of biases within our own thinking. Transparency, jointly set targets and open discussion will combine with technology to help to build a world that we really want to live in.”
* Artificial Intelligence in Europe, Finland, Outlook for 2019 and Beyond, EY